TXR - Programming Language (Version 296)
txr [ options ] [ script-file [ arguments ... ]]
TXR is a general-purpose, multi-paradigm programming language. It comprises two languages integrated into a single tool: a text scanning and extraction language referred to as the TXR Pattern Language (sometimes just "TXR"), and a general-purpose dialect of Lisp called TXR Lisp.
TXR can be used for everything from "one liner" data transformation tasks at the command line, to data scanning and extracting scripts, to full application development in a wide range of areas.
A script written in the TXR Pattern Language, also referred to in this document as a query, specifies a pattern which matches one or more sources of inputs, such as text files. Patterns can consist of large chunks of multiline free-form text, which is matched literally against material in the input sources. Free variables occurring in the pattern (denoted by the @ symbol) are bound to the pieces of text occurring in the corresponding positions. Patterns can be arbitrarily complex, and can be broken down into named pattern functions, which may be mutually recursive.
In addition to embedded variables which implicitly match text, the TXR pattern language supports a number of directives, for matching text using regular expressions, for continuing a match in another file, for searching through a file for the place where an entire subquery matches, for collecting lists, and for combining subqueries using logical conjunction, disjunction and negation, and numerous others.
Patterns can contain actions which transform data and generate output. These actions can be embedded anywhere within the pattern-matching logic. A common structure for small TXR scripts is to perform a complete matching session at the top of the script, and then deal with processing and reporting at the bottom.
The TXR Lisp language can be used from within TXR scripts as an embedded language, or completely standalone. It supports functional, imperative and object-oriented programming, and provides numerous data types such as symbols, strings, vectors, hash tables with weak reference support, lazy lists, and arbitrary-precision ("bignum") integers. It has an expressive foreign function interface (FFI) for calling into libraries and other software components that support C-language-style calls.
TXR Lisp source files as well as individual functions can be optionally compiled for execution on a virtual machine that is built into TXR. Compiled files execute and load faster, and resist reverse-engineering. Standalone application delivery is possible.
TXR is free software offered under the two-clause BSD license which places almost no restrictions on redistribution, and allows every conceivable use, of the whole software or any constituent part, royalty-free, free of charge, and free of any restrictions.
If TXR is given no arguments, it will enter into an interactive mode. See the INTERACTIVE LISTENER section for a description of this mode. When TXR enters interactive mode this way, it prints a one-line banner announcing the program name and version, and one line of help text instructing the user how to exit.
If TXR is invoked under the name txrlisp, it behaves as if the --lisp option had been specified before any other option. Similarly, if TXR is invoked under the name txrvm, it behaves as if the --compiled option had been given.
Unless the -c or -f options are present, the first non-option argument is treated as a script-file which is executed. This is described after the following descriptions of all of the options. Any additional arguments have no fixed meaning; they are available to the TXR query or TXR Lisp application for specifying input files to be processed, or other meanings under the control of the application.
Options which don't take an argument may be combined together. The -v and -q options are mutually exclusive. Of these two, the one which occurs in the rightmost position in the argument list dominates. The -c and -f options are also mutually exclusive; if both are specified, it is a fatal error.
Normally, if this stream is connected to a terminal device, it is automatically marked as having the real-time property when TXR starts up (see the functions stream-set-prop and real-time-stream-p). The -n option suppresses this behavior; the *stdin* stream remains ordinary.
The TXR pattern language reads standard input via a lazy list, created by applying the lazy-stream-cons function to the *stdin* stream. If that stream is marked real-time, then the lazy list which is returned by that function has behaviors that are better suited for scanning interactive input. A more detailed explanation is given under the description of this function.
If the -n option is effect and TXR enters into the interactive listener, the listener operates in plain mode instead of the visual mode. The listener reads buffered lines from the operating system without any character-based editing features or history navigation. In plain mode, no prompts appear and no terminal control escape sequences are generated. The only output is the results of evaluation, related diagnostic messages, and any output generated by the evaluated expressions themselves.
V_0_0[0]="a"
V_0_1[0]="b"
V_1_0[0]="c"
V_1_1[0]="d"
V_0_0[1]="e"
V_0_1[1]="f"
V_1_0[1]="g"
V_1_1[1]="h"
With -a 2, it comes out as:
V_0[0][0]="a"
V_1[0][0]="b"
V_0[0][1]="c"
V_1[0][1]="d"
V_0[1][0]="e"
V_1[1][0]="f"
V_0[1][1]="g"
V_1[1][1]="h"
The leftmost bracketed index is the most major index. That is to say, the dimension order is: NAME_m_m+1_..._n[1][2]...[m-1].
Example:
Shell script which uses TXR to read two lines "1" and "2" from standard input, binding them to variables a and b. Standard input is specified as - and the data comes from shell "here document" redirection:
txr -B -c "@a
@b" - <<!
1
2
!
The @; comment syntax can be used for better formatting:
txr -B -c "@;
@a
@b"
The script-file argument becomes optional if at least one -e, -p, -P or -t option is processed.
If the evaluation of every expression evaluated this way terminates normally, and there is no script-file argument, then TXR terminates with a successful status, instead of entering the interactive listener. The -i option can be used to request the listener.
The requested value of N can be too low, in which case TXR will complain and exit with an unsuccessful termination status. This indicates that TXR refuses to be compatible with such an old version. Users requiring the behavior of that version will have to install an older version of TXR which supports that behavior, or even that exact version.
If the option is specified more than once, the behavior is not specified.
Compatibility can also be requested via the TXR_COMPAT environment variable instead of the -C option.
For more information, see the COMPATIBILITY section.
Note that --lisp and --compiled influence how the argument of the -f option is treated, but only if they precede that option.
If the file has a recognized suffix: ".tl", ".tlo", ".tlo.gz", ".txr" or ".txr_profile", then these options have no effect. The suffix determines the interpretation of the content. Moreover, no suffix search takes place: only the given path name is tried.
After the options, the remaining arguments are treated as follows.
If neither the -f nor the -c options were specified, then the first argument is treated as the script-file. If no arguments are present, then TXR enters interactive mode, provided that none of the -e, -p, -P or -t options had been processed, in which case it instead terminates.
The TXR Pattern Language has features for implicitly treating the subsequent command-line arguments as input files. It follows the convention that an argument consisting of a single - (dash) character specifies that standard input is to be used, instead of opening a file. If the query does not use the @(next) directive to select an alternative data source, and a pattern-matching construct is processed which demands data, then the first argument will be opened as a data source. Arguments not opened as data sources can be assigned alternative meanings and uses, or can be ignored entirely, under control of the query.
Specifying standard input as a source with an explicit - argument is unnecessary. If no arguments are present, then TXR scans standard input by default. This was not true in versions of TXR prior to 171; see the COMPATIBILITY section.
TXR begins by reading the script, which is given as the contents of the argument of the -c option, or else as the contents of an input source specified by the -f option or by the script-file argument. If -f or the script-file argument specify - (dash) then the script is read from standard input.
In the case of the TXR pattern language, the entire query is scanned, internalized, and then begins executing, if it is free of syntax errors. (TXR Lisp is processed differently, form by form.) On the other hand, the pattern language reads data files in a lazy manner. A file isn't opened until the query demands material from that file, and then the contents are read on demand, not all at once.
The suffix of the script-file is significant. If the name has no suffix, or if it has a ".txr" suffix, then it is assumed to be in the TXR pattern language. If it has the ".tl" suffix, then it is assumed to be TXR Lisp. The --lisp and --compiled options change the treatment of unsuffixed script file names, causing them to be interpreted as TXR Lisp source or compiled TXR Lisp, respectively.
If a file name is specified which does not have a recognized suffix, and names a file which doesn't exist, then TXR adds the ".txr" suffix and tries again. If that doesn't exist, another attempt is made with the ".tlo" suffix, which will be treated as as a TXR Lisp compiled file. If that doesn't exist, then ".tlo.gz" is tried, expected to be a file compressed in gzip format. Finally, if that doesn't exist, the ".tl" suffix is tried, which will be treated as containing TXR Lisp source. If either the --lisp or --compiled option has been specified, then TXR skips trying the ".txr" suffix, and tries only ".tlo" followed by ".tlo.gz" and ".tl".
A TXR Lisp file is processed as if by the load macro: forms from the file are read and evaluated. If the forms do not terminate the TXR process or throw an exception, and there are no syntax errors, then TXR terminates successfully after evaluating the last form. If syntax errors are encountered in a form, then TXR terminates unsuccessfully. TXR Lisp is documented in the section TXR LISP.
If a query file is specified, but no file arguments, it is up to the query to open a file, pipe or standard input via the @(next) directive prior to attempting to make a match. If a query attempts to match text, but has run out of files to process, the match fails.
TXR sends errors and verbose logs to the standard error device. The following paragraphs apply when TXR is run without enabling verbose mode with -v, or the printing of variable bindings with -B or -a.
If the command-line arguments are incorrect, TXR issues an error diagnostic and terminates with a failed status.
If the script-file specifies a query, and the query has a malformed syntax, TXR likewise issues error diagnostics and terminates with a failed status.
If the query fails due to a mismatch, TXR terminates with a failed status. No diagnostics are issued.
If the query is well-formed, and matches, then TXR issues no diagnostics, and terminates with a successful status.
In verbose mode (option -v), TXR issues diagnostics on the standard error device even in situations which are not erroneous.
In bindings-printing mode (options -B or -a), TXR prints the word false if the query fails, and exits with a failed termination status. If the query succeeds, the variable bindings, if any, are output on standard output.
If the script-file is TXR Lisp, then it is processed form by form. Each top-level Lisp form is evaluated after it is read. If any form is syntactically malformed, TXR issues diagnostics and terminates unsuccessfully. This is somewhat different from how the pattern language is treated: a script in the pattern language is parsed in its entirety before being executed.
A query may contain comments which are delimited by the sequence @; and extend to the end of the line. Whitespace can occur between the @ and ;. A comment which begins on a line swallows that entire line, as well as the newline which terminates it. In essence, the entire comment line disappears. If the comment follows some material in a line, then it does not consume the newline. Thus, the following two queries are equivalent:
The comment after the @a does not consume the newline, but the comment which follows does. Without this intuitive behavior, line comment would give rise to empty lines that must match empty lines in the data, leading to spurious mismatches.
Instead of the ; character, the # character can be used. This is an obsolescent feature.
TXR has several features which support use of the hash-bang convention for creating apparently standalone executable programs.
This removal allows for TXR queries to be turned into standalone executable programs in the POSIX environment using the hash-bang mechanism. Unlike most interpreters, TXR applies special processing to the #! line, which is described below, in the section Argument Generation with the Null Hack.
Shell session example: create a simple executable program called "twoline.txr" and run it. This assumes TXR is installed in /usr/bin.
$ cat > hello.txr
#!/usr/bin/txr
@(bind a "Hey")
@(output)
Hello, world!
@(end)
$ chmod a+x hello.txr
$ ./hello.txr
Hello, world!
When this plain hash-bang line is used, TXR receives the name of the script as an argument. Therefore, it is not possible to pass additional options to TXR. For instance, if the above script is invoked like this
$ ./hello.txr -B
the -B option isn't processed by TXR, but treated as an additional argument, just as if txr script-file -B had been executed directly.
This behavior is useful if the script author wants not to expose the TXR options to the user of the script.
However, the hash-bang line can use the -f option:
#!/usr/bin/txr -f
Now, the name of the script is passed as an argument to the -f option, and TXR will look for more options after that, so that the resulting program appears to accept TXR options. Now we can run
$ ./hello.txr -B
Hello, world!
a="Hey"
The -B option is honored.
#!/usr/bin/txr -B -f
To support systems like this, TXR supports the special argument --args, as well as an extended version, --eargs. With --args, it is possible to encode multiple arguments into one argument. The --args option must be followed by a separator character, chosen by the programmer. The characters after that are split into multiple arguments on the separator character. The --args option is then removed from the argument list and replaced with these arguments, which are processed in its place.
Example:
#!/usr/bin/txr --args:-B:-f
The above has the same behavior as
#!/usr/bin/txr -B -f
on a system which supports multiple arguments in the hash-bang line. The separator character is the colon, and so the remainder of that argument, -B:-f, is split into the two arguments -B -f.
The --eargs option is similar to --args, but must be followed by one more argument. After --eargs performs the argument splitting in the same manner as --args, any of the arguments which it produces which are the two-character sequence {} are replaced with that following argument. Whether or not the replacement occurs, that following argument is then removed.
Example:
#!/usr/bin/txr --eargs:-B:{}:--foo:42
This has an effect which cannot be replicated in any known implementation of the hash-bang mechanism. Suppose that this hash-bang line is placed in a script called script.txr. When this script is invoked with arguments, as in:
script.txr a b c
then TXR is invoked similarly to:
/usr/bin/txr --eargs:-B:{}:--foo:42 script.txr a b c
Then, when --eargs processing takes place, firstly the argument sequence
-B {} --foo 42
is produced by splitting into four fields using the : (colon) character as the separator. Then, within these four fields, all occurrences of {} are replaced with the following argument script.txr, resulting in:
-B script.txr --foo 42
Furthermore, that script.txr argument is removed from the remaining argument list.
The four arguments are then substituted in place of the original --eargs:-B:{}:--foo:42 syntax.
The resulting TXR invocation is, therefore:
/usr/bin/txr -B script.txr --foo 42 a b c
Thus, --eargs allows some arguments to be encoded into the interpreter script, such that script name is inserted anywhere among them, possibly multiple times. Arguments for the interpreter can be encoded, as well as arguments to be processed by the script.
#!/usr/bin/env txr
Here, the env utility searches for the txr program in the directories indicated by the PATH variable, which liberates the script from having to encode the exact location where the program is installed. However, if the operating system allows only one argument in the hash-bang mechanism, then no arguments can be passed to the program.
To mitigate this problem, TXR supports a special feature in its hash-bang support. If the hash-bang line contains a null byte, then the text from after the null byte until the end of the line is split into fields using the space character as a separator, and these fields are inserted into the command line. This manipulation happens during command-line processing, i.e. prior to the execution of the file. If this processing is applied to a file that is specified using the -f option, then the arguments which arise from the special processing are inserted after that option and its argument. If this processing is applied to the file which is the first non-option argument, then the options are inserted before that argument. However, care is taken not to process that argument a second time. In either situation, processing of the command-line options continues, and the arguments which are processed next are the ones which were just inserted. This is true even if the options had been inserted as a result of processing the first non-option argument, which would ordinarily signal the termination of option processing.
In the following examples, it is assumed that the script is named, and invoked, as /home/jenny/foo.txr, and is given arguments --bar abc, and that txr resolves to /usr/bin/txr. The <NUL> code indicates a literal ASCII NUL character (the zero byte).
Basic example:
#!/usr/bin/env txr<NUL>-a 3
Here, env searches for txr, finding it in /usr/bin. Thus, including the executable name, TXR receives this full argument list:
/usr/bin/txr /home/jenny/foo.txr --bar abc
The first non-option argument is the name of the script. TXR opens the script, and notices that it begins with a hash-bang line. It consumes the hash-bang line and finds the null byte inside it, retrieving the character string after it, which is "-a 3". This is split into the two arguments -a and 3, which are then inserted into the command line ahead of the the script name. The effective command line then becomes:
/usr/bin/txr -a 3 /home/jenny/foo.txr --bar abc
Command-line option processing continues, beginning with the -a option. After the option is processed, /home/jenny/foo.txr is encountered again. This time it is not opened a second time; it signals the end of option processing, exactly as it would immediately do if it hadn't triggered the insertion of any arguments.
Advanced example: use env to invoke txr, passing options to the interpreter and to the script:
#!/usr/bin/env txr<NUL>--eargs:-C:175:{}:--debug
This example shows how --eargs can be used in conjunction with the null hack. When txr begins executing, it receives the arguments
/usr/bin/txr /home/jenny/foo.txr
The script file is opened, and the arguments delimited by the null character in the hash-bang line are inserted, resulting in the effective command line:
/usr/bin/txr --eargs:-C:175:{}:--debug /home/jenny/foo.txr
Next, --eargs is processed in the ordinary way, transforming the command line into:
/usr/bin/txr -C 175 /home/jenny/foo.txr --debug
The name of the script file is encountered, and signals the end of option processing. Thus txr receives the -C option, instructing it to emulate some behaviors from version 175, and the /home/jenny/foo.txr script receives --debug as its argument: it executes with the *args* list containing one element, the character string "--debug".
The hash-bang null-hack feature was introduced in TXR 177. Previous versions ignore the hash-bang line, performing no special processing. Where a risk exists that programs which depend on the feature might be executed by an older version of TXR, care must be taken to detect and handle that situation, either by means of the txr-version variable, or else by some logic which infers that the processing of the hash-bang line hasn't been performed.
It is possible to use the Hash-Bang Null Hack, such that the resulting executable program recognizes TXR options. This is made possible by a special behavior in the processing of the -f option.
For instance, suppose that the effect of the following familiar hash-bang line is required:
#!/path/to/txr -f
However, suppose there is also a requirement to use the env utility to find TXR. Furthermore, the operating system allows only one hash-bang argument. Using the Null Hack, this is rewritten as:
#!/usr/bin/env txr<NUL>-f
then if the script is invoked with arguments -a b c, the command line will ultimately be transformed into:
/path/to/txr -f /path/to/scriptfile -i a b c
which allows TXR to process the -i option, leaving a, b and c as arguments for the script.
However, note that there is a subtle issue with the -f option that has been inserted via the Null Hack: namely, this insertion happens after TXR has opened the script file and read the hash-bang line from it. This means that when the inserted -f option is being processed, the script file is already open. A special behavior occurs. The -f option processing notices that the argument to -f is identical to the pathname of name of the script file that TXR has already opened for processing. The -f option and its argument are then skipped.
Outside of directives, whitespace is significant in TXR queries, and represents a pattern match for whitespace in the input. An extent of text consisting of an undivided mixture of tabs and spaces is a whitespace token.
Whitespace tokens match a precisely identical piece of whitespace in the input, with one exception: a whitespace token consisting of precisely one space has a special meaning. It is equivalent to the regular expression @/[ ]+/: match an extent of one or more spaces (but not tabs!). Multiple consecutive spaces do not have this meaning.
Thus, the query line "a b" (one space between a and b) matches "a b" with any number of spaces between the two letters.
For matching a single space, the syntax @\ can be used (backslash-escaped space).
It is more often necessary to match multiple spaces than to match exactly one space, so this rule simplifies many queries and inconveniences only a few.
In output clauses, string and character literals and quasiliterals, a space token denotes a space.
Query material which is not escaped by the special character @ is literal text, which matches input character for character. Text which occurs at the beginning of a line matches the beginning of a line. Text which starts in the middle of a line, other than following a variable, must match exactly at the current position, where the previous match left off. Moreover, if the text is the last element in the line, its match is anchored to the end of the line.
An empty query line matches an empty line in the input. Note that an empty input stream does not contain any lines, and therefore is not matched by an empty line. An empty line in the input is represented by a newline character which is either the first character of the file, or follows a previous newline-terminated line.
Input streams which end without terminating their last line with a newline are tolerated, and are treated as if they had the terminator.
Text which follows a variable has special semantics, described in the section Variables below.
A query may not leave a line of input partially matched. If any portion of a line of input is matched, it must be entirely matched, otherwise a matching failure results. However, a query may leave unmatched lines. Matching only four lines of a ten-line file is not a matching failure. The eof directive can be used to explicitly match the end of a file.
In the following example, the query matches the text, even though the text has an extra line.
In the following example, the query fails to match the text, because the text has extra material on one line that is not matched:
Needless to say, if the text has insufficient material relative to the query, that is a failure also.
To match arbitrary material from the current position to the end of a line, the "match any sequence of characters, including empty" regular expression @/.*/ can be used. Example:
In this example, the query matches, since the regular expression matches the string "of data". (See the Regular Expressions section below.)
Another way to do this is:
Control characters may be embedded directly in a query (with the exception of newline characters). An alternative to embedding is to use escape syntax. The following escapes are supported:
abcd@\
@\ efg
is equivalent to the line
abcd efg
The two spaces before the @\ in the second line are consumed. The spaces after are preserved.
Note that if a newline is embedded into a query line with @\n, this does not split the line into two; it's embedded into the line and thus cannot match anything. However, @\n may be useful in the @(cat) directive and in @(output).
TXR represents text internally using wide characters, which are used to represent Unicode code points. Script source code, as well as all data sources, are assumed to be in the UTF-8 encoding. In TXR and TXR Lisp source, extended characters can be used directly in comments, literal text, string literals, quasiliterals and regular expressions. Extended characters can also be expressed indirectly using hexadecimal or octal escapes. On some platforms, wide characters may be restricted to 16 bits, so that TXR can only work with characters in the BMP (Basic Multilingual Plane) subset of Unicode.
TXR does not use the localization features of the system library; its handling of extended characters is not affected by environment variables like LANG and L_CTYPE. The program reads and writes only the UTF-8 encoding.
TXR deals with UTF-8 separately in its parser and in its I/O streams implementation.
TXR's text streams perform UTF-8 conversion internally, such that TXR applications use Unicode code points.
In text streams, invalid UTF-8 bytes are treated as follows. When an invalid byte is encountered in the middle of a multibyte character, or if the input ends in the middle of a multibyte character, or if an invalid character is decoded, such as an overlong from, or code in the range U+DC00 through U+DCFF, the UTF-8 decoder returns to the starting byte of the ill-formed multibyte character, and extracts just one byte, mapping that byte to the Unicode character range U+DC00 through U+DCFF, producing that code point as the decoded result. The decoder is then reset to its initial state and begins decoding at the following byte, where the same algorithm is repeated.
Furthermore, because TXR internally uses a null-terminated character representation of strings which easily interoperates with C language interfaces, when a null character is read from a stream, TXR converts it to the code U+DC00. On output, this code converts back to a null byte, as explained in the previous paragraph. By means of this representational trick, TXR can handle textual data containing null bytes.
In contrast to the above, the TXR parser scans raw UTF-8 bytes from a binary stream, rather than using a text stream. The parser performing its own recognition of UTF-8 sequences in certain language constructs, using a UTF-8 decoder only when processing certain kinds of tokens.
Comments are read without regard for encoding, so invalid encoding bytes in comments are not detected. A comment is simply a sequence of bytes terminated by a newline.
Invalid UTF-8 encountered while scanning identifiers and character names in character literal (hash-backslash) syntax is diagnosed as a syntax error.
UTF-8 in string literals is treated in the same way as UTF-8 in text streams. Invalid UTF-8 bytes are mapped into code points in the U+DC000 through U+DCFF range, and incorporated as such into the resulting string object which the literal denotes. The same remarks apply to regular-expression literals.
In place of a piece of text (see section Text above), a regular-expression directive may be used, which has the following syntax:
@/RE/
where the RE part enclosed in slashes represents regular-expression syntax (described in the section Regular Expressions below).
Long regular expressions can be broken into multiple lines using a backslash-newline sequence. Whitespace before the sequence or after the sequence is not significant, so the following two are equivalent:
@/reg \
ular/
@/regular/
There may not be whitespace between the backslash and newline.
Whereas literal text simply represents itself, regular expression denotes a (potentially infinite) set of texts. The regular-expression directive matches the longest piece of text (possibly empty) which belongs to the set denoted by the regular expression. The match is anchored to the current position; thus if the directive is the first element of a line, the match is anchored to the start of a line. If the regular-expression directive is the last element of a line, it is anchored to the end of the line also: the regular expression must match the text from the current position to the end of the line.
Even if the regular expression matches the empty string, the match will fail if the input is empty, or has run out of data. For instance suppose the third line of the query is the regular expression @/.*/, but the input is a file which has only two lines. This will fail: the data has no line for the regular expression to match. A line containing no characters is not the same thing as the absence of a line, even though both abstractions imply an absence of characters.
Like text which follows a variable, a regular-expression directive which follows a variable has special semantics, described in the section Variables below.
Much of the query syntax consists of arbitrary text, which matches file data character for character. Embedded within the query may be variables and directives which are introduced by a @ character. Two consecutive @@ characters encode a literal @.
A variable-matching or substitution directive is written in one of several ways:
@sident
@{bident}
@*sident
@*{bident}
@{bident /regex/}
@{bident (fun [arg ...])}
@{bident number}
@{bident bident}
The forms with an * indicate a long match, see Longest Match below. The forms with the embedded regexp /regex/ or function or number have special semantics; see Positive Match below.
The identifier t cannot be used as a name; it is a reserved symbol which denotes the value true. An attempt to use the variable @t will result in an exception. The symbol nil can be used where a variable name is required syntactically, but it has special semantics, described in a section below.
A sident is a "simple identifier" form which is not delimited by braces.
A sident consists of any combination of one or more letters, numbers, and underscores. It may not look like a number, so that for instance 123 is not a valid sident, but 12A is valid. Case is sensitive, so that FOO is different from foo, which is different from Foo.
The braces around an identifier can be used when material which follows would otherwise be interpreted as being part of the identifier. When a name is enclosed in braces it is a bident.
The following additional characters may be used as part of a bident which are not allowed in a sident:
! $ % & * + - < = > ? \ ~
Moreover, most Unicode characters beyond U+007F may appear in a bident, with certain exceptions. A character may not be used if it is any of the Unicode space characters, a member of the high or low surrogate region, a member of any Unicode private-use area, or is either of the two characters U+FFFE and U+FFFF. These situations produce a syntax error. Invalid UTF-8 in an identifier is also a syntax error.
The rule still holds that a name cannot look like a number so +123 is not a valid bident but these are valid: a->b, *xyz*, foo-bar.
The syntax @FOO_bar introduces the name FOO_bar, whereas @{FOO}_bar means the variable named "FOO" followed by the text "_bar". There may be whitespace between the @ and the name, or opening brace. Whitespace is also allowed in the interior of the braces. It is not significant.
If a variable has no prior binding, then it specifies a match. The match is determined from some current position in the data: the character which immediately follows all that has been matched previously. If a variable occurs at the start of a line, it matches some text at the start of the line. If it occurs at the end of a line, it matches everything from the current position to the end of the line.
If a variable is one of the plain forms
@sident
@{bident}
@*sident
@*{bident}
then this is a "negative match". The extent of the matched text (the text bound to the variable) is determined by looking at what follows the variable, and ranges from the current position to some position where the following material finds a match. This is why this is called a "negative match": the spanned text which ends up bound to the variable is that in which the match for the trailing material did not occur.
A variable may be followed by a piece of text, a regular-expression directive, a function call, a directive, another variable, or nothing (i.e. occurs at the end of a line). These cases are described in detail below.
the variable a is considered to be followed by ":@/foo/bcd e".
If a variable is followed by text, then the extent of the negative match is determined by searching for the first occurrence of that text within the line, starting at the current position.
The variable matches everything between the current position and the matching position (not including the matching position). Any whitespace which follows the variable (and is not enclosed inside braces that surround the variable name) is part of the text. For example:
In the above example, the pattern text "a b " matches the data "a b ". So when the @FOO variable is processed, the data being matched is the remaining "c d e f". The text which follows @FOO is " e f". This is found within the data "c d e f" at position 3 (counting from 0). So positions 0–2 ("c d") constitute the matching text which is bound to FOO.
If the variable is followed by a function call, or a directive, the extent is determined by scanning the text for the first position where a match occurs for the entire remainder of the line. (For a description of functions, see Functions.)
For example:
@foo@(bind a "abc")xyz
Here, @foo will match the text from the current position to where "xyz" occurs, even though there is a @(bind) directive. Furthermore, if more material is added after the "xyz", it is part of the search. Note the difference between the following two:
@foo@/abc/@(func)
@foo@(func)@/abc/
In the first example, @foo matches the text from the current position until the match for the regular expression "abc". @(func) is not considered when processing @foo. In the second example, @foo matches the text from the current position until the position which matches the function call, followed by a match for the regular expression. The entire sequence @(func)@/abc/ is considered.
However, what if an unbound variable with no modifier is followed by another variable? The behavior depends on the nature of the other variable.
If the other variable is also unbound, and also has no modifier, this is a semantic error which will cause the query to fail. A diagnostic message will be issued, unless operating in quiet mode via -q. The reason is that there is no way to bind two consecutive variables to an extent of text; this is an ambiguous situation, since there is no matching criterion for dividing the text between two variables. (In theory, a repetition of the same variable, like @FOO@FOO, could find a solution by dividing the match extent in half, which would work only in the case when it contains an even number of characters. This behavior seems to have dubious value.)
An unbound variable may be followed by one which is bound. The bound variable is effectively replaced by the text which it denotes, and the logic proceeds accordingly.
It is possible for a variable to be bound to a regular expression. If x is an unbound variable and y is bound to a regular expression RE, then @x@y means @x@/RE/. A variable v can be bound to a regular expression using, for example, @(bind v #/RE/).
The @* syntax for longest match is available. Example:
Here, FOO is matched with "xyz", based on the delimiting around the colon. The colon in the pattern then matches the colon in the data, so that BAR is considered for matching against "defxyz". BAR is followed by FOO, which is already bound to "xyz". Thus "xyz" is located in the "defxyz" data following "def", and so BAR is bound to "def".
If an unbound variable is followed by a variable which is bound to a list, or nested list, then each character string in the list is tried in turn to produce a match. The first match is taken.
An unbound variable may be followed by another unbound variable which specifies a regular expression or function call match. This is a special case called a "double variable match". What happens is that the text is searched using the regular expression or function. If the search fails, then neither variable is bound: it is a matching failure. If the search succeeds, then the first variable is bound to the text which is skipped by the search. The second variable is bound to the text matched by the regular expression or function. Example:
This is treated just like the variable followed by directive. No semantic error is identified, even if both variables are unbound. Here, @var2 matches everything at the current position, and so @var1 ends up bound to the empty string.
Example 1: b matches at position 0 and a binds the empty string:
Example 2: *a specifies longest match (see Longest Match below), and so it takes everything:
In the former example, the match extends to the rightmost occurrence of "cd", and so FOO receives "b cdcdcd". In the latter example, the * syntax isn't used, and so a leftmost match takes place. The extent covers only the "b ", stopping at the first "cd" occurrence.
There are syntactic variants of variable syntax which have an embedded expression enclosed with the variable in braces:
@{bident /regex/}
@{bident (fun [args ...])}
@{bident number}
@{bident bident}
These specify a variable binding that is driven by a positive match derived from a regular expression, function or character count, rather than from trailing material (which is regarded as a "negative" match, since the variable is bound to material which is skipped in order to match the trailing material).
The positive match syntax is processed without considering any following syntax, and therefore may be followed by an unbound variable.
In the @{bident /regex/} form, the match extends over all characters from the current position which match the regular expression regex. (See the Regular Expressions section below.) If the variable already has a value, the text extracted by the regular expression must exactly match the variable.
In the @{bident (fun [args ...])} form, the match extends over lines or characters which are matched by the call to the function, if the call succeeds. Thus @{x (y z w)} is just like @(y z w), except that the region of text skipped over by @(y z w) is also bound to the variable x. Except in one special case, the matching takes place horizontally within the current line, and the spanned range of text is treated as a string. The exception is that if the @{bident (fun [args ...])} appears as the only element of a line, and fun has a binding as a vertical function, then the function is invoked in the same manner as it would be by the @(fun [args ...]) syntax. Then the variable indicated by bident is bound to the list of lines matched by the function call. Pattern functions are described in the Functions section below. The function is invoked even if the variable already has a value. The text matched by the function must match the variable.
In the @{bident number} form, the match processes a field of text which consists of the specified number of characters, which must be a nonnegative number. If the data line doesn't have that many characters starting at the current position, the match fails. A match for zero characters produces an empty string. The text which is actually bound to the variable is all text within the specified field, but excluding leading and trailing whitespace. If the field contains only spaces, then an empty string is extracted. This fixed-field extraction takes place whether or not the variable already has a binding. If it already has a binding, then it must match the extracted, trimmed text.
The @{bident bident} syntax allows the number or regex modifier to come from a variable. The variable must be bound and contain a nonnegative integer or regular expression. For example, @{x y} behaves like @{x 3} if y is bound to the integer 3. It is an error if y is unbound.
Just like in the Common Lisp language, the names nil and t are special.
nil symbol stands for the empty list object, an object which marks the end of a list, and Boolean false. It is synonymous with the syntax () which may be used interchangeably with nil in most constructs.
In TXR Lisp, nil and t cannot be used as variables. When evaluated, they evaluate to themselves.
In the TXR pattern language, nil can be used in the variable binding syntax, but does not create a binding; it has a special meaning. It allows the variable-matching syntax to be used to skip material, in ways similar to the skip directive.
The nil symbol is also used as a block name, both in the TXR pattern language and in TXR Lisp. A block named nil is considered to be anonymous.
Names beginning with the : (colon) character are keyword symbols. These also stand for themselves and may not be used as variables. Keywords are useful for labeling information and situations.
Regular expressions are a language for specifying sets of character strings. Through the use of pattern-matching elements, a regular expression is able to denote an infinite set of texts. TXR contains an original implementation of regular expressions, which supports the following syntax:
Any character which is not a regular-expression operator, a backslash escape, or the slash delimiter, denotes a one-position match of that character itself.
Any of the special characters, including the delimiting /, and the backslash, can be escaped with a backslash to suppress its meaning and denote the character itself.
Furthermore, all of the same escapes that are described in the section Special Characters in Text above are supported — the difference is that in regular expressions, the @ character is not required, so for example a tab is coded as \t rather than @\t. Octal and hex character escapes can be optionally terminated by a semicolon, which is useful if the following characters are octal or hex digits not intended to be part of the escape.
Only the above escapes are supported. Unlike in some other regular-expression implementations, if a backlash appears before a character which isn't a regex special character or one of the supported escape sequences, it is an error. This wasn't true of historic versions of TXR. See the COMPATIBILITY section.
| Operators | Class | Associativity |
| (R) [] | primary | |
| R? R+ R* R%... | postfix | left-to-right |
| R1R2 | catenation | left-to-right |
| ~R ...%R | unary | right-to-left |
| R1&R2 | intersection | left-to-right |
| R1|R2 | union | left-to-right |
The % operator is like a postfix operator with respect to its left operand, but like a unary operator with respect to its right operand. Thus a~b%c~d is a(~(b%(c(~d)))), demonstrating right-to-left associativity, where all of b% may be regarded as a unary operator being applied to c~d. Similarly, a?*+%b means (((a?)*)+)%b, where the trailing %b behaves like a postfix operator.
In TXR, regular expression matches do not span multiple lines. The regex language has no feature for multiline matching. However, the @(freeform) directive allows the remaining portion of the input to be treated as one string in which line terminators appear as explicit characters. Regular expressions may freely match through this sequence.
It's possible for a regular expression to match an empty string. For instance, if the next input character is z, facing the regular expression /a?/, there is a zero-character match: the regular expression's state machine can reach an acceptance state without consuming any characters. Examples:
In the first example, variable @A is followed by a regular expression which can match an empty string. The expression faces the letter z at position 0 in the data line. A zero-character match occurs there, therefore the variable A takes on the empty string. The @/.*/ regular expression then consumes the line.
Similarly, in the second example, the /a?/ regular expression faces a z, and thus yields an empty string which is bound to A. Variable @B consumes the entire line.
The third example requests the longest match for the variable binding. Thus, a search takes place for the rightmost position where the regular expression matches. The regular expression matches anywhere, including the empty string after the last character, which is the rightmost place. Thus variable A fetches the entire line.
For additional information about the advanced regular-expression operators, see NOTES ON EXOTIC REGULAR EXPRESSIONS below.
If the @ escape character is followed by an open parenthesis or square bracket, this is taken to be the start of a TXR Lisp compound expression.
The TXR language has the unusual property that its syntactic elements, so-called directives, are Lisp compound expressions. These expressions not only enclose syntax, but expressions which begin with certain symbols de facto behave as tokens in a phrase structure grammar. For instance, the expression @(collect) begins a block which must be terminated by the expression @(end), otherwise there is a syntax error. The collect expression can contain arguments which modify the behavior of the construct, for instance @(collect :gap 0 :vars (a b)). In some ways, this situation might be compared to HTML, in which an element such as <a> must be terminated by </a> and can have attributes such as <a href="...">.
Compound expressions contain subexpressions which are other compound expressions or literal objects of various kinds. Among these are: symbols, numbers, string literals, character literals, quasiliterals and regular expressions. These are described in the following sections. Additional kinds of literal objects exist, which are discussed in the TXR LISP section of the manual.
Some examples of compound expressions are:
(banana)
(a b c (d e f))
( a (b (c d) (e ) ))
("apple" #\b #\space 3)
(a #/[a-z]*/ b)
(_ `@file.txt`)
Symbols occurring in a compound expression follow a slightly more permissive lexical syntax than the bident in the syntax @{bident} introduced earlier. The / (slash) character may be part of an identifier, or even constitute an entire identifier. In fact a symbol inside a directive is a lident. This is described in the Symbol Tokens section under TXR LISP. A symbol must not be a number; tokens that look like numbers are treated as numbers and not symbols.
Character literals are introduced by the #\ (hash-backslash) syntax, which is either followed by a character name, the letter x followed by hex digits, the letter o followed by octal digits, or a single character. Valid character names are:
nul linefeed return
alarm newline esc
backspace vtab space
tab page pnul
For instance #\esc denotes the escape character.
This convention for character literals is similar to that of the Scheme language. Note that #\linefeed and #\newline are the same character. The #\pnul character is specific to TXR and denotes the U+DC00 code in Unicode; the name stands for "pseudo-null", which is related to its special function. For more information about this, see the section "Character Handling and International Characters".
String literals are delimited by double quotes. A double quote within a string literal is encoded using \" and a backslash is encoded as \\. Backslash escapes like \n and \t are recognized, as are hexadecimal escapes like \xFF or \xabc and octal escapes like \123. Ambiguity between an escape and subsequent text can be resolved by adding a semicolon delimiter after the escape: "\xabc;d" is a string consisting of the character U+0ABC followed by "d". The semicolon delimiter disappears. To write a literal semicolon immediately after a hex or octal escape, write two semicolons, the first of which will be interpreted as a delimiter. Thus, "\x21;;" represents "!;".
Note that the source code syntax of TXR string literals is specified in UTF-8, which is decoded into an internal string representation consisting of code points. The numeric escape sequences are an abstract syntax for specifying code points, not for specifying bytes to be inserted into the UTF-8 representation, even if they lie in the 8-bit range. Bytes cannot be directly specified, other than literally. However, when a TXR string object is encoded to UTF-8, every code point lying in the range U+DC00 through U+DCFF is converted to a single byte by taking the low-order eight bits of its value. By manipulating code points in this special range, TXR programs can reproduce arbitrary byte sequences in text streams. Also note that the \u escape sequence for specifying code points found in some languages is unnecessary and absent, since the existing hexadecimal and octal escapes satisfy this requirement. More detailed information is given in the earlier section Character Handling and International Characters.
If the line ends in the middle of a literal, it is an error, unless the last character is a backslash. This backslash is a special escape which does not denote a character; rather, it indicates that the string literal continues on the next line. The backslash is deleted, along with whitespace which immediately precedes it, as well as leading whitespace in the following line. The escape sequence "\ " (backslash space) can be used to encode a significant space.
Example:
"foo \
bar"
"foo \
\ bar"
"foo\ \
bar"
The first string literal is the string "foobar". The second two are "foo bar".
A word list literal (WLL) provides a convenient way to write a list of strings when such a list can be given as whitespace-delimited words.
There are two flavors of the WLL: the regular WLL which begins with #" (hash, double quote) and the splicing list literal which begins with #*" (hash, star, double quote).
Both types are terminated by a double quote, which may be escaped as \" in order to include it as a character. All the escaping conventions used in string literals can be used in word literals.
Unlike in string literals, whitespace (tabs and spaces) is not significant in word literals: it separates words. A whitespace character may be escaped with a backslash in order to include it as a literal character.
Just like in string literals, an unescaped newline character is not allowed. A newline preceded by a backslash is permitted. Such an escaped backslash, together with any leading and trailing unescaped whitespace, is removed and replaced with a single space.
Example:
#"abc def ghi" --> notates ("abc" "def" "ghi")
#"abc def \
ghi" --> notates ("abc" "def" "ghi")
#"abc\ def ghi" --> notates ("abc def" "ghi")
#"abc\ def\ \
\ ghi" --> notates ("abc def " " ghi")
A splicing word literal differs from a word literal in that it does not produce a list of string literals, but rather it produces a sequence of string literals that is merged into the surrounding syntax. Thus, the following two notations are equivalent:
(1 2 3 #*"abc def" 4 5 #"abc def")
(1 2 3 "abc" "def" 4 5 ("abc" "def"))
The regular WLL produced a single list object, but the splicing WLL expanded into multiple string literal objects.
Quasiliterals are similar to string literals, except that they may contain variable references denoted by the usual @ syntax. The quasiliteral represents a string formed by substituting the values of those variables into the literal template. If a is bound to "apple" and b to "banana", the quasiliteral `one @a and two @{b}s` represents the string "one apple and two bananas". A backquote escaped by a backslash represents itself. Unlike in directive syntax, two consecutive @ characters do not code for a literal @, but cause a syntax error. The reason for this is that compounding of the @ syntax is meaningful. Instead, there is a \@ escape for encoding a literal @ character. Quasiliterals support the full output variable syntax. Expressions within variable substitutions follow the evaluation rules of TXR Lisp. This hasn't always been the case: see the COMPATIBILITY section.
Quasiliterals can be split into multiple lines in the same way as ordinary string literals.
The quasiword list literals (QLLs) are to quasiliterals what WLLs are to ordinary literals. (See the above section Word List Literals.)
A QLL combines the convenience of the WLL with the power of quasistrings.
Just as in the case of WLLs, there are two flavors of the QLL: the regular QLL which begins with #` (hash, backquote) and the splicing QLL which begins with #*` (hash, star, backquote).
Both types are terminated by a backquote, which may be escaped as \` in order to include it as a character. All the escaping conventions used in quasiliterals can be used in QLLs.
Unlike in quasiliterals, whitespace (tabs and spaces) is not significant in QLLs: it separates words. A whitespace character may be escaped with a backslash in order to include it as a literal character.
A newline is not permitted unless escaped. An escaped newline works exactly the same way as it does in WLLs.
Note that the delimiting into words is done before the variable substitution. If the variable a contains spaces, then #`@a` nevertheless expands into a list of one item: the string derived from a.
Examples:
#`abc @a ghi` --> notates (`abc` `@a` `ghi`)
#`abc @d@e@f \
ghi` --> notates (`abc` `@d@e@f` `ghi`)
#`@a\ @b @c` --> notates (`@a @b` `@c`)
A splicing QLL differs from an ordinary QLL in that it does not produce a list of quasiliterals, but rather it produces a sequence of quasiliterals that is merged into the surrounding syntax.
TXR supports integers and floating-point numbers.
An integer literal is made up of digits 0 through 9, optionally preceded by a + or - sign. The character , (comma) may appear between digits, as a visual separator of no semantic significance. The digit sequence must start and end with a digit. Runs of consecutive commas are permitted. Commas outside of the digit sequence are interpreted as the Lisp unquote syntax.
Compatibility node: support for separator commas appeared in TXR 283. Older TXR versions will interpret commas in the middle of numeric constants as instances of the unquote syntax.
Examples:
123
-34
+0
-0
+234483527304983792384729384723234
-1,000,000,001
1,2,3,,4 ;; equivalent to 1234
Examples that are not integer tokens:
,123 ;; equivalent to (sys:unquote 123)
123,a ;; equivalent to 123, followed by (sys:unquote a)
-,1 ;; symbol - followed by (sys:unquote 1)
An integer constant can also be specified in hexadecimal using the prefix #x followed by an optional sign, followed by hexadecimal digits: 0 through 9 and the uppercase or lowercase letters A through F:
#xFF ;; 255
#x-ABC ;; -2748
These digits may contain separator commas, just as in the case of the decimal integer:
#xFFFF,FFFF,FFFF
Similarly, octal numbers are supported with the prefix #o followed by octal digits:
#o777 ;; 511
#o123,456 ;; 42797
and binary numbers can be written with a #b prefix:
#b1110 ;; 14
#b1111,1111 ;; 255
A comma between the radix prefix and digits is a syntax error:
#x,DEF5,549C ;; Syntax error
#b,1001,1101 ;; Likewise
Note that the #b prefix is also used for buffer literals.
A floating-point literal is marked by the inclusion of a decimal point, the scientific E notation, or both. It is an optional sign, followed by a mantissa consisting of digits, a decimal point, more digits, and then an optional E notation consisting of the letter e or E, an optional + or - sign, and then digits indicating the exponent value. In the mantissa, the digits are not optional. At least one digit must either precede the decimal point or follow it. That is to say, a decimal point by itself is not a floating-point constant.
The digits of the mantissa may include separator commas, in the same manner as decimal integer literals, in both the integer and fractional part. The digits of the exponent may not include separator commas.
Examples:
.123
123.
1E-3
20E40
.9E1
9.E19
-.5
+3E+3
1.E5
1,123,456.935,342E+013
Examples which are not floating-point constant tokens:
. ;; dot token, not a number
123E ;; the symbol 123E
1.0E- ;; syntax error: invalid floating point constant
1.0E ;; syntax error: invalid floating point constant
1.E ;; syntax error: invalid floating point literal
.e ;; syntax error: dot token followed by symbol
,1.0 ;; equivalent to (sys:unquote 1.0)
In TXR there is a special "dotdot" token consisting of two consecutive periods. An integer constant followed immediately by dotdot is recognized as such; it is not treated as a floating constant followed by a dot. That is to say, 123.. does not mean 123. . (floating point 123.0 value followed by dot token). It means 123 .. (integer 123 followed by .. token).
Dialect Note: unlike in Common Lisp, 123. is not an integer, but the floating-point number 123.0.
Integers within a certain small range centered on zero have fixnum type. Values in the fixnum range fit into a Lisp value directly, not requiring heap allocation. A value which is implemented as a reference to a heap-allocated object is called boxed, whereas a self-contained value not referencing any storage elsewhere is called unboxed. Thus values in the fixnum are unboxed; those outside of the range have bignum type instead, and are boxed. The variables fixnum-min and fixnum-max indicate the range.
Floating-point values are all unboxed if TXR is built with "NaN boxing" enabled, otherwise they are all boxed. The Lisp expression (eq (read "0.0") (read "0.0")) returns t under NaN boxing, indicating that the two instances of 0.0 are the same object. In the absence of NaN boxing, the two read calls produce distinct, boxed representations of 0.0, which compare unequal under eq. (The expression (eq 0.0 0.0) may not be relied upon if it is compiled, since compilation may deduplicate identical boxed literals, leading to a false positive.)
Comments of the form @; were introduced earlier. Inside compound expressions, another convention for comments exists: Lisp comments, which are introduced by the ; (semicolon) character and span to the end of the line.
Example:
@(foo ; this is a comment
bar ; this is another comment
)
This is equivalent to @(foo bar).
When a TXR Lisp compound expression occurs in TXR preceded by a @, it is a directive.
Directives which are based on certain symbols are, additionally, involved in a phrase-structure syntax which uses Lisp expressions as if they were tokens.
For instance, the directive
@(collect)
not only denotes a compound expression with the collect symbol in its head position, but it also introduces a syntactic phrase which requires a matching @(end) directive. In other words, @(collect) is not only an expression, but serves as a kind of token in a higher-level, phrase-structure grammar.
Effectively, collect is a reserved symbol in the TXR language. A TXR program cannot use this symbol as the name of a pattern function due to its role in the syntax. The symbol has no reserved role in TXR Lisp.
Usually if this type of directive occurs alone in a line, not preceded or followed by other material, it is involved in a "vertical" (or line-oriented) syntax.
If such a directive is embedded in a line (has preceding or trailing material) then it is in a horizontal syntactic and semantic context (character-oriented).
There is an exception: the definition of a horizontal function looks like this:
@(define name (arg))body material@(end)
Yet, this is considered one vertical item, which means that it does not match a line of data. (This is necessary because all horizontal syntax matches something within a line of data, which is undesirable for definitions.)
Many directives exhibit both horizontal and vertical syntax, with different but closely related semantics. Some are vertical only, some are horizontal only.
A summary of the available directives follows:
A collect is an anonymous block.
Named filters are stored in the hash table held in the Lisp special variable *filters*.
Some directives contain subexpressions which are evaluated. Two distinct styles of evaluations occur in TXR: bind expressions and Lisp expressions. Which semantics applies to an expression depends on the syntactic context in which it occurs: which position in which directive.
The evaluation of TXR Lisp expressions is described in the TXR LISP section of the manual.
Bind expressions are so named because they occur in the @(bind) directive. TXR pattern function invocations also treat argument expressions as bind expressions.
The @(rebind), @(set), @(merge), and @(deffilter) directives also use bind expression evaluation. Bind expression evaluation also occurs in the argument position of the :tlist keyword in the @(next) directive.
Unlike Lisp expressions, bind expressions do not support operators. If a bind expression is a nested list structure, it is a template denoting that structure. Any symbol in any position of that structure is interpreted as a variable. When the bind expression is evaluated, those corresponding positions in the template are replaced by the values of the variables.
Anywhere where a variable can appear in a bind expression's nested list structure, a Lisp expression can appear preceded by the @ character. That Lisp expression is evaluated and its value is substituted into the bind expression's template.
Moreover, a Lisp expression preceded by @ can be used as an entire bind expression. The value of that Lisp expression is then taken as the bind expression value.
Any object in a bind expression which is not a nested list structure containing Lisp expressions or variables denotes itself literally.
In the following examples, the variables a and b are assumed to have the string values "foo" and "bar", respectively.
The -> notation indicates the value of each expression.
a -> "foo"
(a b) -> ("foo" "bar")
((a) ((b) b)) -> (("foo") (("bar") "bar"))
(list a b) -> error: unbound variable list
@(list a b) -> ("foo" "bar") ;; Lisp expression
(a @[b 1..:]) -> ("foo" "ar") ;; Lisp eval of [b 1..:]
(a @(+ 2 2)) -> ("foo" 4) ;; Lisp eval of (+ 2 2)
#(a b) -> (a b) ;; Vector literal, not list.
[a b] -> error: unbound variable dwim
The last example above [a b] is a notation equivalent to (dwim a b) and so follows similarly to the example involving list.
The next directive indicates that the remaining directives in the current block are to be applied against a new input source.
It can only occur by itself as the only element in a query line, and takes various arguments, according to these possibilities:
@(next)
@(next source [:nothrow] [:noclose])
@(next :args)
@(next :env)
@(next :list lisp-expr)
@(next :tlist bind-expr)
@(next :string lisp-expr)
@(next :var var)
@(next nil)
The lone @(next) without arguments specifies that subsequent directives will match inside the next file in the argument list which was passed to TXR on the command line.
If source is given, it must be a TXR Lisp expression which denotes an input source. Its value may be a string or an input stream. For instance, if variable A contains the text "data", then @(next A) means switch to the file called "data", and @(next `@A.txt`) means to switch to the file "data.txt". The directive @(next (open-command `git log`)) switches to the input stream connected to the output of the git log command.
If the input source cannot be opened for whatever reason, TXR throws an exception (see Exceptions below). An unhandled exception will terminate the program. Often, such a drastic measure is inconvenient; if @(next) is invoked with the :nothrow keyword, then if the input source cannot be opened, the situation is treated as a simple match failure. The :nothrow keyword also ensures that when the stream is later closed, which occurs when the lazy list reads all of the available data, the implicit call to the close-stream function specifies nil as the argument value to that function's throw-on-error-p parameter. This :nothrow mechanism does not suppress all exceptions related to the processing of that stream; unusual conditions encountered during the reading of data from the stream may throw exceptions.
When the subsequent directives which follow @(next) are processed, the directive terminates, and any stream which had been opened for source is closed. If the :noclose keyword is present, then this is prevented; the stream remains open. Note: keeping the stream open may be necessary if the @(data) directive is used to capture the input list into a variable whose value is used after the @(next) directive terminates, because the input list is lazy, and may depend on the stream continuing to be open.
The variant @(next :args) means that the remaining command-line arguments are to be treated as a data source. For this purpose, each argument is considered to be a line of text. The argument list does include that argument which specifies the file that is currently being processed or was most recently processed. As the arguments are matched, they are consumed. This means that if a @(next) directive without arguments is executed in the scope of @(next :args), it opens the file named by the first unconsumed argument.
To process arguments, and then continue with the original file and argument list, wrap the argument processing in a @(block). When the block terminates, the input source and argument list are restored to what they were before the block.
The variant @(next :env) means that the list of process environment variables is treated as a source of data. It looks like a text file stream consisting of lines of the form "name=value". If this feature is not available on a given platform, an exception is thrown.
The syntax @(next :list lisp-expr) treats TXR Lisp expression lisp-expr as a source of text. The value of lisp-expr is flattened to a simple list in a way similar to the @(flatten) directive. The resulting list is treated as if it were the lines of a text file: each element of the list must be a string, which represents a line. If the strings happen contain embedded newline characters, they are a visible constituent of the line, and do not act as line separators.
The syntax @(next :tlist bind-expr) is similar to @(next :list ...) except that bind-expr is not a TXR Lisp expression, but a TXR bind expression.
The syntax @(next :var var) requires var to be a previously bound variable. The value of the variable is retrieved and treated like a list, in the same manner as under @(next :list ...). Note that @(next :var x) is not always the same as @(next :tlist x), because :var x strictly requires x to be a TXR variable, whereas the x in :tlist x is an expression which can potentially refer to Lisp variable.
The syntax @(next :string lisp-expr) treats expression lisp-expr as a source of text. The value of the expression must be a string. Newlines in the string are interpreted as line terminators.
A string which is not terminated by a newline is tolerated, so that:
@(next :string "abc")
@a
binds a to "abc". Likewise, this is also the case with input files and other streams whose last line is not terminated by a newline.
However, watch out for empty strings, which are analogous to a correctly formed empty file which contains no lines:
@(next :string "")
@a
This will not bind a to ""; it is a matching failure. The behavior of :list is different. The query
@(next :list "")
@a
binds a to "". The reason is that under :list the string "" is flattened to the list ("") which is not an empty input stream, but a stream consisting of one empty line.
The @(next nil) variant indicates that the following subquery is applied to empty data, and the list of data sources from the command line is considered empty. This directive is useful in front of TXR code which doesn't process data sources from the command line, but takes command-line arguments. The @(next nil) incantation absolutely prevents TXR from trying to open the first command-line argument as a data source.
Note that the @(next) directive only redirects the source of input over the scope of subquery in which the that directive appears. For example, the following query looks for the line starting with "xyz" at the top of the file "foo.txt", within a some directive. After the @(end) which terminates the @(some), the "abc" is matched in the previous input stream which was in effect before the @(next) directive:
@(some)
@(next "foo.txt")
xyz@suffix
@(end)
abc
However, if the @(some) subquery successfully matched "xyz@suffix" within the file foo.text, there is now a binding for the suffix variable, which is visible to the remainder of the entire query. The variable bindings survive beyond the clause, but the data stream does not.
The skip directive considers the remainder of the query as a search pattern. The remainder is no longer required to strictly match at the current line in the current input stream. Rather, the current stream is searched, starting with the current line, for the first line where the entire remainder of the query will successfully match. If no such line is found, the skip directive fails. If a matching position is found, the remainder of the query is processed from that point.
The remainder of the query can itself contain skip directives. Each such directive performs a recursive subsearch.
Skip comes in vertical and horizontal flavors. For instance, skip and match the last line:
Skip and match the last character of the line:
@(skip)@{last 1}@(eol)
The skip directive has two optional arguments, which are evaluated as TXR Lisp expressions. If the first argument evaluates to an integer, its value limits the range of lines scanned for a match. Judicious use of this feature can improve the performance of queries.
Example: scan until "size: @SIZE" matches, which must happen within the next 15 lines:
@(skip 15)
size: @SIZE
Without the range limitation, skip will keep searching until it consumes the entire input source. In a horizontal skip, the range-limiting numeric argument is expressed in characters, so that
abc@(skip 5)def
means: there must be a match for "abc" at the start of the line, and then within the next five characters, there must be a match for "def".
Sometimes a skip is nested within a collect, or following another skip. For instance, consider:
@(collect)
begin @BEG_SYMBOL
@(skip)
end @BEG_SYMBOL
@(end)
The above collect iterates over the entire input. But, potentially, so does the embedded skip. Suppose that "begin x" is matched, but the data has no matching "end x". The skip will search in vain all the way to the end of the data, and then the collect will try another iteration back at the beginning, just one line down from the original starting point. If it is a reasonable expectation that an end x occurs 15 lines of a "begin x", this can be specified instead:
@(collect)
begin @BEG_SYMBOL
@(skip 15)
end @BEG_SYMBOL
@(end)
If the symbol nil is used in place of a number, it means to scan an unlimited range of lines; thus, @(skip nil) is equivalent to @(skip).
If the symbol :greedy is used, it changes the semantics of the skip to longest match semantics. For instance, match the last three space-separated tokens of the line:
@(skip :greedy) @a @b @c
Without :greedy, the variable @c may match multiple tokens, and end up with spaces in it, because nothing follows @c and so it matches from any position which follows a space to the end of the line. Also note the space in front of @a. Without this space, @a will get an empty string.
A line-oriented example of greedy skip: match the last line without using @(eof):
@(skip :greedy)
@last_line
There may be a second numeric argument. This specifies a minimum number of lines to skip before looking for a match. For instance, skip 15 lines and then search indefinitely for begin ...:
@(skip nil 15)
begin @BEG_SYMBOL
The two arguments may be used together. For instance, the following matches if and only if the 15th line of input starts with begin :
@(skip 1 15)
begin @BEG_SYMBOL
Essentially, @(skip 1 n) means "hard skip by n lines". @(skip 1 0) is the same as @(skip 1), which is a noop, because it means: "the remainder of the query must match starting on the next line", or, more briefly, "skip exactly zero lines", which is the behavior if the skip directive is omitted altogether.
Here is one trick for grabbing the fourth line from the bottom of the input:
@(skip)
@fourth_from_bottom
@(skip 1 3)
@(eof)
Or using greedy skip:
@(skip :greedy)
@fourth_from_bottom
@(skip 1 3)
Non-greedy skip with the @(eof) directive has a slight advantage because the greedy skip will keep scanning even though it has found the correct match, then backtrack to the last good match once it runs out of data. The regular skip with explicit @(eof) will stop when the @(eof) matches.
The skip directive can consume considerable CPU time when multiple skips are nested. Consider:
This is actually nesting: the second and third skips occur within the body of the first one, and thus this creates nested iteration. TXR is searching for the combination of skips which match the pattern of lines A, B and C with backtracking behavior. The outermost skip marches through the data until it finds A followed by a pattern match for the second skip. The second skip iterates to find B followed by the third skip, and the third skip iterates to find C. If A and B are only one line each, then this is reasonably fast. But suppose there are many lines matching A and B, giving rise to a large number of combinations of skips which match A and B, and yet do not find a match for C, triggering backtracking. The nested stepping which tries the combinations of A and B can give rise to a considerable running time.
One way to deal with the problem is to unravel the nesting with the help of blocks. For example:
@(block)
@ (skip)
A
@(end)
@(block)
@ (skip)
B
@(end)
@(skip)
C
Now the scope of each skip is just the remainder of the block in which it occurs. The first skip finds A, and then the block ends. Control passes to the next block, and backtracking will not take place to a block which completed (unless all these blocks are enclosed in some larger construct which backtracks, causing the blocks to be re-executed.
This rewrite is not equivalent, and cannot be used for instance in backreferencing situations such as:
@;
@; Find three lines anywhere in the input which are identical.
@;
@(skip)
@line
@(skip)
@line
@(skip)
@line
This example depends on the nested search-within-search semantics.
The trailer directive introduces a trailing portion of a query or subquery which matches input material normally, but in the event of a successful match, does not advance the current position. This can be used, for instance, to cause @(collect) to match partially overlapping regions.
Trailer can be used in vertical context:
@(trailer)
directives
...
or horizontal:
@(trailer) directives ...
A vertical trailer prevents the vertical input position from advancing as it is matched by directives, whereas a horizontal trailer prevents the horizontal position from advancing. In other words, trailer performs matching without consuming the input, providing a lookahead mechanism.
Example:
@(collect)
@line
@(trailer)
@(skip)
@line
@(end)
This script collects each line which has a duplicate somewhere later in the input. Without the @(trailer) directive, this does not work properly for inputs like:
111
222
111
222
Without @(trailer), the first duplicate pair constitutes a match which spans over the 222. After that pair is found, the matching continues after the second 111.
With the @(trailer) directive in place, the collect body, on each iteration, only consumes the lines matched prior to @(trailer).
The freeform directive provides a useful alternative to TXR's line-oriented matching discipline. The freeform directive treats all remaining input from the current input source as one big line. The query line which immediately follows freeform is applied to that line.
The syntax variations are:
@(freeform)
... query line ..
@(freeform
number)
... query line ..
@(freeform string)
... query line ..
@(freeform number string)
... query line ..
where number and string denote TXR Lisp expressions which evaluate to an integer or string value, respectively.
If number and string are both present, they may be given in either order.
If the number argument is given, its value limits the range of lines which are combined together. For instance @(freeform 5) means to only consider the next five lines to be one big line. Without this argument, freeform is "bottomless". It can match the entire file, which creates the risk of allocating a large amount of memory.
If the string argument is given, it specifies a custom line terminator. The default terminator is "\n". The terminator does not have to be one character long.
Freeform does not convert the entire remainder of the input into one big line all at once, but does so in a dynamic, lazy fashion, which takes place as the data is accessed. So at any time, only some prefix of the data exists as a flat line in which newlines are replaced by the terminator string, and the remainder of the data still remains as a list of lines.
After the subquery is applied to the virtual line, the unmatched remainder of that line is broken up into multiple lines again, by looking for and removing all occurrences of the terminator string within the flattened portion.
Care must be taken if the terminator is other than the default "\n". All occurrences of the terminator string are treated as line terminators in the flattened portion of the data, so extra line breaks may be introduced. Likewise, in the yet unflattened portion, no breaking takes place, even if the text contains occurrences of the terminator string. The extent of data which is flattened, and the amount of it which remains, depends entirely on the query line underneath @(flatten).
In the following example, lines of data are flattened using $ as the line terminator.
The data is turned into the virtual line 1$2:3$4$. The @a$@b: subquery matches the 1$2: portion, binding a to "1", and b to "2". The remaining portion 3$4$ is then split into separate lines again according to the line terminator $i:
3
4
Thus the remainder of the query
@c
@d
faces these lines, binding c to 3 and d to 4. Note that since the data does not contain dollar signs, there is no ambiguity; the meaning may be understood in terms of the entire data being flattened and split again.
In the following example, freeform is used to solve a tokenizing problem. The Unix password file has fields separated by colons. Some fields may be empty. Using freeform, we can join the password file using ":" as a terminator. By restricting freeform to one line, we can obtain each line of the password file with a terminating ":", allowing for a simple tokenization, because now the fields are colon-terminated rather than colon-separated.
Example:
@(next "/etc/passwd")
@(collect)
@(freeform 1 ":")
@(coll)@{token /[^:]*/}:@(end)
@(end)
The fuzz directive allows for an imperfect match spanning a set number of lines. It takes two arguments, both of which are TXR Lisp expressions that should evaluate to integers:
@(fuzz m n)
...
This expresses that over the next n query lines, the matching strictness is relaxed a little bit. Only m out of those n lines have to match. Afterward, the rest of the query follows normal, strict processing.
In the degenerate situation where there are fewer than n query lines following the fuzz directive, then m of them must succeed anyway. (If there are fewer than m, then this is impossible.)
The line and chr directives perform binding between the current input line number or character position within a line, against an expression or variable:
@(line 42)
@(line x)
abc@(chr 3)def@(chr y)
The directive @(line 42) means "match the current input line number against the integer 42". If the current line is 42, then the directive matches, otherwise it fails. line is a vertical directive which doesn't consume a line of input. Thus, the following matches at the beginning of an input stream, and x ends up bound to the first line of input:
@(line 1)
@(line 1)
@(line 1)
@x
The directive @(line x) binds variable x to the current input line number, if x is an unbound variable. If x is already bound, then the value of x must match the current line number, otherwise the directive fails.
The chr directive is similar to line except that it's a horizontal directive, and matches the character position rather than the line position. Character positions are measured from zero, rather than one. chr does not consume a character. Hence the two occurrences of chr in the following example both match, and x takes the entire line of input:
The argument of line or chr may be an @-delimited Lisp expression. This is useful for matching computed lines or character positions:
The name directive performs a binding between the name of the current data source and a variable or bind expression:
If na is an unbound variable, it is bound and takes on the name of the data source, such as a file name. If na is bound, then it has to match the name of the data source, otherwise the directive fails.
The directive @(name "data.txt") fails unless the current data source has that name.
The data directive performs a binding between the unmatched data at the current position, and and a variable or bind expression. The unmatched data takes the form of a list of strings:
@(data d)
The binding is performed on object equality. If d is already bound, a matching failure occurs unless d contains the current unmatched data.
Matching the current data has various uses.
For instance, two branches of pattern matching can, at some point, bind the current data into different variables. When those paths join, the variables can be bound together to create the assertion that the current data had been the same at those points:
@(all)
@ (skip)
foo
@ (skip)
bar
@ (data x)
@(or)
@ (skip)
xyzzy
@ (skip)
bar
@ (data y)
@(end)
@(require (eq x y))
Here, two branches of the @(all) match some material which ends in the line bar. However, it is possible that this is a different line. The data directives are used to create an assertion that the data regions matched by the two branches are identical. That is to say, the unmatched data x captured after the first bar and the unmatched data y captured after the second bar must be the same object in order for @(require (eq x y)) to succeed, which implies that the same bar was matched in both branches of the @(all).
Another use of data is simply to gain access to the trailing remainder of the unmatched input in order to print it, or do some special processing on it.
The tprint Lisp function is useful for printing the unmatched data as newline-terminated lines:
@(data remainder)
@(do (tprint remainder))
The eof directive, if not given any argument, matches successfully when no more input is available from the current input source.
In the following example, the line variable captures the text "One-line file" and then since that is the last line of input, the eof directive matches:
If the data consisted of two or more lines, eof would fail.
The eof directive may be given a single argument, which is a pattern that matches the termination status of the input source. This is useful when the input source is a process pipe. For the purposes of eof, sources which are not process pipes have the symbol t as their termination status.
In the following example, which assumes the availability of a POSIX shell command interpreter in the host system, the variable a captures the string "a" and the status variable captures the integer value 5, which is the termination status of the command:
@(next (open-command "echo a; exit 5"))
@a
@(eof status)
These directives, called the parallel directives, combine multiple subqueries, which are applied at the same input position, rather than to consecutive input.
They come in vertical (line mode) and horizontal (character mode) flavors.
In horizontal mode, the current position is understood to be a character position in the line being processed. The clauses advance this character position by moving it to the right. In vertical mode, the current position is understood to be a line of text within the stream. A clause advances the position by some whole number of lines.
The syntax of these parallel directives follows this example:
@(some)
subquery1
.
.
.
@(and)
subquery2
.
.
.
@(and)
subquery3
.
.
.
@(end)
And in horizontal mode:
@(some)subquery1...@(and)subquery2...@(and)subquery3...@(end)
Long horizontal lines can be broken up with line continuations, allowing the above example to be written like this, which is considered a single logical line:
@(some)@\
subquery1...@\
@(and)@\
subquery2...@\
@(and)@\
subquery3...@\
@(end)
The @(some), @(all), @(none), @(maybe), @(cases) or @(choose) must be followed by at least one subquery clause, and be terminated by @(end). If there are two or more subqueries, these additional clauses are indicated by @(and) or @(or), which are interchangeable. The separator and terminator directives also must appear as the only element in a query line.
The choose directive requires keyword arguments. See below.
The syntax supports arbitrary nesting. For example:
QUERY: SYNTAX TREE:
@(all) all -+
@ (skip) +- skip -+
@ (some) | +- some -+
it | | +- TEXT
@ (and) | | +- and
@ (none) | | +- none -+
was | | | +- TEXT
@ (end) | | | +- end
@ (end) | | +- end
a dark | +- TEXT
@(end) *- end
nesting can be indicated using whitespace between @ and the directive expression. Thus, the above is an @(all) query containing a @(skip) clause which applies to a @(some) that is followed by the text line "a dark". The @(some) clause combines the text line "it", and a @(none) clause which contains just one clause consisting of the line "was".
The semantics of the parallel directives is:
The :resolve parameter is for situations when the @(some) directive has multiple clauses that need to bind some common variables to different values: for instance, output parameters in functions. Resolve takes a list of variable name symbols as an argument. This is called the resolve set. If the clauses of @(some) bind variables in the resolve set, those bindings are not visible to later clauses. However, those bindings do emerge out of the @(some) directive as a whole. This creates a conflict: what if two or more clauses introduce different bindings for a variable in the resolve set? This is why it is called the resolve set: conflicts for variables in the resolve set are automatically resolved in favor of later directives.
Example:
@(some :resolve (x))
@ (bind a "a")
@ (bind x "x1")
@(or)
@ (bind b "b")
@ (bind x "x2")
@(end)
Here, the two clauses both introduce a binding for x. Without the :resolve parameter, this would mean that the second clause fails, because x comes in with the value "x1", which does not bind with "x2". But because x is placed into the resolve set, the second clause does not see the "x1" binding. Both clauses establish their bindings independently creating a conflict over x. The conflict is resolved in favor of the second clause, and so the bindings which emerge from the directive are:
a="a"
b="b"
x="x2"
For all of the parallel directives other than @(none) and @(choose), the query advances the input position by the greatest number of lines that match in any of the successfully matching subclauses that are evaluated. The @(none) directive does not advance the input position.
For instance if there are two subclauses, and one of them matches three lines, but the other one matches five lines, then the overall clause is considered to have made a five line match at its position. If more directives follow, they begin matching five lines down from that position.
The syntax of @(require) is:
@(require lisp-expression)
The require directive evaluates a TXR Lisp expression. (See TXR LISP far below.) If the expression yields a true value, then it succeeds, and matching continues with the directives which follow. Otherwise the directive fails.
In the context of the require directive, the expression should not be introduced by the @ symbol; it is expected to be a Lisp expression.
Example:
@; require that 4 is greater than 3
@; This succeeds; therefore, @a is processed
@(require (> (+ 2 2) 3))
@a
The if directive allows for conditional selection of pattern-matching clauses, based on the Boolean results of Lisp expressions.
A variant of the if directive is also available for use inside an output clauses, where it similarly allows for the conditional selection of output clauses.
The syntax of the if directive can be exemplified as follows:
@(if lisp-expr)
.
.
.
@(elif lisp-expr)
.
.
.
@(elif lisp-expr)
.
.
.
@(else)
.
.
.
@(end)
The @(elif) and @(else) clauses are all optional. If @(else) is present, it must be last, before @(end), after any @(elif) clauses. Any of the clauses may be empty.
@(if (> (length str) 42))
foo: @a @b
@(else)
{@c}
@(end)
In this example, if the length of the variable str is greater than 42, then matching continues with "foo: @a b", otherwise it proceeds with {@c}.
More precisely, how the if directive works is as follows. The Lisp expressions are evaluated in order, starting with the if expression, then the elif expressions if any are present. If any Lisp expression yields a true result (any value other than nil) then evaluation of Lisp expressions stops. The corresponding clause of that Lisp expression is selected and pattern matching continues with that clause. The result of that clause (its success or failure, and any newly bound variables) is then taken as the result of the if directive. If none of the Lisp expressions yield true, and an else clause is present, then that clause is processed and its result determines the result of the if directive. If none of the Lisp expressions yield true, and there is no else clause, then the if directive is deemed to have trivially succeeded, allowing matching to continue with whatever directive follows it.
The @(output) directive supports the embedding of Lisp expressions, whose values are interpolated into the output. In particular, Lisp if expressions are useful. For instance @(if expr "A" "B") reproduces A if expr yields a true value, otherwise B. Yet the @(if) directive is also supported in @(output). How the apparent conflict between the two is resolved is that the two take different numbers of arguments. An @(if) which has no arguments at all is a syntax error. One that has one argument is the head of the if directive syntax which must be terminated by @(end) and which takes the optional @(elif) and @(else) clauses. An @(if) which has two or more arguments is parsed as a self-contained Lisp expression.
Sometimes text is structured as items that can appear in an arbitrary order. When multiple matches need to be extracted, there is a combinatorial explosion of possible orders, making it impractical to write pattern matches for all the possible orders.
The gather directive is for these situations. It specifies multiple clauses which all have to match somewhere in the data, but in any order.
For further convenience, the lines of the first clause of the gather directive are implicitly treated as separate clauses.
The syntax follows this pattern:
@(gather)
one-line-query1
one-line-query2
.
.
.
one-line-queryN
@(and)
multi
line
query1
.
.
.
@(and)
multi
line
query2
.
.
.
@(end)
The multiline clauses are optional. The gather directive takes keyword parameters, see below.
Similarly to collect, gather has an optional until/last clause:
@(gather)
...
@(until)
...
@(end)
How gather works is that the text is searched for matches for the single-line and multiline queries. The clauses are applied in the order in which they appear. Whenever one of the clauses matches, any bindings it produces are retained and it is removed from further consideration. Multiple clauses can match at the same text position. The position advances by the longest match from among the clauses which matched. If no clauses match, the position advances by one line. The search stops when all clauses are eliminated, and then the cumulative bindings are produced. If the data runs out, but unmatched clauses remain, the directive fails.
Example: extract several environment variables, which do not appear in a particular order:
@(next :env)
@(gather)
USER=@USER
HOME=@HOME
SHELL=@SHELL
@(end)
If the until or last clause is present and a match occurs, then the matches from the other clauses are discarded and the gather terminates. The difference between until/last is that any bindings bindings established in last are retained, and the input position is advanced past the matching material. The until/last clause has visibility to bindings established in the previous clauses in that same iteration, even though those bindings end up thrown away.
For consistency, the :mandatory keyword is supported in the until/last clause of gather. The semantics of using :mandatory in this situation is tricky. In particular, if it is in effect, and the gather terminates successfully by collecting all required matches, it will trigger a failure. On the other hand, if the until or last clause activates before all required matches are gathered, a failure also occurs, whether or not the clause is :mandatory.
Meaningful use of :mandatory requires that the gather be open-ended; it must allow some (or all) variables not to be required. The presence of the option means that for gather to succeed, all required variables must be gathered first, but then termination must be achieved via the until/last clause before all gather clauses are satisfied.
Example:
@(gather :vars (a b c (d "foo")))
...
@(end)
Here, a, b and c are required variables, and d is optional, with the default value given by the Lisp expression "foo".
The presence of :vars changes the behavior in three ways.
Firstly, even if all the clauses in the gather match successfully and are eliminated, the directive will fail if the required variables do not have bindings. It doesn't matter whether the bindings are existing, or whether they are established by gather.
Secondly, if some of the clauses of gather did not match, but all of the required variables have bindings, then the directive succeeds. Without the presence of :vars, it would fail in this situation.
Thirdly, if gather succeeds (all required variables have bindings), then all of the optional variables which do not have bindings are given bindings to their default values.
The expressions which give the default values are evaluated whenever the gather directive is evaluated, whether or not their values are used.
The syntax of the collect directive is:
@(collect)
... lines of subquery
@(end)
or with an until or last clause:
@(collect)
... lines of subquery: main clause
@(until)
... lines of subquery: until clause
@(end)
@(collect)
... lines of subquery: main clause
@(last)
... lines of subquery: last clause
@(end)
The repeat symbol may be specified instead of collect, which changes the meaning:
@(repeat)
... lines of subquery
@(end)
The @(repeat) syntax is equivalent to @(collect :vars nil) and doesn't take the :vars clause. It accepts other collect parameters.
The subquery is matched repeatedly, starting at the current line. If it fails to match, it is tried starting at the subsequent line. If it matches successfully, it is tried at the line following the entire extent of matched data, if there is one. Thus, the collected regions do not overlap. (Overlapping behavior can be obtained: see the @(trailer) directive.)
Unless certain keywords are specified, or unless the collection is explicitly failed with @(fail), it always succeeds, even if it collects nothing, and even if the until/last clause never finds a match.
If no until/last clause is specified, and the collect is not limited using parameters, the collection is unbounded: it consumes the entire data file.
If an until/last clause is specified, the collection stops when that clause matches at the current position.
If an until clause terminates collect, no bindings are collected at that position, even if the main clause matches at that position also. Moreover, the position is not advanced. The remainder of the query begins matching at that position.
If a last clause terminates collect, the behavior is different. Any bindings captured by the main clause are thrown away, just like with the until clause. However, the bindings in the last clause itself survive, and the position is advanced to skip over that material.
Example:
The line 42 is not collected, even though it matches @a. Furthermore, the @(until) does not advance the position, so variable c takes 42.
If the @(until) is changed to @(last) the output will be different:
The 42 is not collected into a list, just like before. But now the binding captured by @b emerges. Furthermore, the position advances so variable now takes 6.
The binding variables within the clause of a collect are treated specially. The multiple matches for each variable are collected into lists, which then appear as array variables in the final output.
Example:
The query matches the data in three places, so each variable becomes a list of three elements, reported as an array.
Variables with list bindings may be referenced in a query. They denote a multiple match. The -D command-line option can establish a one-dimensional list binding.
The clauses of collect may be nested. Variable matches collated into lists in an inner collect are again collated into nested lists in the outer collect. Thus an unbound variable wrapped in N nestings of @(collect) will be an N-dimensional list. A one-dimensional list is a list of strings; a two-dimensional list is a list of lists of strings, etc.
It is important to note that the variables which are bound within the main clause of a collect, that is, the variables which are subject to collection, appear, within the collect, as normal one-value bindings. The collation into lists happens outside of the collect. So for instance in the query:
The left @x establishes a binding for some material preceding an equal sign. The right @x refers to that binding. The value of @x is different in each iteration, and these values are collected. What finally comes out of the collect clause is a single variable called x which holds a list containing each value that was ever instantiated under that name within the collect clause.
Also note that the until clause has visibility over the bindings established in the main clause. This is true even in the terminating case when the until clause matches, and the bindings of the main clause are discarded.
Within the @(collect) syntax, it is possible to specify keyword parameters for additional control of the behavior. A keyword parameter consist of a keyword symbol followed by an argument, enclosed within the @(collect) syntax. The following are the supported keywords.
@(collect :maxgap 5)
specifies that the gap between the current position and the first match for the body of the collect, or between consecutive matches can be no longer than five lines. A :maxgap value of 0 means that the collected regions must be adjacent and must match right from the starting position. For instance:
@(collect :maxgap 0)
M @a
@(end)
means: from here, collect consecutive lines of the form "M ...". This will not search for the first such line, nor will it skip lines which do not match this form.
means: collect every other line starting with the current line.
@(collect :lines 2)
foo: @a
bar: @b
baz: @c
@(end)
The above collect will look for a match only twice: at the current position, and one line down.
If there is an existing binding for variable prior to the processing of the collect, then the variable is shadowed.
The binding is collected in the same way as other bindings that are established in the collect body.
The repetition count only increments after a successful match.
The variable is visible to the collect's until/last clause. If that clause is being processed after a successful match of the body, then variable holds an integer value. If the body fails to match, then the until/last clause sees a binding for variable with a value of nil.
The :vars keyword allows the query writer to add discipline the collect body.
The argument to :vars is a list of variable specs. A variable spec is either a symbol, denoting a required variable, or a (symbol default-value) pair, where default-value is a Lisp expression whose value specifies a default value for the variable, which is optional.
When a :vars list is specified, it means that only the given variables can emerge from the successful collect. Any newly introduced bindings for other variables do not propagate. More precisely, whenever the collect body matches successfully, the following three rules apply:
In the event that collect does not match anything, the variables specified in :vars, whether required or optional, are all bound to empty lists. These bindings are established after the processing of the until/last clause, if present.
Example:
@(collect :vars (a b (c "foo")))
@a @c
@(end)
Here, if the body "@a @c" matches, an error will be thrown because one of the mandatory variables is b, and the body neglects to produce a binding for b.
Example:
@(collect :vars (a (c "foo")))
@a @b
@(end)
Here, if "@a @b" matches, only a will be collected, but not b, because b is not in the variable list. Furthermore, because there is no binding for c in the body, a binding is created with the value "foo", exactly as if c matched such a piece of text.
In the following example, the assumption is that THIS NEVER MATCHES is not found anywhere in the input but the line THIS DOES MATCH is found and has a successor which is bound to a. Because the body did not match, the :vars a and b should be bound to empty lists. But a is bound by the last clause to some text, so this takes precedence. Only b is bound to an empty list.
@(collect :vars (a b))
THIS NEVER MATCHES
@(last)
THIS DOES MATCH
@a
@(end)
The following means: do not allow any variables to propagate out of any iteration of the collect and therefore collect nothing:
@(collect :vars nil)
...
@(end)
Instead of writing @(collect :vars nil), it is possible to write @(repeat). @(repeat) takes all collect keywords, except for :vars. There is a @(repeat) directive used in @(output) clauses; that is a different directive.
The until/last clause supports the option keyword :mandatory, exemplified by the following:
@(collect)
...
@(last :mandatory)
...
@(end)
This means that the collect must be terminated by a match for the until/last clause, or else by an explicit @(accept).
Specifically, the collect cannot terminate due to simply running out of data, or exceeding a limit on the number of matches that may be collected. In those situations, if an until or last clause is present with :mandatory, the collect is deemed to have failed.
The coll directive is the horizontal version of collect. Whereas collect works with multiline clauses on line-oriented material, coll works within a single line. With coll, it is possible to recognize repeating regularities within a line and collect lists.
Regular-expression-based Positive Match variables work well with coll.
Example: collect a comma-separated list, terminated by a space.
Here, the variable A is bound to tokens which match the regular expression /[^, ]+/: nonempty sequence of characters other than commas or spaces.
Like collect, coll searches for matches. If no match occurs at the current character position, it tries at the next character position. Whenever a match occurs, it continues at the character position which follows the last character of the match, if such a position exists.
If not bounded by an until clause, it will exhaust the entire line. If the until clause matches, then the collection stops at that position, and any bindings from that iteration are discarded. Like collect, coll also supports an until/last clause, which propagates variable bindings and advances the position. The :mandatory keyword is supported.
coll clauses nest, and variables bound within a coll are available to clauses within the rest of the coll clause, including the until/last clause, and appear as single values. The final list aggregation is only visible after the coll clause.
The behavior of coll leads to difficulties when a delimited variable are used to match material which is delimiter separated rather than terminated. For instance, entries in a comma-separated files usually do not appear as "a,b,c," but rather "a,b,c".
So for instance, the following result is not satisfactory:
The 5 is missing because it isn't followed by a space, which the text-delimited variable match "@a " looks for. After matching "4 ", coll continues to look for matches, and doesn't find any. It is tempting to try to fix it like this:
The problem now is that the regular expression / ?/ (match either a space or nothing), matches at any position. So when it is used as a variable delimiter, it matches at the current position, which binds the empty string to the variable, the extent of the match being zero. In this situation, the coll directive proceeds character by character. The solution is to use positive matching: specify the regular expression which matches the item, rather than a trying to match whatever follows. The collect directive will recognize all items which match the regular expression:
The until clause can specify a pattern which, when recognized, terminates the collection. So for instance, suppose that the list of items may or may not be terminated by a semicolon. We must exclude the semicolon from being a valid character inside an item, and add an until clause which recognizes a semicolon:
Whether followed by the semicolon or not, the items are collected properly.
Note that the @(end) is followed by a semicolon. That's because when the @(until) clause meets a match, the matching material is not consumed.
This repetition can be avoided by using @(last) instead of @(until) since @(last) consumes the terminating material.
Instead of the above regular-expression-based approach, this extraction problem can also be solved with cases:
The @(coll) directive takes the :vars keyword.
The shorthand @(rep) may be used instead of @(coll :vars nil). @(rep) takes all keywords, except :vars.
The flatten directive can be used to convert variables to one-dimensional lists. Variables which have a scalar value are converted to lists containing that value. Variables which are multidimensional lists are flattened to one-dimensional lists.
Example (without @(flatten)):
Example (with @(flatten)):
The syntax of merge follows the pattern:
@(merge destination [sources ...])
destination is a variable, which receives a new binding. sources are bind expressions.
The merge directive provides a way of combining collected data from multiple nested lists in a way which normalizes different nesting levels among the sources. This directive is useful for combining the results from collects at different levels of nesting into a single nested list such that parallel elements are at equal depth.
A new binding is created for the destination variable, which holds the result of the operation.
The merge directive performs its special function if invoked with at least three arguments: a destination and two sources.
The one-argument case @(merge x) binds a new variable x and initializes it with the empty list and is thus equivalent to @(bind x). Likewise, the two-argument case @(merge x y) is equivalent to @(bind x y), establishing a binding for x which is initialized with the value of y.
To understand what merge does when two sources are given, as in @(merge C A B), we first have to define a property called depth. The depth of an atom such as a string is defined as 1. The depth of an empty list is 0. The depth of a nonempty list is one plus the depth of its deepest element. So for instance "foo" has depth 1, ("foo") has depth 2, and ("foo" ("bar")) has depth three.
We can now define a binary (two argument) merge(A, B) function as follows. First, merge(A, B) normalizes the values A and B to produce a pair of values which have equal depth, as defined above. If either value is an atom it is first converted to a one-element list containing that atom. After this step, both values are lists; and the only way an argument has depth zero is if it is an empty list. Next, if either value has a smaller depth than the other, it is wrapped in a list as many times as needed to give it equal depth. For instance if A is (a) and B is (((("b" "c") ("d" "e)))) then A is converted to (((("a")))). Finally, the list values are appended together to produce the merged result. In the case of the preceding two example values, the result is: (((("a"))) ((("b" "c") ("d" "e)))). The result is stored into a the newly bound destination variable C.
If more than two source arguments are given, these are merged by a left-associative reduction, which is to say that a three argument merge(X, Y, Z) is defined as merge(merge(X, Y), Z). The leftmost two values are merged, and then this result is merged with the third value, and so on.
The cat directive converts a list variable into a single piece of text. The syntax is:
@(cat var [sep])
The sep argument is a Lisp expression whose value specifies a separating piece of text. If it is omitted, then a single space is used as the separator.
Example:
The syntax of the bind directive is:
@(bind pattern bind-expression {keyword value}*)
The bind directive is a kind of pattern match, which matches one or more variables given in pattern against a value produced by the bind-expression on the right.
Variable names occurring in the pattern expression may refer to bound or unbound variables.
All variable references occurring in bind-expression must have a value.
Binding occurs as follows. The tree structure of pattern and the value of bind-expression are considered to be parallel structures.
Any variables in pattern which are unbound receive a new binding, which is initialized with the structurally corresponding piece of the object produced by bind-expression.
Any variables in pattern which are already bound must match the corresponding part of the value of bind-expression, or else the bind directive fails. Variables which are already bound are not altered, retaining their current values even if the matching is inexact.
The simplest bind is of one variable against itself, for instance binding A against A:
@(bind A A)
This will throw an exception if A is not bound. If A is bound, it succeeds, since A matches itself.
The next simplest bind binds one variable to another:
@(bind A B)
Here, if A is unbound, it takes on the same value as B. If A is bound, it has to match B, or the bind fails. Matching means that either
The right-hand side does not have to be a variable. It may be some other object, like a string, quasiliteral, regexp, or list of strings, etc. For instance,
@(bind A "ab\tc")
will bind the string "ab\tc" to the variable A if A is unbound. If A is bound, this will fail unless A already contains an identical string. However, the right-hand side of a bind cannot be an unbound variable, nor a complex expression that contains unbound variables.
The left-hand side of bind can be a nested list pattern containing variables. The last item of a list at any nesting level can be preceded by a . (dot), which means that the variable matches the rest of the list from that position.
Suppose that the list A contains ("now" "now" "brown" "cow"). Then the directive @(bind (H N . C) A), assuming that H, N and C are unbound variables, will bind H to "how", code N to "now", and C to the remainder of the list ("brown" "cow").
Example: suppose that the list A is nested to two dimensions and contains (("how" "now") ("brown" "cow")). Then @(bind ((H N) (B C)) A) binds H to "how", N to "now", B to "brown" and C to "cow".
The dot notation may be used at any nesting level. it must be followed by an item. The forms (.) and (X .) are invalid, but (. X) is valid and equivalent to X.
The number of items in a left pattern match must match the number of items in the corresponding right side object. So the pattern () only matches an empty list. The notations () and nil mean exactly the same thing.
The symbols nil, t and keyword symbols may be used on either side. They represent themselves. For example @(bind :foo :bar) fails, but @(bind :foo :foo) succeeds since the two sides denote the same keyword symbol object.
In this example, suppose A contains "foo" and B contains bar. Then @(bind (X (Y Z)) (A (B "hey"))) binds X to "foo", Y to "bar" and Z to "hey". This is because the bind-expression produces the object ("foo" ("bar" "hey")) which is then structurally matched against the pattern (X (Y Z)), and the variables receive the corresponding pieces.
@(bind "a" "A" :lfilt :upcase)
produces a match, since the left side is the same as the right after filtering through the :upcase filter.
For example, the following produces a match:
@(bind "A" "a" :rfilt :upcase)
For a description of filters, see Output Filtering below.
Compound filters like (:fromhtml :upcase) are supported with all these keywords. The filters apply across arbitrary patterns and nested data.
Example:
@(bind (a b c) ("A" "B" "C"))
@(bind (a b c) (("z" "a") "b" "c") :rfilt :upcase)
Here, the first bind establishes the values for a, b and c, and the second bind succeeds, because the value of a matches the second element of the list ("z" "a") if it is upcased, and likewise b matches "b" and c matches "c" if these are upcased.
TXR Lisp forms, introduced by @ may be used in the bind-expression argument of bind, or as the entire form. This is consistent with the rules for bind expressions.
TXR Lisp forms can be used in the pattern expression also.
Example:
@(bind a @(+ 2 2))
@(bind @(+ 2 2) @(* 2 2))
Here, a is bound to the integer 4. The second bind then succeeds because the forms (+ 2 2) and (* 2 2) produce equal values.
The syntax of the set directive is:
@(set pattern bind-expression)
The set directive syntactically resembles bind, but is not a pattern match. It overwrites the previous values of variables with new values from the right-hand side. Each variable that is assigned must have an existing binding: set will not induce binding.
Examples follow.
Store the value of A back into A, an operation with no effect:
@(set A A)
Exchange the values of A and B:
@(set (A B) (B A))
Store a string into A:
@(set A "text")
Store a list into A:
@(set A ("line1" "line2"))
Destructuring assignment. A ends up with "A", B ends up with ("B1" "B2") and C binds to ("C1" "C2").
@(bind D ("A" ("B1" "B2") "C1" "C2"))
@(bind (A B C) (() () ()))
@(set (A B . C) D)
Note that set does not support a TXR Lisp expression on the left side, so the following are invalid syntax:
@(set @(+ 1 1) @(* 2 2))
@(set @b @(list "a"))
The second one is erroneous even though there is a variable on the left. Because it is preceded by the @ escape, it is a Lisp variable, and not a pattern variable.
The set directive also doesn't support Lisp expressions in the pattern, which must consist only of variables.
The syntax of the rebind directive is:
@(rebind pattern bind-expression)
The rebind directive resembles bind. It combines the semantics of local and bind into a single directive. The bind-expression is evaluated in the current environment, and its value remembered. Then a new environment is produced in which all the variables specified in pattern are absent. Then, the pattern is newly bound in that environment against the previously produced value, as if using bind.
The old environment with the previous variables is not modified; it continues to exist. This is in contrast with the set directive, which mutates existing bindings.
rebind makes it easy to create temporary bindings based on existing bindings.
@(define pattern-function (arg))
@;; inside a pattern function:
@(rebind recursion-level @(+ recursion-level 1))
@;; ...
@(end)
When the function terminates, the previous value of recursion-level is restored. The effect is less verbose and more efficient than the following equivalent
@(define pattern-function (arg))
@;; inside a pattern function:
@(local temp)
@(set temp recursion-level)
@(local recursion-level)
@(set recursion-level @(+ temp 1))
@;; ...
@(end)
Like bind, rebind supports nested patterns, such as
@(rebind (a (b c)) (1 (2 3))
but it does not support any keyword arguments. The filtering features of bind do not make sense in rebind because the variables are always reintroduced into an environment in which they don't exist, whereas filtering applies in situations when bound variables are matched against values.
The rebind directive also doesn't support Lisp expressions in the pattern, which must consist only of variables.
The forget has two spellings: @(forget) and @(local).
The arguments are one or more symbols, for example:
@(forget a)
@(local a b c)
this can be written
@(local a)
@(local a b c)
Directives which follow the forget or local directive no longer see any bindings for the symbols mentioned in that directive, and can establish new bindings.
It is not an error if the bindings do not exist.
It is strongly recommended to use the @(local) spelling in functions, because the forgetting action simulates local variables: for the given symbols, the machine forgets any earlier variables from outside of the function, and consequently, any new bindings for those variables belong to the function. (Furthermore, functions suppress the propagation of variables that are not in their parameter list, so these locals will be automatically forgotten when the function terminates.)
The syntax of @(do) is:
@(do lisp-expression*)
The do directive evaluates zero or more TXR Lisp expressions. (See TXR LISP far below.) The value of the expression is ignored, and matching continues with the directives which follow the do directive, if any.
In the context of the do directive, the expression should not be introduced by the @ symbol; it is expected to be a Lisp expression.
Example:
@; match text into variables a and b, then insert into hash table h
@(bind h @(hash))
@a:@b
@(do (set [h a] b))
The syntax of @(mdo) is:
@(mdo lisp-expression*)
Like the do directive, mdo (macro-time do) evaluates zero or more TXR Lisp expressions. Unlike do, mdo performs this evaluation immediately upon being parsed. Then it disappears from the syntax.
The effect of @(mdo e0 e1 e2 ...) is exactly like @(do (macro-time e0 e1 e2 ...)) except that do doesn't disappear from the syntax.
Another difference is that do can be used as a horizontal or vertical directive, whereas mdo is only vertical.
The in-package directive shares the same syntax and semantics as the TXR Lisp macro of the same name:
(in-package name)
The in-package directive is evaluated immediately upon being parsed, leaving no trace in the syntax tree of the surrounding TXR query.
It causes the *package* special variable to take on the package denoted by name.
The directive that name is either a string or symbol. An error exception is thrown if this isn't the case. Otherwise it searches for the package. If the package is not found, an error exception is thrown.
Blocks are useful for terminating parts of a pattern-matching search prematurely, and escaping to a higher level. This makes blocks not only useful for simplifying the semantics of certain pattern matches, but also an optimization tool.
Judicious use of blocks and escapes can reduce or eliminate the amount of backtracking that TXR performs.
The @(block name) directive introduces a named block, except when name is the symbol nil. The @(block) directive introduces an unnamed block, equivalent to @(block nil).
The @(skip) and @(collect) directives introduce implicit anonymous blocks, as do function bodies.
Blocks must be terminated by @(end) and can be vertical:
or horizontal:
The names of blocks are in a distinct namespace from the variable binding space. So @(block foo) is unrelated to the variable @foo.
A block extends from the @(block ...) directive which introduces it, until the matching @(end), and may be empty. For instance:
@(some)
abc
@(block foo)
xyz
@(end)
@(end)
Here, the block foo occurs in a @(some) clause, and so it extends to the @(end) which terminates the block. After that @(end), the name foo is not associated with a block (is not "in scope"). The second @(end) terminates the @(some) block.
The implicit anonymous block introduced by @(skip) has the same scope as the @(skip): it extends over all of the material which follows the skip, to the end of the containing subquery.
Blocks may nest, and nested blocks may have the same names as blocks in which they are nested. For instance:
@(block)
@(block)
...
@(end)
@(end)
is a nesting of two anonymous blocks, and
@(block foo)
@(block foo)
@(end)
@(end)
is a nesting of two named blocks which happen to have the same name. When a nested block has the same name as an outer block, it creates a block scope in which the outer block is "shadowed"; that is to say, directives which refer to that block name within the nested block refer to the inner block, and not to the outer one.
A block normally does nothing. The query material in the block is evaluated normally. However, a block serves as a termination point for @(fail) and @(accept) directives which are in scope of that block and refer to it.
The precise meaning of these directives is:
The @(fail) directive has a vertical and horizontal form.
If the implicit block introduced by @(skip) is terminated in this manner, this has the effect of causing skip itself to fail. In other words, the behavior is as if @(skip)'s search did not find a match for the trailing material, except that it takes place prematurely (before the end of the available data source is reached).
If the implicit block associated with a @(collect) is terminated this way, then the entire collect fails. This is a special behavior, because a collect normally does not fail, even if it matches nothing and collects nothing!
To prematurely terminate a collect by means of its anonymous block, without failing it, use @(accept).
@(accept) communicates the current bindings and input position to the terminated block. These bindings and current position may be altered by special interactions between certain directives and @(accept), described in the following section. Communicating the current bindings and input position means that the block which is terminated by @(accept) exhibits the bindings which were collected just prior to the execution of that @(accept) and the input position which was in effect at that time.
@(accept) has a vertical and horizontal form. In the horizontal form, it communicates a horizontal input position. A horizontal input position thus communicated will only take effect if the block being terminated had been suspended on the same line of input.
If the implicit block introduced by @(skip) is terminated by @(accept), this has the effect of causing the skip itself to succeed, as if all of the trailing material had successfully matched.
If the implicit block associated with a @(collect) is terminated by @(accept), then the collection stops. All bindings collected in the current iteration of the collect are discarded. Bindings collected in previous iterations are retained, and collated into lists in accordance with the semantics of collect.
Example: alternative way to achieve @(until) termination:
@(collect)
@ (maybe)
---
@ (accept)
@ (end)
@LINE
@(end)
This query will collect entire lines into a list called LINE. However, if the line --- is matched (by the embedded @(maybe)), the collection is terminated. Only the lines up to, and not including the --- line, are collected. The effect is identical to:
@(collect)
@LINE
@(until)
---
@(end)
The difference (not relevant in these examples) is that the until clause has visibility into the bindings set up by the main clause.
However, the following example has a different meaning:
@(collect)
@LINE
@ (maybe)
---
@ (accept)
@ (end)
@(end)
Now, lines are collected until the end of the data source, or until a line is found which is followed by a --- line. If such a line is found, the collection stops, and that line is not included in the collection! The @(accept) terminates the process of the collect body, and so the action of collecting the last @LINE binding into the list is not performed.
Example: communication of bindings and input position:
At the point where the accept occurs, the foo block has matched the first line, bound the text "1" to the variable @first. The block is then terminated. Not only does the @first binding emerge from this terminated block, but what also emerges is that the block advanced the data past the first line to the second line. Next, the @(some) directive ends, and propagates the bindings and position. Thus the @second which follows then matches the second line and takes the text "2".
Example: abandonment of @(some) clause by @(accept):
In the following query, the foo block occurs inside a maybe clause. Inside the foo block there is a @(some) clause. Its first subclause matches variable @first and then terminates block foo. Since block foo is outside of the @(some) directive, this has the effect of terminating the @(some) clause:
The second clause of the @(some) directive, namely:
@one
@two
@three
@four
is never processed. The reason is that subclauses are processed in top to bottom order, but the processing was aborted within the first clause the @(accept foo). The @(some) construct never gets the opportunity to match four lines.
If the @(accept foo) line is removed from the above query, the output is different:
Now, all clauses of the @(some) directive have the opportunity to match. The second clause grabs four lines, which is the longest match. And so, the next line of input available for matching is 5, which goes to the @second variable.
If one of the clauses which follow a @(trailer) requests a successful termination to an outer block via @(accept), then @(trailer) intercepts the escape and adjusts the data extent to the position that it was given.
Example:
The variable line3 is bound to "1" because although @(accept) yields a data position which has advanced to the third line, this is intercepted by @(trailer) and adjusted back to the first line. Neglecting to do this adjustment would violate the semantics of trailer.
When the clauses under a next directive are terminated by an accept, such that control passes to a block which surrounds that next, the accept is intercepted by next.
The input position being communicated by the accept is replaced with the original input position in the original stream which is in effect prior to the next directive. The accept transfer is then resumed.
In other words, accept cannot be used to "leak" the new stream out of a next scope.
However, next has no effect on the bindings being communicated.
Example:
@(next "file-x")
@(block b)
@(next "file-y")
@line
@(accept b)
@(end)
Here, the variable line matches the first line of the file "file-y", after which an accept transfer is initiated, targeting block b. This transfer communicates the line binding, as well as the position within file-y, pointing at the second line. However, the accept traverses the next directive, causing it to be abandoned. The special unwinding action within that directive detects this transfer and rewrites the input position to be the original one within the stream associated with "file-x". Note that this special handling exists in order for the behavior to be consistent with what would happen if the @(accept b) were removed, and the block b terminated normally: because the inner next is nested within that block, TXR would backtrack to the previous input position within "file-x".
Example:
@(define fun (a))
@ (bind a "a")
@ (bind b "b")
@ (accept blk)
@(end)
@(block blk)
@(fun x)
this line is skipped by accept
@(end)
Here, the accept initiates a control transfer which communicates the a and b variable bindings which are visible in that scope. This transfer is intercepted by the function, and the treatment of the bindings follows to the same rules as a normal return (which, in the given function, would readily take place if the accept directive were removed). The b variable is suppressed, because b isn't a parameter of the function. Because a is a parameter, and the argument to that parameter is the unbound variable x, the effect is that x is bound to the value of a. When the accept transfer reaches block blk and terminates it, all that emerges is the x binding carrying "a".
If the accept invocation is removed from fun, then the function returns normally, producing the x binding. In that case, the line this line is skipped by accept isn't skipped since the block isn't being terminated; that line must match something.
The processing of the finally block detects that it has been triggered by an accept transfer. Consequently, it retrieves the current input position and bindings from that transfer, and uses that position and those bindings for the processing of the finally clauses.
If the finally clauses succeed, then the new input position and new bindings are installed into the accept control transfer and that transfer resumes.
If the finally clauses fail, then the accept transfer is converted to a fail, with exactly the same block as its destination.
This creates the possibility that an accept in horizontal context targets a vertical block or vice versa, raising the question of how the input position is treated. The semantics of this is defined.
If a horizontal-context accept targets a vertical block, the current position at the target block will be the following line. That is to say, when the horizontal accept occurs, there is a current input line which may have unconsumed material past the current position. If the accept communicates its input position to a vertical context, that unconsumed material is skipped, as if it had been matched and the vertical position is advanced to the next line.
If a horizontal block catches a vertical accept, it rejects that accept's position and stays at the current backtracking position for that block. Only the bindings from the accept are retained.
Functions in TXR are not exactly like functions in mathematics or functional languages, and are not like procedures in imperative programming languages. They are not exactly like macros either. What it means for a TXR function to take arguments and produce a result is different from the conventional notion of a function.
A TXR function may have one or more parameters. When such a function is invoked, an argument must be specified for each parameter. However, a special behavior is at play here. Namely, some or all of the argument expressions may be unbound variables. In that case, the corresponding parameters behave like unbound variables also. Thus TXR function calls can transmit the "unbound" state from argument to parameter.
It should be mentioned that functions have access to all bindings that are visible in the caller; functions may refer to variables which are not mentioned in their parameter list.
With regard to returning, TXR functions are also unconventional. If the function fails, then the function call is considered to have failed. The function call behaves like a kind of match; if the function fails, then the call is like a failed match.
When a function call succeeds, then the bindings emanating from that function are processed specially. Firstly, any bindings for variables which do not correspond to one of the function's parameters are thrown away. Functions may internally bind arbitrary variables in order to get their job done, but only those variables which are named in the function argument list may propagate out of the function call. Thus, a function with no arguments can only indicate matching success or failure, but not produce any bindings. Secondly, variables do not propagate out of the function directly, but undergo a renaming. For each parameter which went into the function as an unbound variable (because its corresponding argument was an unbound variable), if that parameter now has a value, that value is bound onto the corresponding argument.
Example:
@(define collect-words (list))
@(coll)@{list /[^ \t]+/}@(end)
@(end)
The above function collect-words contains a query which collects words from a line (sequences of characters other than space or tab), into the list variable called list. This variable is named in the parameter list of the function, therefore, its value, if it has one, is permitted to escape from the function call.
Suppose the input data is:
Fine summer day
and the function is called like this:
@(collect-words wordlist)
The result (with txr -B) is:
wordlist[0]=Fine
wordlist[1]=summer
wordlist[1]=day
How it works is that in the function call @(collect-words wordlist), wordlist is an unbound variable. The parameter corresponding to that unbound variable is the parameter list. Therefore, that parameter is unbound over the body of the function. The function body collects the words of "Fine summer day" into the variable list, and then yields the that binding. Then the function call completes by noticing that the function parameter list now has a binding, and that the corresponding argument wordlist has no binding. The binding is thus transferred to the wordlist variable. After that, the bindings produced by the function are thrown away. The only enduring effects are:
Another way to understand the parameter behavior is that function parameters behave like proxies which represent their arguments. If an argument is an established value, such as a character string or bound variable, the parameter is a proxy for that value and behaves just like that value. If an argument is an unbound variable, the function parameter acts as a proxy representing that unbound variable. The effect of binding the proxy is that the variable becomes bound, an effect which is settled when the function goes out of scope.
Within the function, both the original variable and the proxy are visible simultaneously, and are independent. What if a function binds both of them? Suppose a function has a parameter called P, which is called with an argument A, which is an unbound variable, and then, in the function, both A and P bound. This is permitted, and they can even be bound to different values. However, when the function terminates, the local binding of A simply disappears (because the symbol A is not among the parameters of the function). Only the value bound to P emerges, and is bound to A, which still appears unbound at that point. The P binding disappears also, and the net effect is that A is now bound. The "proxy" binding of A through the parameter P "wins" the conflict with the direct binding.
Function definition syntax comes in two flavors: vertical and horizontal. Horizontal definitions actually come in two forms, the distinction between which is hardly noticeable, and the need for which is made clear below.
A function definition begins with a @(define ...) directive. For vertical functions, this is the only element in a line.
The define symbol must be followed by a symbol, which is the name of the function being defined. After the symbol, there is a parenthesized optional argument list. If there is no such list, or if the list is specified as () or the symbol nil then the function has no parameters. Examples of valid define syntax are:
@(define foo)
@(define bar ())
@(define match (a b c))
If the define directive is followed by more material on the same line, then it defines a horizontal function:
@(define match-x)x@(end)
If the define is the sole element in a line, then it is a vertical function, and the function definition continues below:
@(define match-x)
x
@(end)
The difference between the two is that a horizontal function matches characters within a line, whereas a vertical function matches lines within a stream. The former match-x matches the character x, advancing to the next character position. The latter match-x matches a line consisting of the character x, advancing to the next line.
Material between @(define) and @(end) is the function body. The define directive may be followed directly by the @(end) directive, in which case the function has an empty body.
Functions may be nested within function bodies. Such local functions have dynamic scope. They are visible in the function body in which they are defined, and in any functions invoked from that body.
The body of a function is an anonymous block. (See Blocks above.)
If a horizontal function is defined as the only element of a line, it may not be followed by additional material. The following construct is erroneous:
@(define horiz (x))@foo:@bar@(end)lalala
This kind of definition is actually considered to be in the vertical context, and like other directives that have special effects and that do not match anything, it does not consume a line of input. If the above syntax were allowed, it would mean that the line would not only define a function but also match lalala. This would, in turn, would mean that the @(define)...@(end) is actually in horizontal mode, and so it matches a span of zero characters within a line (which means that is would require a line of input to match: a surprising behavior for a nonmatching directive!)
A horizontal function can be defined in an actual horizontal context. This occurs if its is in a line where it is preceded by other material. For instance:
X@(define fun)...@(end)Y
This is a query line which must match the text XY. It also defines the function fun. The main use of this form is for nested horizontal functions:
@(define fun)@(define local_fun)...@(end)@(end)
A function of the same name may be defined as both vertical and horizontal. Both functions are available at the same time. Which one is used by a call is resolved by context. See the section Vertical Versus Horizontal Calls below.
A function is invoked by compound directive whose first symbol is the name of that function. Additional elements in the directive are the arguments. Arguments may be symbols, or other objects like string and character literals, quasiliterals ore regular expressions.
Example:
The first call to the function takes the line "one two". The parameter a takes "one" and parameter b takes "two". These are rebound to the arguments first and second. The second call to the function binds the a parameter to the word "ice", and the b is unbound, because the corresponding argument cream is unbound. Thus inside the function, a is forced to match ice. Then a space is matched and b collects the text "milk". When the function returns, the unbound "cream" variable gets this value.
If a symbol occurs multiple times in the argument list, it constrains both parameters to bind to the same value. That is to say, all parameters which, in the body of the function, bind a value, and which are all derived from the same argument symbol must bind to the same value. This is settled when the function terminates, not while it is matching. Example:
Here the query fails because a and b are effectively proxies for the same unbound variable same and are bound to different values, creating a conflict which constitutes a match failure.
A function call which is the only element of the query line in which it occurs is ambiguous. It can go either to a vertical function or to the horizontal one. If both are defined, then it goes to the vertical one.
Example:
Not only does this call go to the vertical function, but it is in a vertical context.
If only a horizontal function is defined, then that is the one which is called, even if the call is the only element in the line. This takes place in a horizontal character-matching context, which requires a line of input which can be traversed:
Example:
The query fails because since @(which fun) is in horizontal mode, it matches characters in a line. Since the function body consists only of @(bind ...) which doesn't match any characters, the function call requires an empty line to match. The line ABC is not empty, and so there is a matching failure. The following example corrects this:
Example:
A call made in a clearly horizontal context will prefer the horizontal function, and only fall back on the vertical one if the horizontal one doesn't exist. (In this fallback case, the vertical function is called with empty data; it is useful for calling vertical functions which process arguments and produce values.)
In the next example, the call is followed by trailing material, placing it in a horizontal context. Leading material will do the same thing:
Example:
As described earlier, variables bound in a function body which are not parameters of the function are discarded when the function returns. However, that, by itself, doesn't make these variables local, because pattern functions have visibility to all variables in their calling environment. If a variable x exists already when a function is called, then an attempt to bind it inside a function may result in a failure. The local directive must be used in a pattern function to list which variables are local.
Example:
@(define path (path))@\
@(local x y)@\
@(cases)@\
(@(path x))@(path y)@(bind path `(@x)@y`)@\
@(or)@\
@{x /[.,;'!?][^ \t\f\v]/}@(path y)@(bind path `@x@y`)@\
@(or)@\
@{x /[^ .,;'!?()\t\f\v]/}@(path y)@(bind path `@x@y`)@\
@(or)@\
@(bind path "")@\
@(end)@\
@(end)
This is a horizontal function which matches a path, which lands into four recursive cases. A path can be parenthesized path followed by a path; it can be a certain character followed by a path, or it can be empty
This function ensures that the variables it uses internally, x and y, do not have anything to do with any inherited bindings for x and y.
Note that the function is recursive, which cannot work without x and y being local, even if no such bindings exist prior to the top-level invocation of the function. The invocation @(path x) causes x to be bound, which is visible inside the invocation @(path y), but that invocation needs to have its own binding of x for local use.
Function definitions may appear in a function. Such definitions are visible in all functions which are invoked from the body (and not necessarily enclosed in the body). In other words, the scope is dynamic, not lexical. Inner definitions shadow outer definitions. This means that a caller can redirect the function calls that take place in a callee, by defining local functions which capture the references.
Example:
Here, the function which is defined which calls fun. A top-level definition of fun is introduced which outputs "top-level fun!". The function callee provides its own local definition of fun which outputs "local fun!" before calling which. When callee is invoked, it calls which, whose @(fun) call is routed to callee's local definition. When which is called directly from the top level, its fun call goes to the top-level definition.
Function indirection may be performed using the
call
directive. If
fun-expr
is an Lisp expression which evaluates to a symbol, and
that symbol names a function which takes no arguments, then
@(call fun-expr)
may be used to invoke the function. Additional
expressions may be supplied which specify arguments.
Example 1:
@(define foo (arg))
@(bind arg "abc")
@(end)
@(call 'foo b)
In this example, the effect is that foo is invoked, and b ends up bound to "abc".
The call directive here uses the 'foo expression to calculate the name of the function to be invoked. (See the quote operator).
This particular call expression can just be replaced by the direct invocation syntax @(foo b).
The power of call lies in being able to specify the function as a value which comes from elsewhere in the program, as in the following example.
@(define foo (arg))
@(bind arg "abc")
@(end)
@(bind f @'foo)
@(call f b)
Here the call directive obtains the name of the function from the f variable.
Note that function names are resolved to functions in the environment that is apparent at the point in execution where the call takes place. The directive @(call f args ...) is precisely equivalent to @(s args ...) if, at the point of the call, f is a variable which holds the symbol s and symbol s is defined as a function. Otherwise it is erroneous.
The syntax of the load and include directives is:
Where expr is a Lisp expression that evaluates to a string giving the path of the file to load.
Firstly, the path given by expr is converted to an effective path, as follows.
If the value of the *load-path* variable has a current value which is not nil and the path given in expr is pure relative according to the pure-rel-path-p function, then the effective path is interpreted taken relative to the directory portion of the path which is stored in *load-path*.
If *load-path* is nil, or the load path is not pure relative, then the path is taken as-is as the effective path.
Next, an attempt is made to open the file for processing, in almost exactly the same manner as by the TXR Lisp function load. The difference is that if the effective path is unsuffixed, then the .txr suffix is added to it, and that resulting path is tried first, and if it succeeds, then the file is treated as TXR Pattern Language syntax. If that fails, then the suffix .tlo is tried, and so forth, as described for the load function.
If these initial attempts to find the file fail, and the failure is due to the file not being found rather than some other problem such as a permission error, and expr isn't an absolute path according to abs-path-p, then additional attempts are made by searching for the file in the list of directories given in the *load-search-dirs* variable. Details are given in the description of the TXR Lisp load function.
Both the load and include directives bind the *load-path* variable to the path of the loaded file just before parsing syntax from it, The *package* variable is also given a new dynamic binding, whose value is the same as the existing binding. These bindings are removed when the load operation completes, restoring the prior values of these variables. The *load-hooks* variable is given a new dynamic binding, with a nil value.
If the file opened for processing is TXR Lisp source, or a compiled TXR Lisp file, then it is processed in the manner described for the load function.
Different requirements apply to the processing of the file under the load and include directives.
The include directive performs the processing of the file at parse time. If the file being processed is TXR Pattern Language, then it is parsed, and then its syntax replaces the include directive, as if it had originally appeared in its place. If a TXR Lisp source or a compiled TXR Lisp file is processed by include then the include directive is removed from the syntax.
The load directive performs the processing of the file at evaluation time. Evaluation time occurs after a TXR program is read from beginning to end and parsed. That is to say, when a TXR query is parsed, any embedded @(load ...) forms in it are parsed and constitute part of its syntax tree. They are executed when that query is executed, whenever its execution reaches those load directives. When the load directive processes TXR Pattern Language syntax, it parses the file in its entirety and then executes that file's directives against the current input position. Repeated executions of the same load directive result in repeated processing of the file.
Note: the include directive is useful for loading TXR files which contain Lisp macros which are needed by the parent program. The parent program cannot use load to bring in macros because macros are required during expansion, which takes place prior to evaluation time, whereas load doesn't execute until evaluation time.
Note: the load directive doesn't provide access to the value propagated by a return via the load block.
See also: the load function, and the self-path, stdlib and *load-path* variables in TXR Lisp.
A TXR query may perform custom output. Output is performed by output clauses, which may be embedded anywhere in the query, or placed at the end. Output occurs as a side effect of producing a part of a query which contains an @(output) directive, and is executed even if that part of the query ultimately fails to find a match. Thus output can be useful for debugging. An output clause specifies that its output goes to a file, pipe, or (by default) standard output. If any output clause is executed whose destination is standard output, TXR makes a note of this, and later, just prior to termination, suppresses the usual printing of the variable bindings or the word false.
The syntax of the @(output) directive is:
@(output [ destination ] { bool-keyword | keyword value }* )
.
. one or more output directives or lines
.
@(end)
If the directive has arguments, then the first one is evaluated. If it is an object other than a keyword symbol, then it specifies the optional destination. Any remaining arguments after the optional destination are the keyword list. If the destination is missing, then the entire argument list is a keyword list.
The destination argument, if present, is treated as a TXR Lisp expression and evaluated. The resulting value is taken as the output destination. The value may be a string which gives the pathname of a file to open for output. Otherwise, the destination must be a stream object.
The keyword list consists of a mixture of Boolean keywords which do not have an argument, or keywords with arguments.
The following Boolean keywords are supported:
Note that since command pipes are processes that report errors asynchronously, a failing command will not throw an immediate exception that can be suppressed with :nothrow. This is for synchronous errors, like trying to open a destination file, but not having permissions, etc.
The following value keywords are supported:
See the later sections Output Filtering below, and The Deffilter Directive.
The @(push) directive is a variant of @(output) which produces lines of text that are pushed back into the input stream.
This directive supports only the :filter keyword argument.
This directive doesn't take any of the keyword arguments supported by @(output) except for the :filter keyword.
After the execution of a @(push), the next pattern matching syntax that is evaluated now faces the material produced by that @(push) followed by the original input. In order to preserve the line numbering of the original input, @(push) adjusts the line number for the synthetic input by subtracting the number of synthetic lines from the original input's line number. For instance if the original input is line 5, and 7 lines are prepended by @(push), then those lines are numbered -2 to 4.
The input-synthesizing effect of @(push) is visible to a subsequent form in exactly those situations in which an input-consuming effect of a pattern matching directive would also be visible. For instance, a @(push) occurring in the body of a @(collect) can produce input that is visible to the next iteration.
The @(push) directive interacts with the parallel matching directives such as @(some). When multiple parallel clauses match, the input position is advanced by the longest match. Lines pushed into the input by @(push) look like negative advancement. If one clause advances in the input, while another one pushes into it, the push will lose to the advancement and its effect will disappear. If two clauses push varying amounts of material, the shorter push will win.
Swap the first two lines if they start with a colon, changing the colon to a period:
@(maybe) :@a :@b @ (push) .@b .@a @ (end) @(end) @(data capture) @(do (tprint capture))
:hello :there rest of data
.there .hello rest of data
Text in an output clause is not matched against anything, but is output verbatim to the destination file, device or command pipe.
Variables occurring in an output clause do not match anything; instead their contents are output.
A variable being output can be any object. If it is of a type other than a list or string, it will be converted to a string as if by the tostring function in TXR Lisp.
A value which is a sequence is converted to a string in a special way: the elements are individually converted to strings and then they are catenated together. The default separator string for most sequences is a single space: an alternate separation can be specified as an argument in the brace substitution syntax. Empty sequences turn into an empty string. More details are given in the Output Variables: Separation section below.
Lists may be output within @(repeat) or @(rep) clauses. Each nesting of these constructs removes one level of nesting from the list variables that it contains.
In an output clause, the @{name number} variable syntax generates fixed-width field, which contains the variable's text. The absolute value of the number specifies the field width. For instance -20 and 20 both specify a field width of twenty. If the text is longer than the field, then it overflows the field. If the text is shorter than the field, then it is left-adjusted within that field, if the width is specified as a positive number, and right-adjusted if the width is specified as negative.
An output variable may specify a filter which overrides any filter established for the output clause. The syntax for this is @{NAME :filter filterspec}. The filter specification syntax is the same as in the output clause. See Output Filtering below.
When the value of an output variable is a buffer (object of type buf), it is rendered as a sequence of hexadecimal digit pairs, with no line breaks. The digits a through f are rendered in lower case.
As mentioned in the previous section, the value of a variable can be a sequence. The individual elements of a sequence are turned into strings, and then catenated together with the separator, which may be specified as a string modifier in the variable syntax.
For most sequences, the default separator is a space.
When the value of a variable is a character string, and the separator is not specified, the string is output as-is. Effectively, the string is treated as a sequence but with an empty default separator.
When the value of a variable is a buffer, it is rendered in hexadecimal, as described in the previous section. If a separator string modifier is specified, it separates pairs of digits, rather than individual digits.
Example:
@(bind str "string")
@(bind buf #b'cafef00d')
@(output)
@{str[0..3] "--"}
@{buf[0..2] ":"}
@{buf[2..4] "/"}
@(end)
The above example produces the output
s--t--r
ca:fe
f0/0d
Additional syntax is supported in output variables that does not appear in pattern-matching variables.
A square bracket index notation may be used to extract elements or ranges from a variable, which works with strings, vectors and lists. Elements are indexed from zero. This notation is only available in brace-enclosed syntax, and looks like this:
If the variable is a list, it is treated as a list substitution, exactly as if it were the value of an unsubscripted list variable. The elements of the list are converted to strings and catenated together with a separator string between them, the default one being a single space.
An alternate character may be given as a string argument in the brace notation.
Example:
@(bind a ("a" "b" "c" "d"))
@(output)
@{a[1..3] "," 10}
@(end)
The above produces the text "b,c" in a field 10 spaces wide. The [1..3] argument extracts a range of a; the "," argument specifies an alternate separator string, and 10 specifies the field width.
When a variable includes indexing, separation and a field width, the indexing operation is first applied to select a subsequence. Then separation is applied to produce a textual representation. Finally the representation is rendered din the specified field width.
The brace syntax has another syntactic and semantic extension in output clauses. In place of the symbol, an expression may appear. The value of that expression is substituted.
Example:
@(bind a "foo")
@(output)
@{`@a:` -10}
Here, the quasiliteral expression `@a:` is evaluated, producing the string "foo:". This string is printed right-adjusted in a 10 character field.
The repeat directive generates repeated text from a "boilerplate", by taking successive elements from lists. The syntax of repeat is like this:
@(repeat)
.
.
main clause material, required
.
.
special clauses, optional
.
.
@(end)
repeat has four types of special clauses, any of which may be specified with empty contents, or omitted entirely. They are described below.
repeat takes arguments, also described below.
All of the material in the main clause and optional clauses is examined for the presence of variables. If none of the variables hold lists which contain at least one item, then no output is performed, (unless the repeat specifies an @(empty) clause, see below). Otherwise, among those variables which contain nonempty lists, repeat finds the length of the longest list. This length of this list determines the number of repetitions, R.
If the repeat contains only a main clause, then the lines of this clause is output R times. Over the first repetition, all of the variables which, outside of the repeat, contain lists are locally rebound to just their first item. Over the second repetition, all of the list variables are bound to their second item, and so forth. Any variables which hold shorter lists than the longest list eventually end up with empty values over some repetitions.
Example: if the list A holds "1", "2" and "3"; the list B holds "A", "B"; and the variable C holds "X", then
@(repeat)
>> @C
>> @A @B
@(end)
will produce three repetitions (since there are two lists, the longest of which has three items). The output is:
>> X
>> 1 A
>> X
>> 2 B
>> X
>> 3
The last line has a trailing space, since it is produced by "@A @B", where B has an empty value. Since C is not a list variable, it produces the same value in each repetition.
The special clauses are:
The precedence among the clauses which take an iteration is: single > first > modlast > last > mod > main. That is, whenever two or more of these clauses can apply to a repetition, then the leftmost one in this precedence list will be selected. It is possible for all these clauses to be viable for processing the same repetition. If a repeat occurs which has only one repetition, then that repetition is simultaneously the first, only and last repetition. Moreover, it also matches (mod 0 m) and, because it is the last repetition, it matches (modlast 0 m). In this situation, if there is a @(single) clause present, then the repetition shall be processed using that clause. Otherwise, if there is a @(first) clause present, that clause is activated. Failing that, @(modlast) is used if there is such a clause, featuring an n argument of zero. If there isn't, then the @(last) clause is considered, if present. Otherwise, the @(mod) clause is considered if present with an n argument of zero. Otherwise, none of these clauses are present or applicable, and the repetition is processed using the main clause.
The @(empty) clause does not appear in the above precedence list because it is mutually exclusive with respect to the others: it is processed only when there are no iterations, in which case even the main clause isn't active.
The @(repeat) clause supports arguments.
@(repeat
[:counter {symbol | (symbol expr)}]
[:vars ({symbol | (symbol expr)}*)])
The :counter argument designates a symbol which will behave as an integer variable over the scope of the clauses inside the repeat. The variable provides access to the repetition count, starting at zero, incrementing with each repetition. If the argument is given as (symbol expr) then expr is a Lisp expression whose value is taken as a displacement value which is added to each iteration of the counter. For instance :counter (c 1) specifies a counter c which counts from 1.
The :vars argument specifies a list of variable name symbols symbol or else pairs of the form (symbol init-form) consisting of a variable name and Lisp expression. Historically, the former syntax informed repeat about references to variables contained in Lisp code. This usage is no longer necessary as of TXR 243, since the repeat construct walks Lisp code, identifying all free variables. The latter syntax introduces a new pattern variable binding for symbol over the scope of the repeat construct. The init-form specifies a Lisp expression which is evaluated to produce the binding's value.
The repeat directive then processes the list of variables, selecting from it those which have a binding, either a previously existing binding or the one just introduced. For each selected variable, repeat will assume that the variable occurs in the repeat block and contains a list to be iterated.
The variable binding syntax supported by :vars of the form (symbol init-form) provides a solution for situations when it is necessary to iterate over some list, but that list is the result of an expression, and not stored in any variable. A repeat block iterates only over lists emanating from variables; it does not iterate over lists pulled from arbitrary expressions.
Example: output all file names matching the *.txr pattern in the current directory:
@(output)
@(repeat :vars ((name (glob "*.txr"))))
@name
@(end)
@(end)
Prior to TXR 243, the simple variable-binding syntax supported by :vars of the form symbol was needed for situations in which TXR Lisp expressions which referenced variables were embedded in @(repeat) blocks. Variable references embedded in Lisp code were not identified in @(repeat). For instance, the following produced no output, because no variables were found in the repeat body:
@(bind trigraph ("abc" "def" "ghi"))
@(output)
@(repeat)
@(reverse trigraph)
@(end)
@(end)
There is a reference to trigraph but it's inside the (reverse trigraph) Lisp expression that was not processed by repeat. The solution was to mention trigraph in the :vars construct:
@(bind trigraph ("abc" "def" "ghi"))
@(output)
@(repeat :vars (trigraph))
@(reverse trigraph)
@(end)
@(end)
Then the repeat block would iterate over trigraph, producing the output
cba
fed
igh
This workaround is no longer required as of TXR 243; the output is produced by the first example, without :vars.
If a repeat clause encloses variables which hold multidimensional lists, those lists require additional nesting levels of repeat (or rep). It is an error to attempt to output a list variable which has not been decimated into primary elements via a repeat construct.
Suppose that a variable X is two-dimensional (contains a list of lists). X must be nested twice in a repeat. The outer repeat will traverse the lists contained in X. The inner repeat will traverse the elements of each of these lists.
A nested repeat may be embedded in any of the clauses of a repeat, not only in the main clause.
The rep directive is similar to repeat. Whereas repeat is line-oriented, rep generates material within a line. It has all the same clauses, but everything is specified within one line:
@(rep)... main material ... .... special clauses ...@(end)
More than one @(rep) can occur within a line, mixed with other material. A @(rep) can be nested within a @(repeat) or within another @(rep).
Also, @(rep) accepts the same :counter and :vars arguments.
Example 1: show the list L in parentheses, with spaces between the elements, or the word EMPTY if the list is empty:
@(output)
@(rep)@L @(single)(@L)@(first)(@L @(last)@L)@(empty)EMPTY@(end)
@(end)
Here, the @(empty) clause specifies EMPTY. So if there are no repetitions, the text EMPTY is produced. If there is a single item in the list L, then @(single)(@L) produces that item between parentheses. Otherwise if there are two or more items, the first item is produced with a leading parenthesis followed by a space by @(first)(@L and the last item is produced with a closing parenthesis: @(last)@L). All items in between are emitted with a trailing space by the main clause: @(rep)@L.
Example 2: show the list L like Example 1 above, but the empty list is ().
@(output)
(@(rep)@L @(last)@L@(end))
@(end)
This is simpler. The parentheses are part of the text which surrounds the @(rep) construct, produced unconditionally. If the list L is empty, then @(rep) produces no output, resulting in (). If the list L has one or more items, then they are produced with spaces each one, except the last which has no space. If the list has exactly one item, then the @(last) applies to it instead of the main clause: it is produced with no trailing space.
The syntax of the close directive is:
@(close expr)
Where expr evaluates to a stream. The close directive can be used to explicitly close streams created using @(output ... :named var) syntax, as an alternative to @(output :finish expr).
Examples:
Write two lines to "foo.txt" over two output blocks using a single stream:
@(output "foo.txt" :named foo)
Hello,
@(end)
@(output :continue foo)
world!
@(end)
@(close foo)
The same as above, using :finish rather than :continue so that the stream is closed at the end of the second block:
@(output "foo.txt" :named foo)
Hello,
@(end)
@(output :finish foo)
world!
@(end)
Often it is necessary to transform the output to preserve its meaning under the convention of a given data format. For instance, if a piece of text contains the characters < or >, then if that text is being substituted into HTML, these should be replaced by < and >. This is what filtering is for. Filtering is applied to the contents of output variables, not to any template text. TXR implements named filters. Built-in filters are named by keywords, given below. User-defined filters are possible, however. See notes on the deffilter directive below.
Instead of a filter name, the syntax (fun name) can be used. This denotes that the function called name is to be used as a filter. This is described in the next section Function Filters below.
Built-in filters named by keywords:
Examples:
To escape HTML characters in all variable substitutions occurring in an output clause, specify :filter :tohtml in the directive:
@(output :filter :tohtml)
...
@(end)
To filter an individual variable, add the syntax to the variable spec:
@(output)
@{x :filter :tohtml}
@(end)
Multiple filters can be applied at the same time. For instance:
@(output)
@{x :filter (:upcase :tohtml)}
@(end)
This will fold the contents of x to uppercase, and then encode any special characters into HTML. Beware of combinations that do not make sense. For instance, suppose the original text is HTML, containing codes like ". The compound filter (:upcase :fromhtml) will not work because " will turn to " which no longer be recognized by the :fromhtml filter, since the entity names in HTML codes are case-sensitive.
Capture some numeric variables and convert to numbers:
@date @time @temperature @pressure
@(filter :tofloat temperature pressure)
@;; temperature and pressure can now be used in calculations
A function can be used as a filter. For this to be possible, the function must conform to certain rules:
For instance, the following is a valid filter function:
@(define foo_to_bar (in out))
@ (next :string in)
@ (cases)
foo
@ (bind out "bar")
@ (or)
@ (bind out in)
@ (end)
@(end)
This function binds the out parameter to "bar" if the in parameter is "foo", otherwise it binds the out parameter to a copy of the in parameter. This is a simple filter.
To use the filter, use the syntax (:fun foo_to_bar) in place of a filter name. For instance in the bind directive:
@(bind "foo" "bar" :lfilt (:fun foo_to_bar))
The above should succeed since the left side is filtered from "foo" to "bar", so that there is a match.
Function filters can be used in a chain:
@(output :filter (:downcase (:fun foo_to_bar) :upcase))
...
@(end)
Here is a split function which takes an extra argument which specifies the separator:
@(define split (in out sep))
@ (next :list in)
@ (coll)@(maybe)@token@sep@(or)@token@(end)@(end)
@ (bind out token)
@(end)
Furthermore, note that it produces a list rather than a string. This function separates the argument in into tokens according to the separator text carried in the variable sep.
Here is another function, join, which catenates a list:
@(define join (in out sep))
@ (output :into out)
@ (rep)@in@sep@(last)@in@(end)
@ (end)
@(end)
Now here is these two being used in a chain:
@(bind text "how,are,you")
@(output :filter (:fun split ",") (:fun join "-"))
@text
@(end)
Output:
how-are-you
When the filter invokes a function, it generates the first two arguments internally to pass in the input value and capture the output. The remaining arguments from the (:fun ...) construct are also passed to the function. Thus the string objects "," and "-" are passed as the sep argument to split and join.
Note that split puts out a list, which join accepts. So the overall filter chain operates on a string: a string goes into split, and a string comes out of join.
The deffilter directive allows a query to define a custom filter, which can then be used in output clauses to transform substituted data.
The syntax of deffilter is illustrated in this example:
The deffilter symbol must be followed by the name of the filter to be defined, followed by bind expressions which evaluate to lists of strings. Each list must be at least two elements long and specifies one or more texts which are mapped to a replacement text. For instance, the following specifies a telephone keypad mapping from uppercase letters to digits.
@(deffilter alpha_to_phone ("E" "0")
("J" "N" "Q" "1")
("R" "W" "X" "2")
("D" "S" "Y" "3")
("F" "T" "4")
("A" "M" "5")
("C" "I" "V" "6")
("B" "K" "U" "7")
("L" "O" "P" "8")
("G" "H" "Z" "9"))
@(deffilter foo (`@a` `@b`) ("c" `->@d`))
@(bind x ("from" "to"))
@(bind y ("---" "+++"))
@(deffilter sub x y)
The last deffilter has the same effect as the @(deffilter sub ("from" "to") ("---" "+++")) directive.
Filtering works using a longest match algorithm. The input is scanned from left to right, and the longest piece of text is identified at every character position which matches a string on the left-hand side, and that text is replaced with its associated replacement text. The scanning then continues at the first character after the matched text.
If none of the strings matches at a given character position, then that character is passed through the filter untranslated, and the scan continues at the next character in the input.
Filtering is not in-place but rather instantiates a new text, and so replacement text is not re-scanned for more replacements.
If a filter definition accidentally contains two or more repetitions of the same left-hand string with different right-hand translations, the later ones take precedence. No warning is issued.
The syntax of the filter directive is:
@(filter FILTER { VAR }+ )
A filter is specified, followed by one or more variables whose values are filtered and stored back into each variable.
Example: convert a, b, and c to uppercase and HTML encode:
@(filter (:upcase :tohtml) a b c)
The exceptions mechanism in TXR is another disciplined form of nonlocal transfer, in addition to the blocks mechanism (see Blocks above). Like blocks, exceptions provide a construct which serves as the target for a dynamic exit. Both blocks and exceptions can be used to bail out of deep nesting when some condition occurs. However, exceptions provide more complexity. Exceptions are useful for error handling, and TXR in fact maps certain error situations to exception control transfers. However, exceptions are not inherently an error-handling mechanism; they are a structured dynamic control transfer mechanism, one of whose applications is error handling.
An exception control transfer (simply called an exception) is always identified by a symbol, which is its type. Types are organized in a subtype-supertype hierarchy. For instance, the file-error exception type is a subtype of the error type. This means that a file error is a kind of error. An exception handling block which catches exceptions of type error will catch exceptions of type file-error, but a block which catches file-error will not catch all exceptions of type error. A query-error is a kind of error, but not a kind of file-error. The symbol t is the supertype of every type: every exception type is considered to be a kind of t. (Mnemonic: t stands for type, as in any type).
Exceptions are handled using @(catch) clauses within a @(try) directive.
In addition to being useful for exception handling, the @(try) directive also provides unwind protection by means of a @(finally) clause, which specifies query material to be executed unconditionally when the try clause terminates, no matter how it terminates.
The general syntax of the try directive is
@(try)
... main clause, required ...
... optional catch clauses ...
... optional finally clause
@(end)
A catch clause looks like:
@(catch TYPE [ PARAMETERS ])
.
.
.
and also this simple form:
@(catch)
.
.
.
which catches all exceptions, and is equivalent to @(catch t).
A finally clause looks like:
@(finally)
...
.
.
The main clause may not be empty, but the catch and finally may be.
A try clause is surrounded by an implicit anonymous block (see Blocks section above). So for instance, the following is a no-op (an operation with no effect, other than successful execution):
The @(accept) causes a successful termination of the implicit anonymous block. Execution resumes with query lines or directives which follow, if any.
try clauses and blocks interact. For instance, an accept from within a try clause invokes a finally.
How this works: the try block's main clause is @(accept foo). This causes the enclosing block named foo to terminate, as a successful match. Since the try is nested within this block, it too must terminate in order for the block to terminate. But the try has a finally clause, which executes unconditionally, no matter how the try block terminates. The finally clause performs some output, which is seen.
Note that finally interacts with accept in subtle ways not revealed in this example; they are documented in the description of accept under the block directive documentation.
A try directive can terminate in one of three ways. The main clause may match successfully, and possibly yield some new variable bindings. The main clause may fail to match. Or the main clause may be terminated by a nonlocal control transfer, like an exception being thrown or a block return (like the block foo example in the previous section).
No matter how the try clause terminates, the finally clause is processed.
The finally clause is itself a query which binds variables, which leads to questions: what happens to such variables? What if the finally block fails as a query? As well as: what if a finally clause itself initiates a control transfer? Answers follow.
Firstly, a finally clause will contribute variable bindings only if the main clause terminates normally (either as a successful or failed match). If the main clause of the try block successfully matches, then the finally block continues matching at the next position in the data, and contributes bindings. If the main clause fails, then the finally block tries to match at the same position where the main clause failed.
The overall try directive succeeds as a match if either the main clause or the finally clause succeed. If both fail, then the try directive is a failed match.
Example:
In this example, the main clause of the try captures line "1" of the data as variable a, then the finally clause captures "2" as b, and then the query continues with the @c line after try block, so that c captures "3".
Example:
In this example, the main clause of the try fails to match, because the input is not prefixed with "hello ". However, the finally clause matches, binding b to "1". This means that the try block is a successful match, and so processing continues with @c which captures "2".
When finally clauses are processed during a nonlocal return, they have no externally visible effect if they do not bind variables. However, their execution makes itself known if they perform side effects, such as output.
A finally clause guards only the main clause and the catch clauses. It does not guard itself. Once the finally clause is executing, the try block is no longer guarded. This means if a nonlocal transfer, such as a block accept or exception, is initiated within the finally clause, it will not re-execute the finally clause. The finally clause is simply abandoned.
The disestablishment of blocks and try clauses is properly interleaved with the execution of finally clauses. This means that all surrounding exit points are visible in a finally clause, even if the finally clause is being invoked as part of a transfer to a distant exit point. The finally clause can make a control transfer to an exit point which is more near than the original one, thereby "hijacking" the control transfer. Also, the anonymous block established by the try directive is visible in the finally clause.
Example:
@(try)
@ (try)
@ (next "nonexistent-file")
@ (finally)
@ (accept)
@ (end)
@(catch file-error)
@ (output)
file error caught
@ (end)
@(end)
In this example, the @(next) directive throws an exception of type file-error, because the given file does not exist. The exit point for this exception is the @(catch file-error) clause in the outermost try block. The inner block is not eligible because it contains no catch clauses at all. However, the inner try block has a finally clause, and so during the processing of this exception which is headed for @(catch file-error), the finally clause performs an anonymous accept. The exit point for that accept is the anonymous block surrounding the inner try. So the original transfer to the catch clause is thereby abandoned. The inner try terminates successfully due to the accept, and since it constitutes the main clause of the outer try, that also terminates successfully. The "file error caught" message is never printed.
catch clauses establish their associated try blocks as potential exit points for exception-induced control transfers (called "throws").
A catch clause specifies an optional list of symbols which represent the exception types which it catches. The catch clause will catch exceptions which are a subtype of any one of those exception types.
If a try block has more than one catch clause which can match a given exception, the first one will be invoked.
When a catch is invoked, it is understood that the main clause did not terminate normally, and so the main clause could not have produced any bindings.
catch clauses are processed prior to finally.
If a catch clause itself throws an exception, that exception cannot be caught by that same clause or its siblings in the same try block. The catch clauses of that block are no longer visible at that point. Nevertheless, the catch clauses are still protected by the finally block. If a catch clause throws, or otherwise terminates, the finally block is still processed.
If a finally block throws an exception, then it is simply aborted; the remaining directives in that block are not processed.
So the success or failure of the try block depends on the behavior of the catch clause or the finally clause, if there is one. If either of them succeed, then the try block is considered a successful match.
Example:
Here, the try block's main clause is terminated abruptly by a file-error exception from the @(next) directive. This is handled by the catch clause, which binds variable a to the input line "1". Then the finally clause executes, binding b to "2". The try block then terminates successfully, and so @c takes "3".
A catch clause may have parameters following the type name, like this:
@(catch pair (a b))
To write a catch-all with parameters, explicitly write the master supertype t:
Parameters are useful in conjunction with throw. The built-in error exceptions carry one argument, which is a string containing the error message. Using throw, arbitrary parameters can be passed from the throw site to the catch site.
The throw directive generates an exception. A type must be specified, followed by optional arguments, which are bind expressions. For example,
@(throw pair "a" `@file.txt`)
throws an exception of type pair, with two arguments, being "a" and the expansion of the quasiliteral `@file.txt`.
The selection of the target catch is performed purely using the type name; the parameters are not involved in the selection.
Binding takes place between the arguments given in throw and the target catch.
If any catch parameter, for which a throw argument is given, is a bound variable, it has to be identical to the argument, otherwise the catch fails. (Control still passes to the catch, but the catch is a failed match).
If any argument is an unbound variable, the corresponding parameter in the catch is left alone: if it is an unbound variable, it remains unbound, and if it is bound, it stays as is.
If a catch has fewer parameters than there are throw arguments, the excess arguments are ignored:
If a catch has more parameters than there are throw arguments, the excess parameters are left alone. They may be bound or unbound variables.
A throw argument passing a value to a catch parameter which is unbound causes that parameter to be bound to that value.
throw arguments are evaluated in the context of the throw, and the bindings which are available there. Consideration of what parameters are bound is done in the context of the catch.
In the above example, c has a top-level binding to the string "c", but then becomes unbound via forget within the try construct, and rebound to the value "lc". Since the try construct is terminated by a throw, these modifications of the binding environment are discarded. Hence, at the end of the query, variable c ends up bound to the original value "c". The throw still takes place within the scope of the bindings set up by the try clause, so the values of a and c that are thrown are "a" and "lc". However, at the catch site, variable a does not have a binding. At that point, the binding to "a" established in the try has disappeared already. Being unbound, the catch parameter a can take whatever value the corresponding throw argument provides, so it ends up with "lc".
There is a horizontal form of throw. For instance:
abc@(throw e 1)
throws exception e if abc matches.
If throw is used to generate an exception derived from type error and that exception is not handled, TXR will issue diagnostics on the *stderr* stream and terminate. If an exception derived from warning is not handled, TXR will generate diagnostics on the *stderr* stream, after which control returns to the throw directive, and proceeds with the next directive. If an exception not derived from error is thrown, control returns to the throw directive and proceeds with the next directive.
The defex directive allows the query writer to invent custom exception types, which are arranged in a type hierarchy (meaning that some exception types are considered subtypes of other types).
Subtyping means that if an exception type B is a subtype of A, then every exception of type B is also considered to be of type A. So a catch for type A will also catch exceptions of type B. Every type is a supertype of itself: an A is a kind of A. This implies that every type is a subtype of itself also. Furthermore, every type is a subtype of the type t, which has no supertype other than itself. Type nil is a subtype of every type, including itself. The subtyping relationship is transitive also. If A is a subtype of B, and B is a subtype of C, then A is a subtype of C.
defex may be invoked with no arguments, in which case it does nothing:
@(defex)
It may be invoked with one argument, which must be a symbol. This introduces a new exception type. Strictly speaking, such an introduction is not necessary; any symbol may be used as an exception type without being introduced by @(defex):
@(defex a)
Therefore, this also does nothing, other than document the intent to use a as an exception.
If two or more argument symbols are given, the symbols are all introduced as types, engaged in a subtype-supertype relationship from left to right. That is to say, the first (leftmost) symbol is a subtype of the next one, which is a subtype of the next one and so on. The last symbol, if it had not been already defined as a subtype of some type, becomes a direct subtype of the master supertype t. Example:
The first directive defines d as a subtype of e, and e as a subtype of t. The second defines a as a subtype of b, b as a subtype of c, and c as a subtype of d, which is already defined as a subtype of e. Thus a is now a subtype of e. The above can be condensed to:
@(defex a b c d e)
Example:
Exception types have a pervasive scope. Once a type relationship is introduced, it is visible everywhere. Moreover, the defex directive is destructive, meaning that the supertype of a type can be redefined. This is necessary so that something like the following works right:
@(defex gorilla ape)
@(defex ape primate)
These directives are evaluated in sequence. So after the first one, the ape type has the type t as its immediate supertype. But in the second directive, ape appears again, and is assigned the primate supertype, while retaining gorilla as a subtype. This situation could be diagnosed as an error, forcing the programmer to reorder the statements, but instead TXR obliges. However, there are limitations. It is an error to define a subtype-supertype relationship between two types if they are already connected by such a relationship, directly or transitively. So the following definitions are in error:
@(defex a b)
@(defex b c)
@(defex a c)@# error: a is already a subtype of c, through b
@(defex x y)
@(defex y x)@# error: circularity; y is already a supertype of x.
The assert directive requires the remaining query or subquery which follows it to match. If the remainder fails to match, the assert directive throws an exception. If the directive is simply
@(assert)
Then it throws an assertion of type assert, which is a subtype of error. The assert directive also takes arguments similar to the throw directive: an exception symbol and additional arguments which are bind expressions, and may be unbound variables. The following assert directive, if it triggers, will throw an exception of type foo, with arguments 1 and "2":
@(assert foo 1 "2")
Example:
@(collect)
Important Header
----------------
@(assert)
Foo: @a, @b
@(end)
Without the assertion in places, if the Foo: @a, @b part does not match, then the entire interior of the @(collect) clause fails, and the collect continues searching for another match.
With the assertion in place, if the text "Important Header" and its underline match, then the remainder of the collect body must match, otherwise an exception is thrown. Now the program will not silently skip over any Important Header sections due to a problem in its matching logic. This is particularly useful when the matching is varied with numerous cases, and they must all be handled.
There is a horizontal assert directive also. For instance:
abc@(assert)d@x
asserts that if the prefix "abc" is matched, then it must be followed by a successful match for "d@x", or else an exception is thrown.
If the exception is not handled, and is derived from error then TXR issues diagnostics on the *stderr* stream and terminates. If the exception is derived from warning and not handled, TXR issues a diagnostic on *stderr* after which control returns to the assert directive. Control silently returns to the assert directive if an exception of any other kind is not handled.
When control returns to assert due to an unhandled exception, it behaves like a failed match, similarly to the require directive.
The TXR language contains an embedded Lisp dialect called TXR Lisp.
This language is exposed in TXR in a number of ways.
In any situation that calls for an expression, a Lisp expression can be used, if it is preceded by the @ character. The Lisp expression is evaluated and its value becomes the value of that expression. Thus, TXR directives are embedded in literal text using @, and Lisp expressions are embedded in directives using @ also.
Furthermore, certain directives evaluate Lisp expressions without requiring @. These are @(do), @(require), @(assert), @(if) and @(next).
TXR Lisp code can be placed into files. On the command line, TXR treats files with a ".tl", ".tlo" or ".tlo.gz" suffix as TXR Lisp source or compiled code, and the @(load) directive does also.
TXR also provides an interactive listener for Lisp evaluation.
Lastly, TXR Lisp expressions can be evaluated via the command line, using the -e and -p options.
Examples:
Bind variable a to the integer 4:
Bind variable b to the standard input stream. Note that @ is not required on a Lisp variable:
@(bind a *stdin*)
Define several Lisp functions inside @(do):
@(do
(defun add (x y) (+ x y))
(defun occurs (item list)
(cond ((null list) nil)
((atom list) (eql item list))
(t (or (eq (first list) item)
(occurs item (rest list)))))))
Trigger a failure unless previously bound variable answer is greater than 42:
@(require (> (int-str answer) 42)
TXR Lisp is a small and simple dialect, like Scheme, but much more similar to Common Lisp than Scheme. It has separate value and function binding namespaces, like Common Lisp (and thus is a Lisp-2 type dialect), and represents Boolean true and false with the symbols t and nil (note the case sensitivity of identifiers denoting symbols!). Furthermore, the symbol nil is also the empty list, which terminates nonempty lists.
TXR Lisp has lexically scoped local variables and dynamic global variables, similarly to Common Lisp, including the convention that defvar marks symbols for dynamic binding in local scopes. Lexical closures are supported. TXR Lisp also supports global lexical variables via defvarl.
Functions are lexically scoped in TXR Lisp; they can be defined in the pervasive global environment using defun or in local scopes using flet and labels.
Much of the TXR Lisp syntax has been introduced in the previous sections of the manual, since directive forms are based on it. There is some additional syntax that is useful in TXR Lisp programming.
The symbol tokens in TXR Lisp, called a lident (Lisp identifier) has a similar syntax to the bident (braced identifier) in the TXR pattern language. It may consist of all the same characters, as well as the / (slash) character which may not be used in a bident. Thus a lident may consist of these characters, in addition to letters, numbers and underscores:
! $ % & * + - < = > ? \ ~ /
and may not look like a number.
A lident may also include all of the Unicode characters which are permitted in a bident.
The one character which is allowed in a lident but not in a bident is / (forward slash).
A lone / is a valid lident and consequently a symbol token in TXR Lisp. The token /abc/ is also a symbol, and, unlike in a braced expression, is not a regular expression. In TXR Lisp expressions, regular expressions are written with a leading #.
If a symbol name contains a colon, the lident characters, if any, before that colon constitute the package prefix.
For example, the syntax foo:bar denotes bar symbol in the foo package.
It is a syntax error to read a symbol whose package doesn't exist.
If the package exists, but the symbol name doesn't exist in that package, then the symbol is interned in that package.
If the package name is an empty string (the colon is preceded by nothing), the package is understood to be the keyword package. The symbol is interned in that package.
The syntax :test denotes the symbol test in the keyword package, the same as keyword:test.
Symbols in the keyword package are self-evaluating. This means that when a keyword symbol is evaluated as a form, the value of that form is the keyword symbol itself. Exactly two non-keyword symbols also have this special self-evaluating behavior: the symbols t and nil in the user package, whose fully qualified names are usr:t and usr:nil.
The syntax @foo:bar denotes the meta prefix @ being applied to the foo:bar symbol, not to a symbol in the @foo package.
The syntax #:bar denotes an uninterned symbol named bar, described in the next section.
In ANSI Common Lisp, the foo:bar syntax does not intern the symbol bar in the foo package; the symbol must exist and be an exported symbol, or else the syntax is erroneous. In ANSI Common Lisp, the syntax foo::bar does intern foo in the bar package. TXR's package system has no double-colon syntax, and lacks the concept of exported symbols.
Uninterned symbols are written with the #: prefix, followed by zero or more lident characters. When an uninterned symbol is read, a new, unique symbol is constructed, with the specified name. Even if two uninterned symbols have the same name, they are different objects. The make-sym and gensym functions produce uninterned symbols.
"Uninterned" means "not entered into a package". Interning refers to a process which combines package lookup with symbol creation, which ensures that multiple occurrences of a symbol name in written syntax are all converted to the same object: the first occurrence creates the symbol and associates it with its name in a package. Subsequent occurrences do not create a new symbol, but retrieve the existing one.
An expression may be preceded by the @ (at sign) character. If the expression is an atom, then this is a meta-atom, otherwise it is a meta-expression.
When the atom is a symbol, this is also called a meta-symbol and in situations when such a symbol behaves like a variable, it is also referred to as a meta-variable.
When the atom is an integer, the meta-atom expression is called a meta-number.
Meta-atom and meta-expression expressions have no evaluation semantics; evaluating them throws an exception. They play a syntactic role in the op operator, which makes use of meta-variables and meta-numbers, and in structural pattern matching, which uses meta-variables as pattern variables and whose operator vocabulary is based on meta-expressions.
Meta-expressions also appear in the quasiliteral notation.
In other situations, application code may assign meaning to meta syntax as the programmer sees fit.
Meta syntax is defined as a shorthand notation, as follows:
If X is the syntax of an atom, such as a symbol, string or vector, then @X is a shorthand for the expression (sys:var X). Here, sys:var refers to the var symbol in the system-package.
If X is a compound expression, either (...) or [...], then @X is a shorthand for the expression (sys:expr X).
The behavior of @ followed by the syntax of a floating-point constant introduced by a leading decimal point, not preceded by digits, is unspecified. Examples of this are @.123 and @.123E+5.
The behavior of @ followed by the syntax of a floating-point expression in E notation, which lacks a decimal point, is also unspecified. An example of this is @12E5.
It is a syntax error for @ to be followed by what appears to be a floating-point constant consisting of a decimal point flanked by digits on both sides. For instance @1.2 is rejected.
A meta-expression followed by a period, and the syntax of another object is otherwise interpreted as a referencing dot expression. For instance @1.E3 denotes (qref @1 E3) which, in turn, denotes (qref (sys:var 1) E3), even though the unprefixed character sequence 1.E3 is otherwise a floating-point constant.
Unlike other major Lisp dialects, TXR Lisp allows a consing dot with no forms preceding it. This construct simply denotes the form which follows the dot. That is to say, the parser implements the following transformation:
(. expr) -> expr
This is convenient in writing function argument lists that only take variable arguments. Instead of the syntax:
(defun fun args ...)
the following syntax can be used:
(defun fun (. args) ...)
When a lambda form is printed, it is printed in the following style.
(lambda nil ...) -> (lambda () ...)
(lambda sym ...) -> (lambda (. sym) ...)
(lambda (sym) ...) -> (lambda (sym) ...)
In no other circumstances is nil printed as (), or an atom sym as (. sym).
This notation is implemented for the square brackets, according to this transformation:
[. expr] -> (dwim . expr)
This is useful in Structural Pattern Matching, allowing a pattern like
[. @args]
to match a dwim expression and capture all of its arguments in a variable, without having to resort to the internal notation:
Compatibility Note: support for [. expr] was introduced in TXR 282. Older versions do not read the syntax, but do print (dwim . @var) as [. @var] which is then unreadable in those versions, breaking read-print consistency.
A dot token which is flanked by expressions on both sides, without any intervening whitespace, is the referencing dot, and not the consing dot. The referencing dot is a syntactic sugar which translated to the qref syntax ("quoted ref"). When evaluated as a form, this syntax denotes structure access; see Structures. However, it is possible to put this syntax to use for other purposes, in other contexts.
;; a.b may be almost any expressions
a.b <--> (qref a b)
a.b.c <--> (qref a b c)
a.(qref b c) <--> (qref a b c)
(qref a b).c <--> (qref (qref a b) c)
That is to say, this dot operator constructs a qref expression out of its left and right arguments. If the right argument of the dot is already a qref expression (whether produced by another instance of the dot operator, or expressed directly) it is merged. This requires the qref dot operator to be right-to-left associative, so that a.b.c works by first translating b.c to (qref b c), and then adjoining a to produce (qref a b c).
If the referencing dot is immediately followed by a question mark, it forms a single token, which produces the following syntactic variation, in which the following item is annotated as a list headed by the symbol t:
a.?b <--> (t a).b <--> (qref (t a) b)
a.?b.?c <--> (t a).(t b).c <--> (qref (t a) (t b) c)
a.?(b) <--> (t a).(b) <--> (qref (t a) (b))
(a).?b <--> (t (a)).b <--> (qref (t (a)) b)
This syntax denotes null-safe access to structure slots and methods. a.?b means that a may evaluate to nil, in which case the expression yields nil; otherwise, a must evaluate to a struct which has a slot b, and the expression denotes access to that slot. Similarly, a.?(b 1) means that if a evaluates to nil, the expression yields nil; otherwise, a is treated as a struct object whose method b is invoked with argument 1, and the value returned by that method becomes the value of the expression.
Integer tokens cannot be involved in this syntax, because they form floating-point constants when juxtaposed with a dot. Such ambiguous uses of floating-point tokens are diagnosed as syntax errors:
(a.4) ;; error: cramped floating-point literal
(a .4) ;; good: a followed by 0.4
Closely related to the referencing dot syntax is the unbound referencing dot. This is a dot which is flanked by an expression on the right, without any intervening whitespace, but is not preceded by an expression Rather, it is preceded by whitespace, or some punctuation such as [, ( or '. This is a syntactic sugar which translates to uref syntax:
.a <--> (uref a)
.a.b <--> (uref a b)
.a.?b <--> (uref (t a) b)
If the unbound referencing dot is itself combined with a question mark to form the .? token, then the translation to uref is as follows:
.?a <--> (uref t a)
.?a.b <--> (uref t a b)
.?a.?b <--> (uref t a (t b))
When the unbound referencing dot is applied to a dotted expression, this can be understood as a conversion of qref to uref.
Indeed, this is exactly what happens if the unbound dot is applied to an explicit qref expression:
The unbound referencing dot takes its name from the semantics of the uref macro, which produces a function that implements late binding of an object to a method slot. Whereas the expression obj.a.b denotes accessing object obj to retrieve slot a and then accessing slot b of the object from that slot, the expression .a.b. represents a "disembodied" reference: it produces a function which takes an object as an argument and then performs the implied slot referencing on that argument. When the function is called, it is said to bind the referencing to the object. Hence that referencing is "unbound".
Whereas the expression .a produces a function whose argument must be an object, .?a produces a function whose argument may be nil. The function detects this case and returns nil.
Under a quasiquote, form is considered to be a quasiquote template. The template is considered to be a literal structure, except that it may contain the notations ,expr and ,*expr which denote non-constant parts.
A quasiquote gets translated into code which, when evaluated, constructs the structure implied by qq-template, taking into account the unquotes and splices.
A quasiquote also processes nested quasiquotes specially.
If qq-template does not contain any unquotes or splices (which match its level of nesting), or is simply an atom, then ^qq-template is equivalent to 'qq-template . in other words, it is like an ordinary quote. For instance ^(a b ^(c ,d)) is equivalent to '(a b ^(c ,d)). Although there is an unquote ,d it belongs to the inner quasiquote ^(c ,d), and the outer quasiquote does not have any unquotes of its own, making it equivalent to a quote.
Dialect Note: in Common Lisp and Scheme, ^form is written `form, and quasiquotes are also informally known as backquotes. In TXR, the backquote character ` used for quasistring literals.
Note: if a variable is called *x*, then the syntax ,*x* means ,* x*: splice the value of x*. In this situation, whitespace between the comma and the variable name must be used: , *x*.
In other Lisp dialects, like Scheme and ANSI Common Lisp, the equivalent syntax is usually ,@ (comma at). The @ character already has an assigned meaning in TXR, so * is used.
However, * is also a character that may appear in a symbol name, which creates a potential for ambiguity. The syntax ,*abc denotes the application of the ,* splicing operator to the symbolic expression abc; to apply the ordinary non-splicing unquote to the symbol *abc, whitespace must be used: , *abc.
In TXR, the unquoting and splicing forms may freely appear outside of a quasiquote template. If they are evaluated as forms, however, they throw an exception:
,(+ 2 2) ;; error!
',(+ 2 2) --> ,(+ 2 2)
In other Lisp dialects, a comma not enclosed by backquote syntax is treated as a syntax error by the reader.
TXR's quasiquote supports splicing multiple items into a quote, if that quote is itself evaluated via an unquote. Concretely, these two examples produce the same result:
(eval
(eval
(let ((args '(a b c)))
^^(let ((a 1) (b 2) (c 3))
(list ,',*args)))))
-> (1 2 3)
(eval
(eval
(let ((args '(a b c)))
^^(let ((a 1) (b 2) (c 3))
(list ,*',args)))))
-> (1 2 3)
The only difference is that the former example uses ,',*args whereas the latter ,*',args. Thus the former example splices args into the quote as if by (quote ,*args) which is invalid quote syntax if args doesn't expand to exactly one element. This invalid quote syntax is accepted by the quasiquote expander when it occurs in the above unquoting and splicing situation. Effectively, it behaves as if the splice distributes across the quoted unquote, such that all the arguments of the quote end up individually quoted, and spliced into the surrounding list.
The Common Lisp equivalent this combination, ,',@args, works in some Common Lisp implementations, such as CLISP.
'#(1 2 3)
The #(1 2 3) literal is turned into a vector atom right in the TXR parser, and this atom is being quoted: this is (quote atom) syntactically, which evaluates to atom.
When a vector is quasi-quoted, this is a case of ^atom which evaluates to atom.
A vector can be quasiquoted, for example:
^#(1 2 3)
Unquotes can occur within a quasiquoted vector:
(let ((a 42))
^#(1 ,a 3)) ; value is #(1 42 3)
In this situation, the ^#(...) notation produces code which constructs a vector.
The vector in the following example is also a quasivector. It contains unquotes, and though the quasiquote is not directly applied to it, it is embedded in a quasiquote:
(let ((a 42))
^(a b c #(d ,a))) ; value is (a b c #(d 42))
Hash-table literals have two parts: the list of hash construction arguments and the key-value pairs. For instance:
#H((:eql-based) (a 1) (b 2))
where (:eql-based) indicates that this hash table's keys are treated using eql equality, and (a 1) and (b 2) are the key/value entries. Hash literals may be quasiquoted. In quasiquoting, the arguments and pairs are treated as separate syntax; it is not one big list. So the following is not a possible way to express the above hash:
;; not supported: splicing across the entire syntax
(let ((hash-syntax '((:eql-based) (a 1) (b 2))))
^#H(,*hash-syntax))
This is correct:
;; fine: splicing hash arguments and contents separately
(let ((hash-args '(:eql-based))
(hash-contents '((a 1) (b 2))))
^#H(,hash-args ,*hash-contents))
Example:
(eval (let ((a 3)) ^`abc @,a @{,a} @{(list 1 2 ,a)}`))
-> "abc 3 3 1 2 3"
When a struct literal is read, the denoted struct type is constructed as if by a call to make-struct with an empty plist argument, followed by a sequence of assignments which store into each slot the corresponding value expression.
An empty list can be specified as nil or (), which defaults to a hash table based on the eql function, with no weak semantics or user data.
The entire syntax following #H may be an empty list; however, that empty list may not be specified as nil; the empty parentheses notation is required.
The hash table's key-value contents are specified as zero or more two-element lists, whose first element specifies the key and whose second specifies the value. Both expressions are literal objects, not subject to evaluation.
Buffers may be constructed by the make-buf function, and other means such as the ffi-get function.
Note that the #b prefix is also used for binary numbers. In that syntax, it is followed by an optional sign, and then a mixture of one or more of the digits 0 or 1.
A tree node is an object of type tnode. Every tnode has three elements: a key, a left link and a right link. They may be objects of any type. If the tree node literal syntax omits any of these, they default to nil.
The list syntax which follows #T may be empty. If so, it cannot be written as nil.
The first element of the #T syntax, if present, must be a list of zero to three elements. These elements are symbols giving the names of the tree object's key abstraction functions. keyfun specifies the key function which is applied to each element to retrieve its key. If it is omitted, the object shall use the identity function as its key. The lessfun specifies the name of the comparison function by which keys are compared for inequality. It defaults to less. The equalfun specifies the function by which keys are compared for equality. It defaults to equal. A symbol which is specified as the name of any of these three special functions must be an element of the list stored in the special variable *tree-fun-whitelist*, otherwise the string literal is diagnosed as erroneous. Note: this is due to security considerations, since these three functions are executed during the processing of tree syntax.
A tree object is constructed from a tree literal by first creating an empty tree endowed with the three key abstraction functions that are indicated in the syntax, either explicitly or as defaults. Then, every element object is constructed from its respective literal syntax and inserted into the tree.
Duplicate objects are preserved. For instance the tree literal #T(() 1 1 1) specifies a tree with three nodes which have the same key. Duplicates appear in the tree in the order that they appear in the literal.
The implementation of JSON syntax is based on, and intended to conform with the IETF RFC 8259 document. Only TXR's extensions to JSON syntax are described in this manual, as well as the correspondence between JSON syntax and Lisp.
The json-syntax is translated into a TXR Lisp object as follows.
A JSON string corresponds to a Lisp string. A JSON number corresponds to a Lisp floating-point number. A JSON array corresponds to a Lisp vector. A JSON object corresponds to an equal-based hash table.
The JSON Boolean symbols true and false translate to the Lisp symbols t and nil, respectively, those being the standard ones in the usr package.
The JSON symbol null maps to the null symbol in the usr package.
The #Jjson-syntax expression produces the object:
(json quote lisp-object)
where lisp-object is the Lisp value which corresponds to the json-syntax.
Similarly, but with a key difference, the #J^json-syntax expression produces the object:
(json sys:qquote lisp-object)
in which quote has been replaced with sys:qquote.
The json symbol is bound as a macro, which is expanded when a #J expression is evaluated.
The following remarks indicate special treatment and extensions in the processing of JSON. Similar remarks regarding the production of JSON are given under the put-json function.
When an invalid UTF-8 byte is encountered inside a JSON string, its value is mapped into the code point range U+DC01 to U+DCFF. That byte is consumed, and decoding continues with the next byte. This treatment is consistent with the treatment of invalid UTF-8 bytes in TXR Lisp literals and I/O streams. If the valid UTF-8 byte U+0000 (ASCII NUL) occurs in a JSON string, it is also mapped to U+DC00, TXR's pseudo-null character. This treatment is consistent with TXR string literals and I/O streams.
The JSON escape sequence \u0000 denoting the U+0000 NUL character is also converted to U+DC00.
TXR Lisp does not impose the restriction that the keys in a JSON object must be strings: #J{1:2,true:false} is accepted.
TXR Lisp allows the circle notation to occur within JSON syntax. See the section Notation for Circular and Shared Structure.
TXR Lisp supports the extension of Lisp comments in JSON. When the ; character (semicolon) occurs in the middle of JSON syntax, outside of a token, that character and all characters until the end of the line constitute a comment that is discarded. TXR Lisp never produces comments when printing JSON.
TXR Lisp allows for JSON syntax to be quasiquoted, and provides two extensions for writing unquotes and splicing unquotes. Within a JSON quasiquote, the ~ (tilde) character introduces a Lisp expression whose value is to be substituted at that point. Thus, the tilde serves the role of the unquoting comma used in Lisp quasiquotes. Splicing is indicated by the character sequence ~*, which introduces a Lisp expression that is expected to produce a list, whose elements are interpolated into the JSON value.
Note: quasiquoting allows Lisp values to be introduced into the resulting object which are outside of the JSON type system, such as integers, characters, symbols or structures. These objects have no representation in JSON syntax.
;; Basic JSON:
#Jtrue -> t
#Jfalse -> nil
(list #J true #Jtrue #Jfalse) -> (t t nil)
#J[1, 2, 3.14] -> #(1.0 2.0 3.14)
#J{"foo":"bar"} -> #H(() ("foo" "bar"))
;; Quoting JSON shows the json expression
'#Jfalse -> (json quote ())
'#Jtrue -> (json quote t)
'#J["a", true, 3.0] -> (json quote #("a" t 3.0))
'#J^[~(+ 2 2), 3] -> (json sys:qquote #(,(+ 2 2) 3.0))
:; Circle notation:
#J[#1="abc", #1#, #1#] -> #("abc" "abc" "abc")
;; JSON Quasiquote:
#J^[~*(list 1.0 2.0 3.0), ~(* 2.0 2), 5.0]
--> #(1.0 2.0 3.0 4.0 5.0)
;; Lisp quasiquote around JSON quote: requires evaluation round.
^#J[~*(list 1.0 2.0 3.0), ~(* 2.0 2), 5.0]
--> (json quote #(1.0 2.0 3.0 4.0 5.0))
(eval ^#J[~*(list 1.0 2.0 3.0), ~(* 2.0 2), 5.0])
--> #(1.0 2.0 3.0 4.0 5.0)
;; Comment extension
#J[1, ; Comment inside JSON.
2, ; Another one.
3] ; Lisp comment outside of JSON.
--> #(1.0 2.0 3.0)
That is to say, A .. B translates to (rcons A B), and so for instance (a b .. (c d) e .. f . g) means (a (rcons b (c d)) (rcons e f) . g).
The rcons function constructs a range object, which denotes a pair of values. Range objects are most commonly used for referencing subranges of sequences.
For instance, if L is a list, then [L 1 .. 3] computes a sublist of L consisting of elements 1 through 2 (counting from zero).
Note that if this notation is used in the dot position of an improper list, the transformation still applies. That is, the syntax (a . b .. c) is valid and produces the object (a . (rcons b c)) which is another way of writing (a rcons b c), which is quite probably nonsense.
The notation's .. operator associates right to left, so that a..b..c denotes (rcons a (rcons b c)).
Note that range objects are not printed using the dotdot notation. A range literal has the syntax of a two-element list, prefixed by #R. (See Range Literals above.)
In any context where the dotdot notation may be used, and where it is evaluated to its value, a range literal may also be specified. If an evaluated dotdot notation specifies two constant expressions, then an equivalent range literal can replace it. For instance the form [L 1 .. 3] can also be written [L #R(1 3)]. The two are syntactically different, and so if these expressions are being considered for their syntax rather than value, they are not the same.
For instance if foo is a variable which holds a function object, then [foo 3] can be used to call it, instead of (call foo 3). If foo is a vector, then [foo 3] retrieves the fourth element, like (vecref foo 3). Indexing over lists, strings and hash tables is possible, and the notation is assignable.
Furthermore, any arguments enclosed in [] which are symbols are treated according to a modified namespace lookup rule.
More details are given in the documentation for the dwim operator.
The first position of an ordinary Lisp-2 style compound form, is expected to have a function or operator name. Then arguments follow. There may also be an expression in the dotted position, if the form is a function call.
If the form is a function call then the arguments are evaluated. If any of the arguments are symbols, they are treated according to Lisp-2 namespacing rules.
A function name may be a symbol, or else any of the syntactic forms given in the description of the function func-get-name.
If there is an expression in the dotted position of a function call expression, it is also evaluated, and the resulting value is involved in the function call in a special way.
Firstly, note that a compound form cannot be used in the dot position, for obvious reasons, namely that (a b c . (foo z)) does not mean that there is a compound form in the dot position, but denotes an alternate spelling for (a b c foo z), where foo behaves as a variable.
If the dot position of a compound form is an atom, then the behavior may be understood according to the following transformations:
(f a b c ... . x) --> (apply (fun f) a b c ... x)
[f a b c ... . x] --> [apply f a b c ... x]
In addition to atoms, meta-expressions and meta-symbols can appear in the dot position, even though their underlying syntax is actually a compound expression. This is made to work according to a transformation pattern which superficially resembles the above one for atoms:
(f a b c ... . @x) --> (apply (fun f) a b c ... @x)
However, in this situation, the @x is a notation denoting the expression (sys:var x) and thus the entire form is a proper list, not a dotted list. With the underlying syntax revealed, the transformation looks like this:
(f a b c ... sys:var x) --> (apply (fun f) a b c ... (sys:var @x))
That is to say, the TXR Lisp form expander reacts to the presence of a sys:var or sys:expr atom in embedded in the form. That symbol and the items which follow it are wrapped in an additional level of nesting, converted into a single compound form element.
Effectively, in all these cases, the dot notation constitutes a shorthand for apply.
Examples:
;; a contains 3
;; b contains 4
;; c contains #(5 6 7)
;; s contains "xyz"
(foo a b . c) ;; calls (foo 3 4 5 6 7)
(foo a) ;; calls (foo 3)
(foo . s) ;; calls (foo #\x #\y #\z)
(list . a) ;; yields 3
(list a . b) ;; yields (3 . 4)
(list a . c) ;; yields (3 5 6 7)
(list* a c) ;; yields (3 . #(5 6 7))
(cons a . b) ;; error: cons isn't variadic.
(cons a b . c) ;; error: cons requires exactly two arguments.
[foo a b . c] ;; calls (foo 3 4 5 6 7)
[c 1] ;; indexes into vector #(5 6 7) to yield 6
(call (op list 1 . @1) 2) ;; yields (1 . 2)
Note that the atom in the dot position of a function call may be a symbol macro. Since the semantics works as if by transformation to an apply form in which the original dot position atom is an ordinary argument, the symbol macro may produce a compound form.
Thus:
(symacrolet ((x 2))
(list 1 . x)) ;; yields (1 . 2)
(symacrolet ((x (list 1 2)))
(list 1 . x)) ;; yields (1 1 2)
That is to say, the expansion of x is not substituted into the form (list 1 . x) but rather the transformation to apply syntax takes place first, and so the substitution of x takes place in a form resembling (apply (fun list) 1 x).
Dialect Note:
In some other Lisp dialects like ANSI Common Lisp, the improper list syntax may not be used as a function call; a function called apply (or similar) must be used for application even if the expression which gives the trailing arguments is a symbol. Moreover, applying sequences other than lists is not supported.
TXR Lisp allows macros to be called using forms which are improper lists. These forms are simply destructured by the usual macro parameter list destructuring. To be callable this way, the macro must have an argument list which specifies a parameter match in the dot position. This dot position must either match the terminating atom of the improper list form, or else match the trailing portion of the improper list form.
For instance if a macro mac is defined as
(defmacro mac (a b . c) ...)
then it may not be invoked as (mac 1 . 2) because the required argument b is not satisfied, and so the 2 argument cannot match the dot position c as required. The macro may be called as (mac 1 2 . 3) in which case c receives the form 3. If it is called as (mac 1 2 3 . 4) then c receives the improper list form 3 . 4.
TXR Lisp supports a printed notation called circle notation which accurately articulates the representation of objects which contain shared substructures as well as circular references. The notation is supported as a means of input, and is also optionally produced as output, controlled by the *print-circle* variable.
Ordinarily, shared substructure in printed objects is not evident, except in the case of multiple occurrences of interned symbols, in whose semantics it is implicit that they refer to the same object. Other shared structure is printed as separate copies which look like distinct objects. For instance, the object produced by (let ((shared '(1 2))) (list shared shared)) is printed as ((1 2) (1 2)), where it is not clear that the two occurrences of (1 2) are actually the same object. Under the circle notation, this object can be represented as (#5=(1 2) #5#). The #5= part introduces a reference label, associating the arbitrarily chosen nonnegative integer 5 with the object which follows. The subsequent notation #5# simply refers to the object labeled by 5, reproducing that object by reference. The result is a two-element list which has the same (1 2) in two places.
Circular structure presents a greater challenge to printing: namely, if it is printed by a naive recursive descent, it results in infinite output, and possibly stack exhaustion due to recursion. The circle notation detects and handles circular references. For instance, the object produced by (let ((c (list 1))) (rplacd c c)) produces a circular list which looks like an infinite list of 1's: (1 1 1 1 ...). This cannot be printed. However, under the circle notation, it can be represented as #1=(1 . #1#). The entire object itself is labeled by the integer 1. Then, enclosed within the syntax of that labeled object itself, a reference occurs to the label. This circular label reference represents the corresponding circular reference in the object.
A detailed description of the notational elements follows:
There may be no more than one definition for a given label within the syntactic scope being parsed, otherwise a syntax error occurs. In TXR pattern language code, an entire source file is parsed as one unit, and so scope for the circular notation's references is the entire source file. Files processed by @(include) have their own scope. The scope for labels in TXR Lisp source code is the top-level expression in which they appear. Consequently, references in one TXR Lisp top-level expression cannot reach definitions in another.
Note:
Circular notation can span hash-table literals. The syntax #1=#H((:eql-based) (#1# #1#)) denotes an eql-based hash table which contains one entry, in which that same table itself is both the key and value. This kind of circularity is not supported for equal-based hash tables. The analogous syntax #1=#H(() (#1# #1#)) produces a hash table in an inconsistent state.
Dialect Note:
Circle notation is taken from Common Lisp, intended to be unsurprising to users familiar with that language. The implementation is based on descriptions in the ANSI Common Lisp document, judiciously taking into account the content of the X3J13 Cleanup Issues named PRINT-CIRCLE-STRUCTURE:USER-FUNCTIONS-WORK and PRINT-CIRCLE-SHARED:RESPECT-PRINT-CIRCLE.
This is useful for temporarily "commenting out" an expression.
Notes:
Whereas it is valid for a TXR Lisp source file to be empty, it is a syntax error if a TXR Lisp source file contains nothing but one or more objects which are each suppressed by a preceding #;. In the interactive listener, an input line consisting of nothing but commented-out objects is similarly a syntax error.
The notation does not cascade; consecutive occurrences of #; trigger a syntax error.
The notation interacts with the circle notation. Firstly, if an object which is erased by #; contains circular-referencing instances of the label notation, those instances refer to nil. Secondly, commented-out objects may introduce labels which are subsequently referenced in expr. An example of the first situation occurs in:
#;(#1=(#1#))
Here the #1# label is a circular reference because it refers to an object which is a parent of the object which contains that reference. Such a reference is only satisfied by a "backpatching" process once the entire surrounding syntax is processed to the top level. The erasure perpetrated by #; causes the #1# label reference to be replaced by nil, and therefore the labeled object is the object (nil).
An example of the second situation is
#;(#2=(a b c)) #2#
Here, even though the expression (#2=(a b c)) is suppressed, the label definition which it has introduced persists into the following object, where the label reference #2# resolves to (a b c).
A combination of the two situations occurs in
#;(#1=(#1#)) #1#
which yields (nil). This is because the #1= label is available; but the earlier #1# reference, being a circular reference inside an erased object, had lapsed to nil.
In ancient Lisp in the 1960's, it was not possible to apply the operations car and cdr to the nil symbol (empty list), because it is not a cons cell. In the InterLisp dialect, this restriction was lifted: these operations were extended to accept nil (and return nil). The convention was adopted in other Lisp dialects such as MacLisp and eventually in Common Lisp. Thus there exists an object which is not a cons, yet which takes car and cdr.
In TXR Lisp, this relaxation is extended further. For the sake of convenience, the operations car and cdr, are made to work with strings and vectors:
(cdr "") -> nil
(car "") -> nil
(car "abc") -> #\a
(cdr "abc") -> "bc"
(cdr #(1 2 3)) -> #(2 3)
(car #(1 2 3)) -> 1
Moreover, structure types which define the methods car, cdr and nullify can also be treated in the same way.
The ldiff function is also extended in a special way. When the right parameter a non-list sequence, then it uses the equal equality test rather than eq for detecting the tail of the list.
(ldiff "abcd" "cd") -> (#\a #\b)
The ldiff operation starts with "abcd" and repeatedly applies cdr to produce "bcd" and "cd", until the suffix is equal to the second argument: (equal "cd" "cd") yields true.
Operations based on car, cdr and ldiff, such as keep-if and remq extend to strings and vectors.
Most derived list processing operations such as remq or mapcar obey the following rule: the returned object follows the type of the leftmost input list object. For instance, if one or more sequences are processed by mapcar, and the leftmost one is a character string, the function is expected to return characters, which are converted to a character string. However, in the event that the objects produced cannot be assembled into that type of sequence, a list is returned instead.
For example [mapcar list "ab" "12"] returns ((#\a #\b) (#\1 #\2)), because a string cannot hold lists of characters. However [mappend list "ab" "12"] returns "a1b2".
The lazy versions of these functions such as mapcar* do not have this behavior; they produce lazy lists.
TXR Lisp implements a unified paradigm for iterating over sequence-like container structures and abstract spaces such as bounded and unbounded ranges of integers. This concept is based around an iterator abstraction which is directly compatible with Lisp cons-cell traversal in the sense that when iteration takes place over lists, the iterator instance is nothing but a cons cell.
An iterator is created using the constructor function iter-begin which takes a single argument. The argument denotes a space to be traversed; the iterator provides the means for that traversal.
When the iter-begin function is applied to a list (a cons cell or the nil object), the return value is that object itself. The remaining functions in the iterator API then behave like aliases for list processing functions. The iter-more function behaves like identity, iter-item behaves like car and iter-step behaves like cdr.
For example, the following loops not only produce identical behavior, but the iter variable steps through the cons cells in the same manner in both:
;; print all symbols in the list (a b c d):
(let ((iter '(a b c d)))
(while iter
(prinl (car iter))
(set iter (cdr iter))))
;; likewise:
(let ((iter (iter-begin '(a b c d))))
(while (iter-more iter)
(prinl (iter-item iter))
(set iter (iter-step iter))))
There are three important differences.
Firstly, both examples will still work if the list (a b c d) is replaced by a different kind of sequence, such as the string "abcd" or the vector #(a b c d). However, the former example will not execute efficiently on these objects. The reason is that the cdr function will construct successive suffixes of the string and list object. That requires not only the allocation of memory, but changes the running time complexity of the loop from linear to quadratic.
Secondly, the former example with car/cdr will not work correctly if the sequence is an empty non-list sequence, like the null string or empty vector. Rectifying this problem requires the nullify function to be used:
;; print all symbols in the list (a b c d):
(let ((iter (nullify "abcd")))
(while iter
(prinl (car iter))
(set iter (cdr iter))))
The nullify function converts empty sequences of all kinds into the empty list nil.
Thirdly, the second example will work even if the input list is replaced with certain objects which are not sequences at all:
;; Print the integers from 0 to 3
(let ((iter (iter-begin 0..4)))
(while (iter-more iter)
(prinl (iter-item iter))
(set iter (iter-step iter))))
;; Print incrementing integers starting at 1,
;; breaking out of the loop after 100.
(let ((iter (iter-begin 1)))
(while (iter-more iter)
(if (eql 100 (prinl (iter-item iter)))
(return))
(set iter (iter-step iter))))
In TXR Lisp, numerous functions that appear as list processing functions in other contemporary Lisp dialects, and historically, are actually sequence processing functions based on the above iterator paradigm.
In TXR Lisp, sequences (strings, vectors and lists) as well as hashes and regular expressions can be used as functions everywhere, not just with the DWIM brackets.
Sequences work as one- or two-argument functions. With a single argument, an element is selected by position and returned. With two arguments, a range is extracted and returned.
Moreover, when a sequence is used as a function of one argument, and the argument is a range object rather than an integer, then the call is equivalent to the two-argument form. This is the basis for array slice syntax like ["abc" 0..1] .
Hashes also work as one or two argument functions, corresponding to the arguments of the gethash function.
A regular expression behaves as a one, two, or three argument function, which operates on a string argument. It returns the leftmost matching substring, or else nil.
Structure objects are callable if they implement the lambda method.
Integers and ranges are callable like functions. They take one argument, which must be a sequence or hash. An integer selects the corresponding element position from the sequence, and a range extracts a slice of its argument.
Example 1:
(mapcar "abc" '(2 0 1)) -> (#\c #\a #\b)
Here, mapcar treats the string "abc" as a function of one argument (since there is one list argument). This function maps the indices 0, 1 and 2 to the corresponding characters of string "abc". Through this function, the list of integer indices (2 0 1) is taken to the list of characters (#\c #\a #\b).
Example 2:
(call '(1 2 3 4) 1..3) -> (2 3)
Here, the shorthand 1 .. 3 denotes (rcons 1 3). A range used as an argument to a sequence performs range extraction: taking a slice starting at index 1, up to and not including index 3, as if by the call (sub '(1 2 3 4) 1 3).
Example 3:
(call '(1 2 3 4) '(0 2)) -> (1 2)
A sequence applied to a list of index arguments is equivalent to using the select function, as if (select '(1 2 3 4) '(0 2)) were called.
Example 4:
(call #/b./ "abcd") -> "bc"
Here, the regular expression, called as a function, finds the matching substring "bc" within the argument "abcd".
Example 5:
[1 "abcd"] -> #\b
["abcd" 1] -> #\b
An integer used as function indexes into sequence.
This produces the same result as when the sequence is used
as a function with an integer argument.
Example 6:
[1..3 '(a b c d)] -> (b c)
['(a b c d) 1..3] -> (b c)
A range used as a function extracts a slice of its argument.
Similarly to Common Lisp, TXR Lisp is lexically scoped by default, but also has dynamically scoped (a.k.a "special") variables.
When a variable is defined with defvar or defparm, a binding for the symbol is introduced in the global name space, regardless of in what scope the defvar form occurs.
Furthermore, at the time the defvar form is evaluated, the symbol which names the variable is tagged as special.
When a symbol is tagged as special, it behaves differently when it is used in a lexical binding construct like let, and all other such constructs such as function parameter lists. Such a binding is not the usual lexical binding, but a "rebinding" of the global variable. Over the dynamic scope of the form, the global variable takes on the value given to it by the rebinding. When the form terminates, the prior value of the variable is restored. (This is true no matter how the form terminates; even if by an exception.)
Because of this "pervasive special" behavior of a symbol that has been used as the name of a global variable, a good practice is to make global variables have visually distinct names via the "earmuffs" convention: beginning and ending the name with an asterisk.
(defvar *x* 42) ;; *x* has a value of 42
(defun print-x ()
(format t "~a\n" *x*))
(let ((*x* "abc")) ;; this overrides *x*
(print-x)) ;; *x* is now "abc" and so that is printed
(print-x) ;; *x* is 42 again and so "42" is printed
The terms bind and binding are used differently in TXR Lisp compared to ANSI Common Lisp. In TXR Lisp binding is an association between a symbol and an abstract storage location. The association is registered in some namespace, such as the global namespace or a lexical scope. That storage location, in turn, contains a value. In ANSI Lisp, a binding of a dynamic variable is the association between the symbol and a value. It is possible for a dynamic variable to exist, and not have a value. A value can be assigned, which creates a binding. In TXR Lisp, an assignment is an operation which transfers a value into a binding, not one which creates a binding.
In ANSI Lisp, a dynamic variable can exist which has no value. Accessing the value signals a condition, but storing a value is permitted; doing so creates a binding. By contrast, in TXR Lisp a global variable cannot exist without a value. If a defvar form doesn't specify a value, and the variable doesn't exist, it is created with a value of nil.
Unlike ANSI Common Lisp, TXR Lisp has global lexical variables in addition to special variables. These are defined using defvarl and defparml. The only difference is that when variables are introduced by these macros, the symbols are not marked special, so their binding in lexical scopes is not altered to dynamic binding.
Many variables in TXR Lisp's standard library are global lexicals. Those which are special variables obey the "earmuffs" convention in their naming. For instance s-ifmt, log-emerg and sig-hup are global lexicals, because they provide constant values for which overriding doesn't make sense. On the other hand the standard output stream variable *stdout* is special. Overriding it over a dynamic scope is useful, as a means of redirecting the output of functions which write to the *stdout* stream.
In Common Lisp, defparm is known as defparameter.
The TXR Lisp feature known as syntactic places allows programs to use the syntax of a form which is used to access a value from an environment or object, as an expression which denotes a place where a value may be stored.
They are almost exactly the same concept as "generalized references" in Common Lisp, and are related to "lvalues" in languages in the C family, or "designators" in Pascal.
A symbol is a is a syntactic place if it names a variable. If a is a variable, then it may be assigned using the set operator: the form (set a 42) causes a to have the integer value 42.
A compound expression can be a syntactic place, if its leftmost constituent is as symbol which is specially registered, and if the form has the correct syntax for that kind of place, and suitable semantics. Such an expression is a compound place.
An example of a compound place is a car form. If c is an expression denoting a cons cell, then (car c) is not only an expression which retrieves the value of the car field of the cell. It is also a syntactic place which denotes that field as a storage location. Consequently, the expression (set (car c) "abc") stores the character string "abc" in that location. Although the same effect can be obtained with (rplaca c "abc") the syntactic place frees the programmer from having to remember different update functions for different kinds of places. There are various other advantages. TXR Lisp provides a plethora of operators for modifying a place in addition to set. Subject to certain usage restrictions, these operators work uniformly on all places. For instance, the expression (rotate (car x) [str 3] y) causes three different kinds of places to exchange contents, while the three expressions denoting those places are evaluated only once. New kinds of place update macros like rotate are quite easily defined, as are new kinds of compound places.
When a function call form such as the above (car x) is a syntactic place, then the function is called an accessor. This term is used throughout this document to denote functions which have associated syntactic places.
Syntactic places can be macros (global and lexical), including symbol macros. So for instance in (set x 42) the x place can actually be a symbolic macro which expands to, say, (cdr y). This means that the assignment is effectively (set (cdr y) 42).
Syntactic places, as well as operators upon syntactic places, are both open-ended. Code can be written quite easily in TXR Lisp to introduce new kinds of places, as well as new place-mutating operators. New places can be introduced with the help of the defplace, define-accessor or defset macros, or possibly the define-place-macro macro in simple cases when a new syntactic place can be expressed as a transformation to the syntax of an existing place. Three ways exist for developing new place update macros (place operators). They can be written using the ordinary macro definer ordinary macro definer defmacro, with the help of special utility macros called with-update-expander, with-clobber-expander, and with-delete-expander. They can also be written using defmacro in conjunction with the operators placelet or placelet*. Simple update macros similar to inc and push can be written compactly using define-modify-macro.
Unlike generalized references in Common Lisp, TXR Lisp syntactic places support the concept of deletion. Some kinds of places can be deleted, which is an action distinct from (but does not preclude) being overwritten with a value. What exactly it means for a place to be deleted, or whether that is even permitted, depends on the kind of place. For instance a place which denotes a lexical variable may not be deleted, whereas a global variable may be. A place which denotes a hash-table entry may be deleted, and results in the entry being removed from the hash table. Deleting a place in a list causes the trailing items, if any, or else the terminating atom, to move in to close the gap. Users may define new kinds of places which support deletion semantics.
To bring about their effect, place operators must evaluate one or more places. Moreover, some of them evaluate additional forms which are not places. Which arguments of a place operator form are places and which are ordinary forms depends on its specific syntax. For all the built-in place operators, the position of an argument in the syntax determines whether it is treated as (and consequently required to be) a syntactic place, or whether it is an ordinary form.
All built-in place operators perform the evaluation of place and non-place argument forms in strict left-to-right order.
Place forms are evaluated not in order to compute a value, but in order to determine the storage location. In addition to determining a storage location, the evaluation of a place form may possibly give rise to side effects. Once a place is fully evaluated, the storage location can then be accessed. Access to the storage location is not considered part of the evaluation of a place. To determine a storage location means to compute some hidden referential object which provides subsequent access to that location without the need for a reevaluation of the original place form. (The subsequent access to the place through this referential object may still require a multi-step traversal of a data structure; minimizing such steps is a matter of optimization.)
Place forms may themselves be compounds, which contain subexpressions that must be evaluated. All such evaluation for the built-in places takes place in left to right order.
Certain place operators, such as shift and rotate, exhibit an unspecified behavior with regard to the timing of the access of the prior value of a place, relative to the evaluation of places which occur later in the same place operator form. Access to the prior values may be delayed until the entire form is evaluated, or it may be interleaved into the evaluation of the form. For example, in the form (shift a b c 1), the prior value of a can be accessed and saved as soon as a is evaluated, prior to the evaluation of b. Alternatively, a may be accessed and saved later, after the evaluation of b or after the evaluation of all the forms. This issue affects the behavior of place-modifying forms whose subforms contain side effects. It is recommended that such forms not be used in programs.
Certain place forms are required to have one or more arguments which are themselves places. The prime example of this, and the only example from among built-in syntactic places, are DWIM forms. A DWIM form has the syntax
(dwim obj-place index [alt])
and the square-bracket-notation equivalent:
[obj-place index [alt]]
Note that not only is the entire form a place, denoting some element or element range of obj-place, but there is the added constraint that obj-place must also itself be a syntactic place.
This requirement is necessary, because it supports the behavior that when the element or element range is updated, then obj-place is also potentially updated.
After the assignment (set [obj 0..3] '("forty" "two")) not only is the range of places denoted by [obj 0..3] replaced by the list of strings ("forty" "two") but obj may also be overwritten with a new value.
This behavior is necessary because the DWIM brackets notation maintains the illusion of an encapsulated array-like container over several dissimilar types, including Lisp lists. But Lisp lists do not behave as fully encapsulated containers. Some mutations on Lisp lists return new objects, which then have to stored (or otherwise accepted) in place of the original objects in order to maintain the array-like container illusion.
The following is a summary of the built-in place forms, in addition to symbolic places denoting variables. New syntactic place forms can be defined by TXR programs.
(car object)
(first object)
(rest object)
(second object)
(third object)
...
(tenth object)
(last object [num])
(butlast object [num])
(cdr object)
(caar object)
(cadr object)
(cdar object)
(cddr object)
...
(cdddddr object)
(nthcdr index obj)
(nthlast index obj)
(butlastn num obj)
(nth index obj)
(ref seq idx)
(sub sequence [from [to]])
(vecref vec idx)
(chr-str str idx)
(gethash hash key [alt])
(hash-userdata hash)
(dwim obj-place index [alt])
(dwim integer obj-place ) ;; integers are callable
(dwim range obj-place ) ;; ranges are callable
(sub-list obj [from [to]])
(sub-vec obj [from [to]])
(sub-str str [from [to]])
[obj-place index [alt]] ;; equivalent to dwim
[integer obj-place ]
[range obj-place ]
(symbol-value symbol-valued-form)
(symbol-function function-name-valued-form)
(symbol-macro symbol-valued-form)
(fun function-name)
(force promise)
(errno)
(slot struct-obj slot-name-valued-form)
(qref struct-obj slot-name) ;; by macro-expansion to (slot ...)
struct-obj.slot-name ;; equivalent to qref
(sock-peer socket)
(sock-opt socket level option [ffi-type])
(carray-sub carray [from [to]])
(sub-buf buf [from [to]])
(left node)
(right node)
(key node)
(read-once node)
The following is a summary of the built-in place mutating macros. They are described in detail in their own sections.
TXR Lisp is a Lisp-2 dialect: it features separate namespaces for functions and variables.
In TXR Lisp, global functions and operator macros coexist, meaning that the same symbol can be defined as both a macro and a function.
There is a global namespace for functions, into which functions can be introduced with the defun macro. The global function environment can be inspected and modified using the symbol-function accessor.
There is a global namespace for macros, into which macros are introduced with the defmacro macro. The global function environment can be inspected and modified using the symbol-macro accessor.
If a name x is defined as both a function and a macro, then an expression of the form (x ...) is expanded by the macro, whereas an expression of the form [x ...] refers to the function. Moreover, the macro can produce a call to the function. The expression (fun x) will retrieve the function object.
There is a global namespace for variables also. The operators defvar and defparm introduce bindings into this namespace. These operators have the side effect of marking a symbol as a special variable, of the symbol are treated as dynamic variables, subject to rebinding. The global variable namespace together with the special dynamic rebinding is called the dynamic environment. The dynamic environment can be inspected and modified using the symbol-value accessor.
The operators defvarl and defparml introduce bindings into the global namespace without marking symbols as special variables. Such bindings are called global lexical variables.
Symbol macros may be defined over the global variable namespace using defsymacro.
Note that whereas a symbol may simultaneously have both a function and macro binding in the global namespace, a symbol may not simultaneously have a variable and symbol macro binding.
In addition to global and dynamic namespaces, TXR Lisp provides lexically scoped binding for functions, variables, macros, and symbol macros. Lexical variable binding are introduced with let, let* or various binding macros derived from these. Lexical functions are bound with flet and labels. Lexical macros are established with macrolet and lexical symbol macros with symacrolet.
Macros receive an environment parameter with which they may expand forms in their correct environment, and perform some limited introspection over that environment in order to determine the nature of bindings, or the classification of forms in those environments. This introspection is provided by lexical-var-p, lexical-fun-p, and lexical-lisp1-binding.
Lexical operator macros and lexical functions can also coexist in the following way. A lexical function shadows a global or lexical macro completely. However, the reverse is not the case. A lexical macro shadows only those uses of a function which look like macro calls. This is succinctly demonstrated by the following form:
(flet ((foo () 43))
(macrolet ((foo () 44))
(list (fun foo) (foo) [foo])))
-> (#<interpreted fun: lambda nil> 44 43)
The (fun foo) and [fun] expressions are oblivious to the macro; the macro expansion process process the symbol foo in those contexts. However the form (foo) is subject to macro-expansion and replaced with 44.
If the flet and macrolet are reversed, the behavior is different:
(macrolet ((foo () 44))
(flet ((foo () 43))
(list (fun foo) (foo) [foo])))
-> (#<interpreted fun: lambda nil> 43 43)
All three forms refer to the function, which lexically shadows the macro.
TXR Lisp expressions can be embedded in the TXR pattern language in various ways. Likewise, the pattern language can be invoked from TXR Lisp. This brings about the possibility that Lisp code attempts to access pattern variables bound in the pattern language. The TXR pattern language can also attempt to access TXR Lisp variables.
The rules are as follows, but they have undergone historic changes. See the COMPATIBILITY section, in particular notes under 138 and 121, and also 124.
A Lisp expression evaluated from the TXR pattern language executes in a null lexical environment. The current set of pattern variables captured up to that point by the pattern language are installed as dynamic variables. They shadow any Lisp global variables (whether those are defined by defvar or defvarl).
In the reverse direction, a variable reference from the TXR pattern language searches the pattern variable space first. If a variable doesn't exist there, then the lookup refers to the TXR Lisp global variable space. The pattern language doesn't see Lisp lexical variables.
When Lisp code is evaluated from the pattern language, the pattern variable bindings are not only installed as dynamic variables for the sake of their visibility from Lisp, but they are also specially stored in a dynamic environment frame. When TXR pattern code is reentered from Lisp, these bindings are picked up from the closest such environment frame, allowing the nested invocation of pattern code to continue with the bindings captured by outer pattern code.
Concisely, in any context in which a symbol has both a binding as a Lisp global variable as well as a pattern variable, that symbol refers to the pattern variable. Pattern variables are propagated through Lisp evaluation into nested invocations of the pattern language.
The pattern language can also reference Lisp variables using the @ prefix, which is a consequence of that prefix introducing an expression that is evaluated as Lisp, the name of a variable being such an expression.
The following sections list all of the special operators, macros and functions in TXR Lisp.
In these sections, syntax is indicated using these conventions:
A compound expression with a symbol as its first element, if intended to be evaluated, denotes either an operator invocation or a function call. This depends on whether the symbol names an operator or a function.
When the form is an operator invocation, the interpretation of the meaning of that form is under the complete control of that operator.
If the compound form is a function call, the remaining forms, if any, denote argument expressions to the function. They are evaluated in left-to-right order to produce the argument values, which are passed to the function. An exception is thrown if there are not enough arguments, or too many. Programs can define named functions with the defun operator
Some operators are macros. There exist predefined macros in the library, and macro operators can also be user-defined using the macro-defining operator defmacro. Operators that are not macros are called special operators.
Macro operators work as functions which are given the source code of the form. They analyze the form, and translate it to another form which is substituted in their place. This happens during a code walking phase called the expansion phase, which is applied to each top-level expression prior to evaluation. All macros occurring in a form are expanded in the expansion phase, and subsequent evaluation takes place on a structure which is devoid of macros. All that remains are the executable forms of special operators, function calls, symbols denoting either variables or themselves, and atoms such as numeric and string literals.
Special operators can also perform code transformations during the expansion phase, but that is not considered macroexpansion, but rather an adjustment of the representation of the operator into a required executable form. In effect, it is post-macro compilation phase.
Note that Lisp forms occurring in TXR pattern language are not individual top-level forms. Rather, the entire TXR query is parsed at the same time, and the macros occurring in its Lisp forms are expanded at that time.
(quote form)
The quote operator, when evaluated, suppresses the evaluation of form, and instead returns form itself as an object. For example, if form is a symbol sym, then the value of (quote sym) is sym itself. Without quote, sym would evaluate to the value held by the variable which is named sym, or else throw an error if there is no such variable. The quote operator never raises an error, if it is given exactly one argument, as required.
The notation 'obj is translated to the object (quote obj) providing a shorthand for quoting. Likewise, when an object of the form (quote obj) is printed, it appears as 'obj.
;; yields symbol a itself, not value of variable a
(quote a) -> a
;; yields three-element list (+ 2 2), not 4.
(quote (+ 2 2)) -> (+ 2 2)
Variables are associations between symbols and storage locations which hold values. These associations are called bindings.
Bindings are held in a context called an environment.
Lexical environments hold local variables, and nest according to the syntactic structure of the program. Lexical bindings are always introduced by a some form known as a binding construct, and the corresponding environment is instantiated during the evaluation of that construct. There also exist bindings outside of any binding construct, in the so-called global environment. Bindings in the global environment can be temporarily shadowed by lexically-established binding in the dynamic environment. See the Special Variables section above.
Certain special symbols cannot be used as variable names, namely the symbols t and nil, and all of the keyword symbols (symbols in the keyword package), which are denoted by a leading colon. When any of these symbols is evaluated as a form, the resulting value is that symbol itself. It is said that these special symbols are self-evaluating or self-quoting, similarly to all other atom objects such as numbers or strings.
When a form consisting of a symbol, other than the above special symbols, is evaluated, it is treated as a variable, and yields the value of the variable's storage location. If the variable doesn't exist, an exception is thrown.
Note: symbol forms may also denote invocations of symbol macros. (See the operators defsymacro and symacrolet). All macros, including symbol macros, which occur inside a form are fully expanded prior to the evaluation of a form, therefore evaluation does not consider the possibility of a symbol being a symbol macro.
(defvar sym [value])
(defparm sym value)
The defvar operator binds a name in the variable namespace of the global environment. Binding a name means creating a binding: recording, in some namespace of some environment, an association between a name and some named entity. In the case of a variable binding, that entity is a storage location for a value. The value of a variable is that which has most recently been written into the storage location, and is also said to be a value of the binding, or stored in the binding.
If the variable named sym already exists in the global environment, the form has no effect; the value form is not evaluated, and the value of the variable is unchanged.
If the variable does not exist, then a new binding is introduced, with a value given by evaluating the value form. If the form is absent, the variable is initialized to nil.
The value form is evaluated in the environment in which the defvar form occurs, not necessarily in the global environment.
The symbols t and nil may not be used as variables, nor can they be keyword symbols (symbols denoted by a leading colon).
In addition to creating a binding, the defvar operator also marks sym as the name of a special variable. This changes what it means to bind that symbol in a lexical binding construct such as the let operator, or a function parameter list. See the section "Special Variables" far above.
The defparm macro behaves like defvar when a variable named sym doesn't already exist.
If sym already denotes a variable binding in the global namespace, defparm evaluates the value form and assigns the resulting value to the variable.
The following equivalence holds:
(defparm x y) <--> (prog1 (defvar x) (set x y))
The defvar and defparm forms return sym.
(defvarl sym [value])
(defparml sym value)
The defvarl and defparml macros behave, respectively, almost exactly like defvar and defparm.
The difference is that these operators do not mark sym as special.
If a global variable sym does not previously exist, then after the evaluation of either of these forms (boundp sym) is true, but (special-var-p sym) isn't.
If sym had been already introduced as a special variable, it stays that way after the evaluation of defvarl or defparml.
(let ({sym | (sym init-form)}*) body-form*)
(let* ({sym | (sym init-form)}*) body-form*)
The let and let* operators introduce a new scope with variables and evaluate forms in that scope. The operator symbol, either let or let*, is followed by a list which can contain any mixture of sym or (sym init-form) pairs. Each sym must be a symbol, and specifies the name of variable to be instantiated and initialized.
The (sym init-form) variant specifies that the new variable sym receives an initial value from the evaluation of init-form. The plain sym variant specifies a variable which is initialized to nil. The init-forms are evaluated in order, by both let and let*.
The symbols t and nil may not be used as variables, and neither can be keyword symbols: symbols denoted by a leading colon.
The difference between let and let* is that in let*, later init-forms are in scope of the variables established by earlier variables in the same let* construct. In plain let, the init-forms are evaluated in a scope which does not include any of the variables.
When the variables are established, the body-forms are evaluated in order. The value of the last body-form becomes the return value of the let. If there are no body-forms, then the return value nil is produced.
The list of variables may be empty.
The list of variables may contain duplicate syms if the operator is let*. In that situation, a given init-form has in scope the rightmost duplicate of any given sym that has been previously established. The body-forms have in scope the rightmost duplicate of any sym in the construct. Therefore, the following form calculates the value 3:
(let* ((a 1)
(a (succ a))
(a (succ a)))
a)
Each duplicate is a separately instantiated binding, and may be independently captured by a lexical closure placed in a subsequent init-form:
(let* ((a 0)
(f1 (lambda () (inc a)))
(a 0)
(f2 (lambda () (inc a))))
(list [f1] [f1] [f1] [f2] [f2] [f2]))
--> (1 2 3 1 2 3)
The preceding example shows that there are two mutable variables named a in independent scopes, each respectively captured by the separate closures f1 and f2. Three calls to f1 increment the first a while the second a retains its initial value.
Under let, the behavior of duplicate variables is unspecified.
Implementation note: the TXR compiler diagnoses and rejects duplicate symbols in let whereas the interpreter ignores the situation.
When the names of a special variables is specified in let or let* remain, a new binding is created for them in the dynamic environment, rather than the lexical environment. In let*, later init-forms are evaluated in a dynamic scope in which previous dynamic variables are established, and later dynamic variables are not yet established. A special variable may appear multiple times in a let*, just like a lexical variable. Each duplicate occurrence extends the dynamic environment with a new dynamic binding. All these dynamic environments are removed when the let or let* form terminates. Dynamic environments aren't captured by lexical closures, but are captured in delimited continuations.
(let ((a 1) (b 2)) (list a b)) -> (1 2)
(let* ((a 1) (b (+ a 1))) (list a b (+ a b))) -> (1 2 3)
(let ()) -> nil
(let (:a nil)) -> error, :a and nil can't be used as variables
TXR Lisp follows ANSI Common Lisp in making let the parallel binding construct, and let* the sequential one. In that language, the situation exists for historic reasons: mainly that let was initially understood as being a macro for an immediately-called lambda where the parameters come into existence simultaneously, receiving the evaluated values of all the argument expressions. The need for sequential binding was recognized later, by which time let was cemented as a parallel binding construct. There are very good arguments for, in a new design, using the let name for the construct which has sequential semantics. Nevertheless, in this matter, TXR Lisp remains compatible with dialects like ANSI CL and Emacs Lisp.
(progv symbols-expr values-expr body-form*)
The progv operator binds dynamic variables, and evaluates the body-forms in the dynamic scope of those bindings. The bindings are removed when the form terminates. The result value is that of the last body-form or else nil if there are no forms.
The symbols-expr and values-expr are expressions which are evaluated. Their values are expected to be lists, of bindable symbols and arbitrary values, respectively. The symbols coming from one list are bound to the values coming from the other list.
If there are more symbols than values, then the extra symbols will appear unbound, as if they were first bound and then hidden using the makunbound function.
If there are more values than symbols, the extra values are ignored.
Note that dynamic binding takes place for the symbols even if they have not been introduced as special variables via defvar or defparm. However, if those symbols appear as expressions denoting variables inside the body-forms, they will not necessarily be treated as dynamic variables. If they have lexical definitions in scope, those will be referenced. Furthermore, the compiler treats undefined variables as global references, and not dynamic.
(progv '(a b) '(1 2) (cons a b)) -> (1 . 2)
(progv '(x) '(1) (let ((x 4)) (symbol-value 'x))) -> 1
(let ((x 'lexical)
(vars (list 'x))
(vals (list 'dynamic)))
(progv vars vals (list x (symbol-value 'x))))
--> (lexical dynamic)
(defun name (param* [: opt-param*] [. rest-param])
body-form)
The defun operator introduces a new function in the global function namespace. The function is similar to a lambda, and has the same parameter syntax and semantics as the lambda operator.
Note that the above syntax synopsis describes only the canonical parameter syntax which remains after parameter list macros are expanded. See the section Parameter List Macros.
Unlike in lambda, the body-forms of a defun are surrounded by a block. The name of this block is the same as the name of the function, making it possible to terminate the function and return a value using (return-from name value). For more information, see the definition of the block operator.
A function may call itself by name, allowing for recursion.
The special symbols t and nil may not be used as function names. Neither can keyword symbols.
It is possible to define methods as well as macros with defun, as an alternative to the defmeth and defmacro forms.
To define a method, the syntax (meth type name) should be used as the argument to the name parameter. This gives rise to the syntax (defun (meth type name) args form*) which is equivalent to the (defmeth type name args form*) syntax.
Macros can be defined using (macro name) as the name parameter of defun. This way of defining a macro doesn't support destructuring; it defines the expander as an ordinary function with an ordinary argument list. To work, the function must accept two arguments: the entire macro call form that is to be expanded, and the macro environment. Thus, the macro definition syntax is (defun (macro name) form env form*) which is equivalent to the (defmacro name (:form form :env env) form*) syntax.
In ANSI Common Lisp, keywords may be used as function names. In TXR Lisp, they may not.
A function defined by defun may coexist with a macro defined by defmacro. This is not permitted in ANSI Common Lisp.
(lambda (param* [: opt-param*] [. rest-param])
body-form)
(lambda rest-param
body-form)
The lambda operator produces a value which is a function. Like in most other Lisps, functions are objects in TXR Lisp. They can be passed to functions as arguments, returned from functions, aggregated into lists, stored in variables, etc.
Note that the above syntax synopsis describes only the canonical parameter syntax which remains after parameter list macros are expanded. See the section Parameter List Macros.
The first argument of lambda is the list of parameters for the function. It may be empty, and it may also be an improper list (dot notation) where the terminating atom is a symbol other than nil. It can also be a single symbol.
The second and subsequent arguments are the forms making up the function body. The body may be empty.
When a function is called, the parameters are instantiated as variables that are visible to the body forms. The variables are initialized from the values of the argument expressions appearing in the function call.
The dotted notation can be used to write a function that accepts a variable number of arguments. There are two ways write a function that accepts only a variable argument list and no required arguments:
(lambda (. rest-param) ...)
(lambda rest-param ...)
(These notations are syntactically equivalent because the list notation (. X) actually denotes the object X which isn't wrapped in any list).
The keyword symbol : (colon) can appear in the parameter list. This is the symbol in the keyword package whose name is the empty string. This symbol is treated specially: it serves as a separator between required parameters and optional parameters. Furthermore, the : symbol has a role to play in function calls: it can be specified as an argument value to an optional parameter by which the caller indicates that the optional argument is not being specified. It will be processed exactly that way.
An optional parameter can also be written in the form (name expr [sym]). In this situation, if the call does not specify a value for the parameter, or specifies a value as the : (colon) keyword symbol, then the parameter takes on the value of the expression expr. This expression is only evaluated when its value is required.
If sym is specified, then sym will be introduced as an additional binding with a Boolean value which indicates whether or not the optional parameter had been specified by the caller.
Each expr that is evaluated is evaluated in an environment in which all of the previous parameters are visible, in addition to the surrounding environment of the lambda. For instance:
(let ((default 0))
(lambda (str : (end (length str)) (counter default))
(list str end counter)))
In this lambda, the initializing expression for the optional parameter end is (length str), and the str variable it refers to is the previous argument. The initializer for the optional variable counter is the expression default, and it refers to the binding established by the surrounding let. This reference is captured as part of the lambda's lexical closure.
Keyword symbols, and the symbols t and nil may not be used as parameter names. The behavior is unspecified if the same symbol is specified more than once anywhere in the parameter list, whether as a parameter name or as the indicator sym in an optional parameter or any combination.
Implementation note: the TXR compiler diagnoses and rejects duplicate symbols in lambda whereas the interpreter ignores the situation.
Note: it is not always necessary to use the lambda operator directly in order to produce an anonymous function.
In situations when lambda is being written in order to simulate partial evaluation, it may be possible to instead make use of the op macro. For instance the function (lambda (. args) [apply + a args]) which adds the values of all of its arguments together, and to the lexically captured variable a can be written more succinctly as (op + a). The op operator is the main representative of a family of operators: lop, ap, ip, do, ado, opip and oand.
In situations when functions are simply combined together, the effect may be achieved using some of the available functional combinators, instead of a lambda. For instance chaining together functions as in (lambda (x) (square (cos x))) is achievable using the chain function: [chain cos square]. The opip operator can also be used: (opip cos square). Numerous combinators are available; see the section Partial Evaluation and Combinators.
When a function is needed which accesses an object, there are also alternatives. Instead of (lambda (obj) obj.slot) and (lambda (obj arg) obj.(slot arg)), it is simpler to use the .slot and .(slot arg) notations. See the section Unbound Referencing Dot. Also see the functions umethod and uslot as well as the related convenience macros umeth and usl.
If a function is needed which partially applies, to some arguments, a method invoked on a specific object, the method function or meth macro may be used. For instance, instead of (lambda (arg) obj.(method 3 arg)), it is possible to write (meth obj 3) except that the latter produces a variadic function.
The following expression returns a function which captures the variable counter. Whenever the returned function is called, it increments counter by one, and returns the incremented value.
(let ((counter 0))
(lambda () (inc counter)))
The following produces a variadic function which requires at least two arguments. The third and subsequent arguments are aggregated into a list passed as the single parameter z:
(lambda (x y . z) (list 'my-arguments-are x y z))
A variadic function with no required arguments. The parameter name for the received arguments appears alone in place of the parameter list.
(lambda args (list 'my-list-of-arguments args))
Same as the previous example, using a dotted notation specific to TXR Lisp.
(lambda (. args) (list 'my-list-of-arguments args))
Note that (. args) is just a written notation equivalent to args and not a different object structure.
Optional arguments:
[(lambda (x : y) (list x y)) 1] -> (1 nil)
[(lambda (x : y) (list x y)) 1 2] -> (1 2)
Passing : (colon symbol) to request default value of optional parameter:
[(lambda (x : (y 42) z) (list x y z)) 1 2 3] -> (1 2 3)
[(lambda (x : (y 42) z) (list x y z)) 1 : 3] -> (1 42 3)
[(lambda (x : (y 42) z) (list x y z)) 1] -> (1 42 nil)
Presence-indicating variable accompanying optional parameter:
[(lambda (x : (y 42 have-y)) (list x y have-y)) 1 2]
-> (1 2 t)
[(lambda (x : (y 42 have-y)) (list x y have-y)) 1]
-> (1 42 nil)
;; defaulting via : is indistinguishable from missing
[(lambda (x : (y 42 have-y)) (list x y have-y)) 1 :]
-> (1 42 nil)
(flet ({(name param-list function-body-form*)}*)
body-form*)
(labels ({(name param-list function-body-form*)}*)
body-form*)
The flet and labels macros bind local, named functions in the lexical scope.
Note that the above syntax synopsis describes only the canonical parameter syntax which remains after parameter list macros are expanded. See the section Parameter List Macros.
The difference between flet and labels is that a function defined by labels can see itself, and therefore recurse directly by name. Moreover, if multiple functions are defined by the same labels construct, they all have each other's names in scope of their bodies. By contrast, a flet-defined function does not have itself in scope and cannot recurse. Multiple functions in the same flet do not have each other's names in their scopes.
More formally, the function-body-forms and param-list of the functions defined by labels are in a scope in which all of the function names being defined by that same labels construct are visible.
Under both labels and flet, the local functions that are defined are lexically visible to the main body-forms.
Note that labels and flet are properly scoped with regard to macros. During macro expansion, function bindings introduced by these macro operators shadow macros defined by macrolet and defmacro.
Furthermore, function bindings introduced by labels and flet also shadow symbol macros defined by symacrolet, when those symbol macros occur as arguments of a dwim form.
See also: the macrolet operator.
The flet and labels macros do not establish named blocks around the body forms of the local functions which they bind. This differs from ANSI Common Lisp, whose local function have implicit named blocks, allowing for return-from to be used.
;; Wastefully slow algorithm for determining evenness.
;; Note:
;; - mutual recursion between labels-defined functions
;; - inner is-even bound by labels shadows the outer
;; one bound by defun so the (is-even n) call goes
;; to the local function.
(defun is-even (n)
(labels ((is-even (n)
(if (zerop n) t (is-odd (- n 1))))
(is-odd (n)
(if (zerop n) nil (is-even (- n 1)))))
(is-even n)))
(call function argument*)
The call function invokes function, passing it the given arguments, if any.
function need not be a function; other kinds of objects can be used in place of functions with various semantics. The details are given in the description of the dwim operator.
Apply lambda to 1 2 arguments, adding them to produce 3:
(call (lambda (a b) (+ a b)) 1 2)
Useless use of call on a named function; equivalent to (list 1 2):
(apply function [arg* trailing-args])
(iapply function [arg* trailing-args])
The apply function invokes function, optionally passing to it an argument list. The return value of the apply call is that of function.
If no arguments are present after function, then function is invoked without arguments.
If one argument is present after function, then it is interpreted as trailing-args. If this is a sequence (a list, vector or string), then the elements of the sequence are passed as individual arguments to function. If trailing-args is not a sequence, then function is invoked with an improper argument list, terminated by the trailing-args atom.
If two or more arguments are present after function, then the last of these arguments is interpreted as trailing-args. The previous arguments represent leading arguments. When the argument list is formed to which function is applied, the leading arguments become individual arguments presented in the same order, followed by arguments taken from the trailing_args list.
Note that if trailing-args value is an atom or an improper list, the function is then invoked with an improper argument list. Only a variadic function may be invoked with an improper argument list. Moreover, all of the function's required and optional parameters must be satisfied by elements of the improper list, such that the terminating atom either matches the rest-param directly (see the lambda operator) or else the rest-param receives an improper list terminated by that atom. To treat the terminating atom of an improper list as an ordinary element which can satisfy a required or optional function parameter, the iapply function may be used, described next.
The iapply function ("improper apply") is similar to apply, except with regard to the treatment of trailing-args. Firstly, under iapply, if trailing-args is an atom other than nil (possibly a sequence, such as a vector or string), then it is treated as an ordinary argument: function is invoked with a proper argument list, whose last element is trailing-args. Secondly, if trailing-args is a list, but an improper list, then the terminating atom of trailing-args becomes an individual argument. This terminating atom is not split into multiple arguments, even if it is a sequence. Thus, in all possible cases, iapply treats an extra non-nil atom as an argument, and never calls function with an improper argument list.
;; '(1 2 3) becomes arguments to list, thus (list 1 2 3).
(apply (fun list) '(1 2 3)) -> (1 2 3)
;; this effectively invokes (list 1 2 3 4)
(apply (fun list) 1 2 '(3 4)) -> (1 2 3 4)
;; this effectively invokes (list 1 2 . 3)
(apply (fun list) 1 2 3)) -> (1 2 . 3)
;; "abc" is separated into characters
;; which become arguments of list
(apply (fun list) "abc") -> (#\a #\b #\c)
Note that some uses of this function that are necessary in other Lisp dialects are not necessary in TXR Lisp. The reason is that in TXR Lisp, improper list syntax is accepted as a compound form, and performs application:
(foo a b . x)
Here, the variables a and b supply the first two arguments for foo. In the dotted position, x must evaluate to a list or vector. The list or vector's elements are pulled out and treated as additional arguments for foo. This syntax can only be used if x is a symbolic form or an atom. It cannot be a compound form, because (foo a b . (x)) and (foo a b x) are equivalent structures.
(fun function-name)
The fun operator retrieves the function object corresponding to a named function in the current lexical environment.
The function-name may be a symbol denoting a named function: a built in function, or one defined by defun.
The function-name may also take any of the forms specified in the description of the func-get-name function. If such a function-name refers to a function which exists, then the fun operator yields that function.
Note: the fun operator does not see macro bindings via their symbolic names with which they are defined by defmacro. However, the name syntax (macro name) may be used to refer to macros. This syntax is documented in the description of func-get-name. It is also possible to retrieve a global macro expander using the function symbol-macro.
(dwim argument*)
(set (dwim obj-place index [alt]) new-value)
(set (dwim {integer | range} obj-place) new-value)
'['argument*']'
(set '['obj-place index [alt]']' new-value)
(set '[{'integer | range} obj-place']' new-value)
The dwim operator's name is an acronym: DWIM may be taken to mean "Do What I Mean", or alternatively, "Dispatch, in a Way that is Intelligent and Meaningful".
The notation [...] is a shorthand which denotes (dwim ...).
Note that since the [ and ] are used in this document for indicating optional syntax, in the above Syntax synopsis the quoted notation '[' and ']' denotes bracket tokens which literally appear in the syntax.
The dwim operator takes a variable number of arguments, which are treated as expressions to be individually macro-expanded and evaluated, using the same rules.
This means that the first argument isn't a function name, but an ordinary expression which can simply compute a function object (or, more generally, a callable object).
Furthermore, for those arguments of dwim which are symbols (after all macro-expansion is performed), the evaluation rules are altered. For the purposes of resolving symbols to values, the function and variable binding namespaces are considered to be merged into a single space, creating a situation that is similar to a Lisp-1 style dialect.
This special Lisp-1 evaluation is not recursively applied. All arguments of dwim which, after macro expansion, are not symbols are evaluated using the normal Lisp-2 evaluation rules. Thus, the DWIM operator must be used in every expression where the Lisp-1 rules for reducing symbols to values are desired.
If a symbol has bindings both in the variable and function namespace in scope, and is referenced by a dwim argument, this constitutes a conflict which is resolved according to two rules. When nested scopes are concerned, then an inner binding shadows an outer binding, regardless of their kind. An inner variable binding for a symbol shadows an outer or global function binding, and vice versa.
If a symbol is bound to both a function and variable in the global namespace, then the variable binding is favored.
Macros do not participate in the special scope conflation, with one exception. What this means is that the space of symbol macros is not folded together with the space of operator macros. An argument of dwim that is a symbol might be symbol macro, variable or function, but it cannot be interpreted as the name of a operator macro.
The exception is this: from the perspective of a dwim form, function bindings can shadow symbol macros. If a function binding is defined in an inner scope relative to a symbol macro for the same symbol, using flet or labels, the function hides the symbol macro. In other words, when macro expansion processes an argument of a dwim form, and that argument is a symbol, it is treated specially in order to provide a consistent name lookup behavior. If the innermost binding for that symbol is a function binding, it refers to that function binding, even if a more outer symbol macro binding exists, and so the symbol is not expanded using the symbol macro. By contrast, in an ordinary form, a symbolic argument never resolves to a function binding. The symbol refers to either a symbol macro or a variable, whichever is nested closer.
If, after macro expansion, the leftmost argument of the dwim is the name of a special operator or macro, the dwim form doesn't denote an invocation of that operator or macro. A dwim form is an invocation of the dwim operator, and the leftmost argument of that operator, if it is a symbol, is treated as a binding to be resolved in the variable or function namespace, like any other argument. Thus [if x y] is an invocation of the if function, not the if operator.
How many arguments are required by the dwim operator depends on the type of object to which the first argument expression evaluates. The possibilities are:
This form is also a syntactic place. If a value is stored to this place, it replaces the element.
The place may also be deleted, which has the effect of removing the element from the sequence, shifting the elements at higher indices, if any, down one element position, and shortening the sequence by one. If the place is deleted, and if sequence is a list, then the sequence form itself must be a place.
This form is implemented using the ref accessor such that, except for the argument evaluation semantics of the DWIM brackets, it is equivalent to using the (ref sequence index) syntax.
This form is also a syntactic place. Storing a value in this place has the effect of replacing the subsequence with a new subsequence. Deleting the place has the effect of removing the specified subsequence from sequence. If sequence is a list, then the sequence form must itself be a place. The new-value argument in a range assignment can be a string, vector or list, regardless of whether the target is a string, vector or list. If the target is a string, the replacement sequence must be a string, or a list or vector of characters.
The semantics is implemented using the sub accessor, such that the following equivalence holds:
[seq from..to] <--> (sub seq from..to)
For this reason, sequence may be any object that is iterable by iter-begin.
This form is equivalent to (select sequence where-index) except when the target of an assignment operation.
This form is a syntactic place if sequence is one. If a sequence is assigned to this place, then elements of the sequence are distributed to the specified locations.
The following equivalences hold between index-sequence-based indexing and the select and replace functions, except that set always returns the value assigned, whereas replace returns its first argument:
[seq idx-seq] <--> (select seq idx-seq)
(set [seq idx-seq] new) <--> (replace seq new idx-seq)
Note that unlike the select function, this does not support [hash index-seq] because since hash keys may be lists, that syntax is indistinguishable from a simple hash lookup where index-seq is the key.
If start is specified, it gives the starting position where the search begins, and if from-end is given, and has a value other than nil, it specifies a search from right to left. These optional arguments have the same conventions and semantics as their equivalents in the search-regst function.
Note that string is always required, and is always the rightmost argument.
Note that the various above forms are not actually cases of the dwim operator but the due to the semantics of the left argument objects being used as functions. All of the semantics described above is available in any situation in which an object is used as a function: for instance, as an argument of the call or apply operators, or the functional argument in mapcar.
Vector and list range indexing is based from zero, meaning that the first element is numbered zero, the second one and so on. zero. Negative values are allowed; the value -1 refers to the last element of the vector or list, and -2 to the second last and so forth. Thus the range 1 .. -2 means "everything except for the first element and the last two".
The symbol t represents the position one past the end of the vector, string or list, so 0..t denotes the entire list or vector, and the range t..t represents the empty range just beyond the last element. It is possible to assign to t..t. For instance:
(defvar list '(1 2 3))
(set [list t..t] '(4)) ;; list is now (1 2 3 4)
The value zero has a "floating" behavior when used as the end of a range. If the start of the range is a negative value, and the end of the range is zero, the zero is interpreted as being the position past the end of the sequence, rather than the first element. For instance the range -1..0 means the same thing as -1..t. Zero at the start of a range always means the first element, so that 0..-1 refers to all the elements except for the last one.
The dwim operator allows for a Lisp-1 flavor of programming in TXR Lisp, which is principally a Lisp-2 dialect.
A Lisp-1 dialect is one in which an expression like (a b) treats both a and b as expressions subject to the same evaluation rules—at least, when a isn't an operator or an operator macro. This means that the symbols a and b are resolved to values in the same namespace. The form denotes a function call if the value of variable a is a function object. Thus in a Lisp-1, named functions do not exist as such: they are just variable bindings. In a Lisp-1, the form (car 1) means that there is a variable called car, which holds a function, which is retrieved from that variable and applied to the 1 argument. In the expression (car car), both occurrences of car refer to the variable, and so this form applies the car function to itself. It is almost certainly meaningless. In a Lisp-2 (car 1) means that there is a function called car, in the function namespace. In the expression (car car) the two occurrences refer to different bindings: one is a function and the other a variable. Thus there can exist a variable car which holds a cons-cell object, rather than the car function, and the form makes sense.
The Lisp-1 approach is useful for functional programming, because it eliminates cluttering occurrences of the call and fun operators. For instance:
;; regular notation
(call foo (fun second) '((1 a) (2 b)))
;; [] notation
[foo second '((1 a) (2 b))]
Lisp-1 dialects can also provide useful extensions by giving a meaning to objects other than functions in the first position of a form, and the dwim/[...] syntax does exactly this.
TXR Lisp is a Lisp-2 because Lisp-2 also has advantages. Lisp-2 programs which use macros naturally achieve hygiene because lexical variables do not interfere with the function namespace. If a Lisp-2 program has a local variable called list, this does not interfere with the hidden use of the function list in a macro expansion in the same block of code. Lisp-1 dialects have to provide hygienic macro systems to attack this problem. Furthermore, even when not using macros, Lisp-1 programmers have to avoid using the names of functions as lexical variable names, if the enclosing code might use them.
The two namespaces of a Lisp-2 also naturally accommodate symbol macros and operator macros. Whereas functions and variables can be represented in a single namespace readily, because functions are data objects, this is not so with symbol macros and operator macros, the latter of which are distinguished syntactically by their position in a form. In a Lisp-1 dialect, given (foo bar), either of the two symbols could be a symbol macro, but only foo can possibly be an operator macro. Yet, having only a single namespace, a Lisp-1 doesn't permit (foo foo), where foo is simultaneously a symbol macro and an operator macro, though the situation is unambiguous by syntax even in Lisp-1. In other words, Lisp-1 dialects do not entirely remove the special syntactic recognition given to the leftmost position of a compound form, yet at the same time they prohibit the user from taking full advantage of it by providing only one namespace.
TXR Lisp provides the "best of both worlds": the DWIM brackets notation provides a model of Lisp-1 computation that is purer than Lisp-1 dialects (since the leftmost argument is not given any special syntactic treatment at all) while the Lisp-2 foundation provides a traditional Lisp environment with its "natural hygiene".
(functionp obj)
The functionp function returns t if obj is a function, otherwise it returns nil.
(copy-fun function)
The copy-fun function produces and returns a duplicate of function, which must be a function.
A duplicate of a function is a distinct function object not eq to the original function, yet which accepts the same arguments and behaves exactly the same way as the original.
If a function contains no captured environment, then a copy made of that function by copy-fun is indistinguishable from the original function in every regard, except for being a distinct object that compares unequal to the original under the eq function.
If a function contains a captured environment, then a copy of that function made by copy-fun has its own copy of that environment. If the copied function changes the values of captured lexical variables, the original function is not affected by these changes and vice versa.
The entire lexical environment is copied; the copy and original function do not share any portion of the environment at any level of nesting.
(progn form*)
(prog1 form*)
The progn operator evaluates each form in left-to-right order, and returns the value of the last form. The value of the form (progn) is nil.
The prog1 operator evaluates each form in left-to-right order, and returns the value of the first form. The value of the form (prog1) is nil.
Various other operators such as let also arrange for the evaluation of a body of forms, the value of the last of which is returned. These operators are said to feature an implicit progn.
These special operators are also functions. The progn function accepts zero or more arguments. It returns its last argument, or nil if called with no arguments. The prog1 function likewise accepts zero or more arguments. It returns its first argument, or nil if called with no arguments.
In ANSI Common Lisp, prog1 requires at least one argument. Neither prog nor prog1 exist as functions.
(prog2 form*)
The prog2 evaluates each form in left-to-right order. The value is that of the second form, if present, otherwise it is nil.
The form (prog2 1 2 3) yields 2. The value of (prog2 1 2) is also 2; (prog2 1) and (prog2) yield nil.
The prog2 symbol also has a function binding. The prog2 function accepts any number of arguments. If invoked with at least two arguments, it returns the second one. Otherwise it returns nil.
In ANSI Common Lisp, prog2 requires at least two arguments. It does not exist as a function.
(cond {(test form*)}*)
The cond operator provides a multi-branching conditional evaluation of forms. Enclosed in the cond form are groups of forms expressed as lists. Each group must be a list of at least one form.
The forms are processed from left to right as follows: the first form, test, in each group is evaluated. If it evaluates true, then the remaining forms in that group, if any, are also evaluated. Processing then terminates and the result of the last form in the group is taken as the result of cond. If test is the only form in the group, then result of test is taken as the result of cond.
If the first form of a group yields nil, then processing continues with the next group, if any. If all form groups yield nil, then the cond form yields nil. This holds in the case that the syntax is empty: (cond) yields nil.
(caseq test-form normal-clause* [else-clause])
(caseql test-form normal-clause* [else-clause])
(casequal test-form normal-clause* [else-clause])
These three macros arrange for the evaluation of test-form, whose value is then compared against the key or keys in each normal-clause. When the value matches a key, then the remaining forms of normal-clause are evaluated, and the value of the last form is returned; subsequent clauses are not evaluated.
If no normal-clause matches, and there is no else-clause, then the value nil is returned. Otherwise, the forms in the else-clause are evaluated, and the value of the last one is returned. If there are no forms, then nil is returned.
If duplicates keys are present in such a way that the value of the test-form matches multiple normal-clauses, it is unspecified which of those clauses is evaluated.
The syntax of a normal-clause takes on these two forms:
(key form*)
where key may be an atom which denotes a single key, or else a list of keys. There is a restriction that the symbol t may not be used as key. The form (t) may be used as a key to match that symbol.
The syntax of an else-clause is:
(t form*)
which resembles a form that is often used as the final clause in the cond syntax.
The three forms of the case construct differ from what type of test they apply between the value of test-form and the keys. The caseq macro generates code which uses the eq function's equality. The caseql macro uses eql, and casequal uses equal.
(let ((command-symbol (casequal command-string
(("q" "quit") 'quit)
(("a" "add") 'add)
(("d" "del" "delete") 'delete)
(t 'unknown))))
...)
(caseq* test-form normal-clause* [else-clause])
(caseql* test-form normal-clause* [else-clause])
(casequal* test-form normal-clause* [else-clause])
The caseq*, caseql*, and casequal* macros are similar to the macros caseq, caseql, and casequal, differing from them in only the following regard. The normal-clause, of these macros has the form (evaluated-key form*) where evaluated-key is either an atom, which is evaluated to produce a key, or else else a compound form, whose elements are evaluated as forms, producing multiple keys. This evaluation takes place at macro-expansion time, in the global environment.
The else-clause works the same way under these macros as under caseq et al.
Note that although in a normal-clause, evaluated-key must not be the atom t, there is no restriction against it being an atom which evaluates to t. In this situation, the value t has no special meaning.
The evaluated-key expressions are evaluated in the order in which they appear in the construct, at the time the caseq*, caseql* or casequal* macro is expanded.
Note: these macros allow the use of variables and global symbol macros as case keys.
(defvarl red 0)
(defvarl green 1)
(defvarl blue 2)
(let ((color blue))
(caseql* color
(red "hot")
((green blue) "cool")))
--> "cool"
(ecaseq test-form normal-clause* [else-clause])
(ecaseql test-form normal-clause* [else-clause])
(ecasequal test-form normal-clause* [else-clause])
(ecaseq* test-form normal-clause* [else-clause])
(ecaseql* test-form normal-clause* [else-clause])
(ecasequal* test-form normal-clause* [else-clause])
These macros are error-catching variants of, respectively, caseq, caseql, casequal, caseq*, caseql* and casequal*.
If the else-clause is present in the invocation of an error-catching case macro, then the the invocation is precisely equivalent to the corresponding non-error-trapping variant.
If the else-clause is missing in the invocation of an error-catching variant, then a default else-clause is inserted which throws an exception of type case-error, derived from error. After this insertion, the semantics follows that of the non-error-trapping variant.
For instance, (ecaseql 3), which has no else-clause, is equivalent to (caseql 3 (t expr)) where expr indicates the inserted expression which throws case-error. However, (ecaseql 3 (t 42)) is simply equivalent to (caseql 3 (t 42)), since it has an else-clause.
Note: the error-catching case macros are intended for situations in which it is a matter of program correctness that every possible value of test-form matches a normal-clause, such that if a failure to match occurs, it indicates a software defect. The error-throwing else-clause helps to ensure that the error situation is noticed. Without this clause, the case macro terminates with a value of nil, which may conceal the defect and delay its identification.
(if cond t-form [e-form])
'['if cond then [else]']'
There exist both an if operator and an if function. A list form with the symbol if in the first position is interpreted as an invocation of the if operator. The function can be accessed using the DWIM bracket notation and in other ways.
The if operator provides a simple two-way-selective evaluation control. The cond form is evaluated. If it yields true then t-form is evaluated, and that form's return value becomes the return value of the if. If cond yields false, then e-form is evaluated and its return value is taken to be that of if. If e-form is omitted, then the behavior is as if e-form were specified as nil.
The if function provides no evaluation control. All of its arguments are evaluated from left to right. If the cond argument is true, then it returns the then argument, otherwise it returns the value of the else argument if present, otherwise it returns nil.
(and form*)
'['and arg*']'
There exist both an and operator and an and function. A list form with the symbol and in the first position is interpreted as an invocation of the operator. The function can be accessed using the DWIM bracket notation and in other ways.
The and operator provides three functionalities in one. It computes the logical "and" function over several forms. It controls evaluation (a.k.a. "short-circuiting"). It also provides an idiom for the convenient substitution of a value in place of nil when some other values are all true.
The and operator evaluates as follows. First, a return value is established and initialized to the value t. The forms, if any, are evaluated from left to right. The return value is overwritten with the result of each form. Evaluation stops when all forms are exhausted, or when nil is stored in the return value. When evaluation stops, the operator yields the return value.
The and function provides no evaluation control: it receives all of its arguments fully evaluated. If it is given no arguments, it returns t. If it is given one or more arguments, and any of them are nil, it returns nil. Otherwise, it returns the value of the last argument.
(and) -> t
(and (> 10 5) (stringp "foo")) -> t
(and 1 2 3) -> 3 ;; shorthand for (if (and 1 2) 3).
(nand form*)
'['nand arg*']'
There exist both a nand macro and a nand function. A list form with the symbol nand in the first position is interpreted as an invocation of the macro. The function can be accessed using the DWIM bracket notation and in other ways.
The nand macro and function are the logical negation of the and operator and function. They are related according to the following equivalences:
(nand f0 f1 f2 ...) <--> (not (and f0 f1 f2 ...))
[nand f0 f1 f2 ...] <--> (not [and f0 f1 f2 ...])
(or form*)
'['or arg*']'
There exist both an or operator and an or function. A list form with the symbol or in the first position is interpreted as an invocation of the operator. The function can be accessed using the DWIM bracket notation and in other ways.
The or operator provides three functionalities in one. It computes the logical "or" function over several forms. It controls evaluation (a.k.a. "short-circuiting"). The behavior of or also provides an idiom for the selection of the first non-nil value from a sequence of forms.
The or operator evaluates as follows. First, a return value is established and initialized to the value nil. The forms, if any, are evaluated from left to right. The return value is overwritten with the result of each form. Evaluation stops when all forms are exhausted, or when a true value is stored into the return value. When evaluation stops, the operator yields the return value.
The or function provides no evaluation control: it receives all of its arguments fully evaluated. If it is given no arguments, it returns nil. If all of its arguments are nil, it also returns nil. Otherwise, it returns the value of the first argument which isn't nil.
(or) -> nil
(or 1 2) -> 1
(or nil 2) -> 2
(or (> 10 20) (stringp "foo")) -> t
(nor form*)
'['nor arg*']'
There exist both a nor macro and a nor function. A list form with the symbol nor in the first position is interpreted as an invocation of the macro. The function can be accessed using the DWIM bracket notation and in other ways.
The nor macro and function are the logical negation of the or operator and function. They are related according to the following equivalences:
(nor f0 f1 f2 ...) <--> (not (or f0 f1 f2 ...))
[nor f0 f1 f2 ...] <--> (not [or f0 f1 f2 ...])
(when expression form*)
(unless expression form*)
The when macro operator evaluates expression. If expression yields true, and there are additional forms, then each form is evaluated. The value of the last form becomes the result value of the when form. If there are no forms, then the result is nil.
The unless operator is similar to when, except that it reverses the logic of the test. The forms, if any, are evaluated if and only if expression is false.
(while expression form*)
(until expression form*)
The while macro operator provides a looping construct. It evaluates expression. If expression yields nil, then the evaluation of the while form terminates, producing the value nil. Otherwise, if there are additional forms, then each form is evaluated. Next, evaluation returns to expression, repeating all of the previous steps.
The until macro operator is similar to while, except that the until form terminates when expression evaluates true, rather than false.
These operators arrange for the evaluation of all their enclosed forms in an anonymous block. Any of the forms, or expression, may use the return operator to terminate the loop, and optionally to specify a result value for the form.
The only way these forms can yield a value other than nil is if the return operator is used to terminate the implicit anonymous block, and is given an argument, which becomes the result value.
(while* expression form*)
(until* expression form*)
The while* and until* macros are similar, respectively, to the macros while and until.
They differ in one respect: they begin by evaluating the forms one time unconditionally, without first evaluating expression. After this evaluation, the subsequent behavior is like that of while or until.
Another way to regard the behavior is that that these forms execute one iteration unconditionally, without evaluating the termination test prior to the first iteration. Yet another view is that these constructs relocate the test from the top of the loop to the bottom of the loop.
(whilet ({sym | (sym init-form)}+)
body-form*)
The whilet macro provides a construct which combines iteration with variable binding.
The evaluation of the form takes place as follows. First, fresh bindings are established for syms as if by the let* operator. It is an error for the list of variable bindings to be empty.
After the establishment of the bindings, the value of the last sym is tested. If the value is nil, then whilet terminates. Otherwise, body-forms are evaluated in the scope of the variable bindings, and then whilet iterates from the beginning, again establishing fresh bindings for the syms, and testing the value of the last sym.
All evaluation takes place in an anonymous block, which can be terminated with the return operator. Doing so terminates the loop. If the whilet loop is thus terminated by an explicit return, a return value can be specified. Under normal termination, the return value is nil.
In the syntax, a small convenience is permitted. Instead of the last (sym init-form) it is permissible for the syntax (init-form) to appear, the sym being omitted. A machine-generated variable is substituted in place of the missing sym and that variable is then initialized from init-form and used as the basis of the test.
;; read lines of text from *stdin* and print them,
;; until the end-of-stream condition:
(whilet ((line (get-line)))
(put-line line))
;; read lines of text from *stdin* and print them,
;; until the end-of-stream condition occurs or
;; a line is identical to the character string "end".
(whilet ((line (get-line))
(more (and line (nequal line "end"))))
(put-line line))
(iflet {({sym | (sym init-form)}+) | atom-form}
then-form [else-form])
(whenlet {({sym | (sym init-form)}+) | atom-form}
body-form*)
The iflet and whenlet macros combine the variable binding of let* with conditional evaluation of if and when, respectively.
In either construct's syntax, a non-compound form atom-form may appear in place of the variable binding list. In this case, atom-form is evaluated as a form, and the construct is equivalent to its respective ordinary if or when counterpart.
If the list of variable bindings is empty, it is interpreted as the atom nil and treated as an atom-form.
If one or more bindings are specified rather than atom-form, then the evaluation of these forms takes place as follows. First, fresh bindings are established for syms as if by the let* operator.
Then, the last variable's value is tested. If it is not nil then the test is true, otherwise false.
In the syntax, a small convenience is permitted. Instead of the last (sym init-form) it is permissible for the syntax (init-form) to appear, the sym being omitted. A machine-generated variable is substituted in place of the missing sym and that variable is then initialized from init-form and used as the basis of the test. This is intended to be useful in situations in which then-form or else-form do not require access to the tested value.
In the case of the iflet operator, if the test is true, the operator evaluates then-form and yields its value. Otherwise the test is false, and if the optional else-form is present, that is evaluated instead and its value is returned. If this form is missing, then nil is returned.
In the case of the whenlet operator, if the test is true, then the body-forms, if any, are evaluated. The value of the last one is returned, otherwise nil if the forms are missing. If the test is false, then evaluation of body-forms is skipped, and nil is returned.
;; dispose of foo-resource if present
(whenlet ((foo-res (get-foo-resource obj)))
(foo-shutdown foo-res)
(set-foo-resource obj nil))
;; Contrast with: above, using when and let
(let ((foo-res (get-foo-resource obj)))
(when foo-res
(foo-shutdown foo-res)
(set-foo-resource obj nil)))
;; print frobosity value if it exceeds 150
(whenlet ((fv (get-frobosity-value))
(exceeds-p (> fv 150)))
(format t "frobosity value ~a exceeds 150\n" fv))
;; same as above, taking advantage of the
;; last variable being optional:
(whenlet ((fv (get-frobosity-value))
((> fv 150)))
(format t "frobosity value ~a exceeds 150\n" fv))
;; yield 4: 3 interpreted as atom-form
(whenlet 3 4)
;; yield 4: nil interpreted as atom-form
(iflet () 3 4)
(condlet
([({ sym | (sym init-form)}+) | atom-form]
body-form*)*)
The condlet macro generalizes iflet.
Each argument is a compound consisting of at least one item: a list of bindings or atom-form. This item is followed by zero or more body-forms.
If there are no body-forms then the situation is treated as if there were a single body-form specified as nil.
The arguments of condlet are considered in sequence, starting with the leftmost.
If the argument's left item is an atom-form then the form is evaluated. If it yields true, then the body-forms next to it are evaluated in order, and the condlet form terminates, yielding the value obtained from the last body-form. If atom-form yields false, then the next argument is considered, if there is one.
If the argument's left item is a list of bindings, then it is processed with exactly the same logic as under the iflet macro. If the last binding contains a true value, then the adjoining body-forms are evaluated in a scope in which all of the bindings are visible, and condlet terminates, yielding the value of the last body-form. Otherwise, the next argument of condlet is considered (processed in a scope in which the bindings produced by the current item are no longer visible).
If condlet runs out of arguments, it terminates and returns nil.
(let ((l '(1 2 3)))
(condlet
;; first arg
(((a (first l) ;; a binding gets 1
(b (second l)) ;; b binding gets 2
(g (> a b)))) ;; last variable g is nil
'foo) ;; not evaluated
;; second arg
(((b (second l) ;; b gets 2
(c (third l)) ;; c gets 3
(g (> b c)))) ;; last variable g is true
'bar))) ;; condlet terminates
--> bar ;; result is bar
(ifa cond then [else])
The ifa macro provides an anaphoric conditional operator resembling the if operator. Around the evaluation of the then and else forms, the symbol it is implicitly bound to a subexpression of cond, a subexpression which is thereby identified as the it-form. This it alias provides a convenient reference to that place or value, similar to the word "it" in the English language, and similar anaphoric pronouns in other languages.
If it is bound to a place form, the binding is established as if using the placelet operator: the form is evaluated only once, even if the it alias is used multiple times in the then or else expressions. Furthermore, the place form is implicitly surrounded with read-once so that the place's value is accessed only once, and multiple references to it refer to a copy of the value cached in a hidden variable, rather than generating multiple accesses to the place. Otherwise, if the form is not a syntactic place it is bound as an ordinary lexical variable to the form's value.
An it-candidate is an an expression viable for having its value or storage location bound to the it symbol. An it-candidate is any expression which is not a constant expression according to the constantp function, and not a symbol.
The ifa macro imposes applies several rules to the cond expression:
(ifa (not expr) then else) -> (ifa expr else then)
which applies likewise for null or false substituted for not. The Boolean inverse function is removed, and the then and else expressions are exchanged.
In all other regards, the ifa macro behaves similarly to if.
The cond expression is evaluated, and, if applicable, the value of, or storage location denoted by the appropriate argument is captured and bound to the variable it whose scope extends over the then form, as well as over else, if present.
If cond yields a true value, then then is evaluated and the resulting value is returned, otherwise else is evaluated if present and its value is returned. A missing else is treated as if it were the nil form.
(ifa t 1 0) -> 1
;; Rule 6: it binds to (* x x), which is
;; the only it-candidate.
(let ((x 6) (y 49))
(ifa (> y (* x x)) ;; it binds to (* x x)
(list it)))
-> (36)
;; Rule 4: it binds to argument of evenp,
;; even though 4 isn't an it-candidate.
(ifa (evenp 4)
(list it))
-> (4)
;; Rule 5:
(ifa (not (oddp 4))
(list it))
-> (4)
;; Rule 7: no candidates: choose leftmost
(let ((x 6) (y 49))
(ifa (< 1 x y)
(list it)))
-> (1)
-> (4)
;; Violation of Rule 1:
;; while is not a function
(ifa (while t (print 42))
(list it))
--> exception!
;; Violation of Rule 2:
(let ((x 6) (y 49))
(ifa (> (* y y y) (* x x)))
(list it))
--> exception!
(conda {(test form*)}*)
The conda operator provides a multi-branching conditional evaluation of forms, similarly to the cond operator. Enclosed in the cond form are groups of forms expressed as lists. Each group must be a list of at least one form.
The conda operator is anaphoric: it expands into a nested structure of zero or more ifa invocations, according to these patterns:
(conda) -> nil
(conda (x y ...) ...) -> (ifa x (progn y ...) (conda ...))
Thus, conda inherits all the restrictions on the test expressions from ifa, as well as the anaphoric it variable feature.
(whena test form*)
The whena macro is similar to the when macro, except that it is anaphoric in exactly the same manner as the ifa macro. It may be understood as conforming to the following equivalence:
(whena x f0 f2 ...) <--> (if x (progn f0 f2 ...))
(dotimes (var count-form [result-form])
body-form*)
The dotimes macro implements a simple counting loop. var is established as a variable, and initialized to zero. count-form is evaluated one time to produce a limiting value, which should be a number. Then, if the value of var is less than the limiting value, the body-forms are evaluated, var is incremented by one, and the process repeats with a new comparison of var against the limiting value possibly leading to another evaluation of the forms.
If var is found to equal or exceed the limiting value, then the loop terminates.
When the loop terminates, its return value is nil unless a result-form is present, in which case the value of that form specifies the return value.
body-forms as well as result-form are evaluated in the scope in which the binding of var is visible.
(each ({(sym init-form)}*) body-form*)
(each* ({(sym init-form)}*) body-form*)
(collect-each ({(sym init-form)}*) body-form*)
(collect-each* ({(sym init-form)}*) body-form*)
(append-each ({(sym init-form)}*) body-form*)
(append-each* ({(sym init-form)}*) body-form*)
These operators establish a loop for iterating over the elements of one or more sequences. Each init-form must evaluate to an iterable object that is suitable as an argument for the iter-begin function. The sequences are then iterated in parallel over repeated evaluations of the body-forms, with each sym variable being assigned to successive elements of its sequence. The shortest list determines the number of iterations, so if any of the init-forms evaluate to an empty sequence, the body is not executed.
If the list of (sym init-form) pairs itself is empty, then an infinite loop is specified.
The body forms are enclosed in an anonymous block, allowing the return operator to terminate the loop prematurely and optionally specify the return value.
The collect-each and collect-each* variants are like each and each*, except that for each iteration, the resulting value of the body is collected into a list. When the iteration terminates, the return value of the collect-each or collect-each* operator is this collection.
The append-each and append-each* variants are like each and each*, except that for each iteration other than the last, the resulting value of the body must be a list. The last iteration may produce either an atom or a list. The objects produced by the iterations are combined together as if they were arguments to the append function, and the resulting value is the value of the append-each or append-each* operator.
The alternate forms denoted by the adorned symbols each*, collect-each* and append-each*, differ from each, collect-each and append-each in the following way. The plain forms evaluate the init-forms in an environment in which none of the sym variables are yet visible. By contrast, the alternate forms evaluate each init-form in an environment in which bindings for the previous sym variables are visible. In this phase of evaluation, sym variables are list-valued: one by one they are each bound to the list object emanating from their corresponding init-form. Just before the first loop iteration, however, the sym variables are assigned the first item from each of their lists.
The semantics of collect-each may be understood in terms of an equivalence to a code pattern involving mapcar:
(collect-each ((x xinit) (mapcar (lambda (x y)
(y yinit)) <--> body)
body) xinit yinit)
The collect-each* variant may be understood in terms of the following equivalence involving let* for sequential binding and mapcar:
(collect-each* ((x xinit) (let* ((x xinit)
(y yinit)) <--> (y yinit))
body) (mapcar (lambda (x y)
body)
x y))
However, note that the let* as well as each invocation of the lambda binds fresh instances of the variables, whereas these operators are permitted to bind a single instance of the variables, which are first initialized with the initializing expressions, and then reused as iteration variables which are stepped by assignment.
The other operators may be understood likewise, with the substitution of the mapdo function in the case of each and each* and of the mappend function in the case of append-each and append-each*.
;; print numbers from 1 to 10 and whether they are even or odd
(each* ((n 1..11) ;; n is just a range object in this scope
(even (collect-each ((m n)) (evenp m))))
;; n is an integer in this scope
(format t "~s is ~s\n" n (if even "even" "odd")))
1 is "odd"
2 is "even"
3 is "odd"
4 is "even"
5 is "odd"
6 is "even"
7 is "odd"
8 is "even"
9 is "odd"
10 is "even"
({for | for*} ({sym | (sym init-form)}*)
([test-form result-form*])
[(inc-form*)]
body-form*)
({for | for*} ({sym | (sym init-form)}*)
([test-form result-form*]))
({for | for*} ({sym | (sym init-form)}*))
The macros for and for* combine variable binding with loop iteration. The first argument is a list of variables with optional initializers, exactly the same as in the let and let* operators. Furthermore, the difference between for and for* is like that between let and let* with regard to this list of variables.
The second variant in the above syntax synopsis shows that when body-forms are absent, then a list of inc-forms which is empty may be omitted from the syntax.
The for and for* macros execute these steps:
({doloop | doloop*}
({ sym | (sym [init-form [step-form])}*)
([test-form result-form*])
tagbody-form*)
The doloop and doloop* macros provide an iteration construct inspired by the ANSI Common Lisp do and do* macros.
Each sym element in the form must be a symbol suitable for use as a variable name.
The tagbody-forms are placed into an implicit tagbody, meaning that a tagbody-form which is an integer, character or symbol is interpreted as a tagbody label which may be the target of a control transfer via the go macro.
The doloop macro binds each sym to the value produced by evaluating the adjacent init-form. Then, in the environment in which these variables now exist, test-form is evaluated. If that form yields nil, then the loop terminates. The result-forms are evaluated, and the value of the last one is returned.
If result-forms are absent, then nil is returned.
If test-form is also absent, then the loop terminates and returns nil.
If test-form produces a true value, then result-forms are not evaluated. Instead, the implicit tagbody consisting of the tagbody-forms is evaluated. If that evaluation terminates normally, the loop variables are then updated by assigning to each sym the value of step-form.
The following defaulting behaviors apply in regard to the variable syntax. For each sym which has an associated init-form but no step-form, the init-form is duplicated and taken as the step-form. Thus a variable specification like (x y) is equivalent to (x y y). If both forms are omitted, then the init-form is taken to be nil, and the step-form is taken to be sym. This means that the variable form (x) is equivalent to (x nil x) which has the effect that x retains its current value when the next loop iteration begins. Lastly, the sym variant is equivalent to (sym) so that x is also equivalent to (x nil x).
The differences between doloop and doloop* are: doloop binds the variables in parallel, similarly to let, whereas doloop* binds sequentially, like let*; moreover, doloop performs the step-form assignments in parallel as if using a single (pset sym0 step-form-0 sym1 step-form-1 ...) form, whereas doloop* performs the assignment sequentially as if using set rather than pset.
The doloop and doloop* macros establish an anonymous block, allowing early return from the loop, with a value, via the return operator.
These macros are substantially different from the ANSI Common Lisp do and do* macros. Firstly, the termination logic is inverted; effectively they implement "while" loops, whereas their ANSI CL counterparts implement "until" loops. Secondly, in the ANSI CL macros, the defaulting of the missing step-form is different. Variables with no step-form are not updated. In particular, this means that the form (x y) is not equivalent to (x y y); the ANSI CL macros do not feature the automatic replication of init-form into the step-form position.
(sum-each ({(sym init-form)}*) body-form*)
(sum-each* ({(sym init-form)}*) body-form*)
(mul-each ({(sym init-form)}*) body-form*)
(mul-each* ({(sym init-form)}*) body-form*)
The macros sum-each, and mul-each behave very similarly to the each operator. Whereas the each operator form returns nil as its result, the sum-each and mul-each forms, if they execute to completion and return normally, return an accumulated value.
The sum-each macro initializes newly instantiated, hidden accumulator variable to the value 0. For each iteration of the loop, the body-forms are evaluated, and are expected to produce a value. This value is added to the current value of the hidden accumulator using the + function, and the result is stored into the accumulator. If sum-each returns normally, then the value of this accumulator becomes its resulting value.
The mul-each macro similarly initializes a hidden accumulator to the value 1. The value from each iteration of the body is multiplied with the accumulator using the * function, and the result is stored into the accumulator. If mul-each returns normally, then the value of this accumulator becomes its resulting value.
The sum-each* and mul-each* variants of the macros implement the sequential scoping rule for the variable bindings, exactly the way each* alters the semantics of each.
The body-forms are enclosed in an implicit anonymous block. If the forms terminate by returning from the anonymous block then these macros terminate with the specified value.
When sum-each* and sum-each are specified with variables whose values specify zero iterations, or with no variables at all, the form terminates with a value of 0. In this situation, mul-each and mul-each* terminate with 1. Note that this behavior differs from each, and its closely-related operators, which loop infinitely when no variables are specified.
It is unspecified whether mul-each and mul-each* continue iterating when the accumulator takes on a value satisfying the zerop predicate.
(each-true ({(sym init-form)}*) body-form*)
(some-true ({(sym init-form)}*) body-form*)
(each-false ({(sym init-form)}*) body-form*)
(some-false ({(sym init-form)}*) body-form*)
These macros iterate zero or more variables over sequences, similarly to the each operator, and calculate logical results, with short-circuiting semantics.
The each-true macro initializes an internal result variable to the t value. It then evaluates the body-forms for each tuple of variable values, replacing the result variable with the value produced by these forms. If that value is nil, the iteration stops. When the iteration terminates normally, the value of the result variable is returned.
If no variables are specified, termination occurs immediately. Note that this is different from the each operator, which iterates indefinitely if no variables are specified.
The body-forms are surrounded by an implicit anonymous block, making it possible to terminate via return or return-from. In these cases, the form terminates with nil or the specified return value. The internal result is ignored.
The some-true macro is similar to each-true, with the following differences. The internal result variable is initialized to nil rather than t. The iteration stops whenever the body-forms produce a true value, and that value is returned.
The each-false and some-false macros are, respectively, similar to each-true and some-true, with one difference. After each iteration, the value produced by the body-forms is logically inverted using the not function prior to being assigned to the result variable.
(each-true ()) -> t
(each-true ((a ()))) -> t
(each-true ((a '(1 2 3))) a) -> 3
(each-true ((a '(1 2 3))
(b '(4 5 6)))
(< a b))
-> t
(each-true ((a '(1 2 3))
(b '(4 0 6)))
(< a b))
-> nil
(some-true ((a '(1 2 3))) a) -> 1
(some-true ((a '(nil 2 3))) a) -> 2
(some-true ((a '(nil nil nil))) a) -> nil
(some-true ((a '(1 2 3))
(b '(4 0 6)))
(< a b))
-> t
(some-true ((a '(1 2 3))
(b '(0 1 2)))
(< a b))
-> nil
(each-false ((a '(1 2 3))
(b '(4 5 6)))
(> a b))
-> t
(each-false ((a '(1 2 3))
(b '(4 0 6)))
(> a b))
-> nil
(some-false ((a '(1 2 3))
(b '(4 0 6)))
(> a b))
-> t
(some-false ((a '(1 2 3))
(b '(0 1 2)))
(> a b))
-> nil
(each-prod ({(sym init-form)}*) body-form*)
(collect-each-prod ({(sym init-form)}*) body-form*)
(append-each-prod ({(sym init-form)}*) body-form*)
The macros each-prod, collect-each-prod and append-each-prod have a similar syntax to each, collect-each and collect-each-prod. However, instead of iterating over sequences in parallel, they iterate over the Cartesian product of the elements from the sequences. The difference between collect-each and collect-each-prod is analogous to that between the functions mapcar and maprod.
Like in the each operator family, the body-forms are surrounded by an anonymous block. If these forms execute a return from this block, then these macros terminate with the specified return value.
When no iterations are performed, including in the case when an empty list of variables is specified, all these macro forms terminate and return nil. Note that this behavior differs from each, and its closely-related operators, which loop infinitely when no variables are specified.
With one caveat noted below, these macros can be understood as providing syntactic sugar according to the pattern established by the following equivalences:
(each-prod (block nil
((x xinit) (let ((#:gx xinit) (#:gy yinit))
(y yinit)) <--> (maprodo (lambda (x y)
body) body)
#:gx #:gy))
(collect-each-prod (block nil
((x xinit) (let ((#:gx xinit) (#:gy yinit))
(y yinit)) <--> (maprod (lambda (x y)
body) body)
#:gx #:gy))
(append-each-prod (block nil
((x xinit) (let ((#:gx xinit) (#:gy yinit))
(y yinit)) <--> (maprend (lambda (x y)
body) body)
#:gx #:gy))
However, note that each invocation of the lambda binds fresh instances of the variables, whereas these operators are permitted to bind a single instance of the variables, which are then stepped by assignment.
(collect-each-prod ((a '(a b c))
(n #(1 2)))
(cons a n))
--> ((a . 1) (a . 2) (b . 1) (b . 2) (c . 1) (c . 2))
(each-prod* ({(sym init-form)}*) body-form*)
(collect-each-prod* ({(sym init-form)}*) body-form*)
(append-each-prod* ({(sym init-form)}*) body-form*)
The macros each-prod*, collect-each-prod* and append-each-prod* are variants of each-prod, collect-each-prod and append-each-prod with sequential binding.
These macros can be understood as providing syntactic sugar according to the pattern established by the following equivalences:
(each-prod* (let* ((x xinit)
((x xinit) (y yinit))
(y yinit)) <--> (maprodo (lambda (x y) body)
body) x y)
(collect-each-prod* (let* ((x xinit)
((x xinit) (y yinit))
(y yinit)) <--> (maprod (lambda (x y) body)
body) x y)
(append-each-prod* (let* ((x xinit)
((x xinit) (y yinit))
(y yinit)) <--> (maprend (lambda (x y) body)
body) x y)
However, note that the let* as well as each invocation of the lambda binds fresh instances of the variables, whereas these operators are permitted to bind a single instance of the variables, which are first initialized with the initializing expressions, and then reused as iteration variables which are stepped by assignment.
(collect-each-prod* ((a "abc")
(b (upcase-str a)))
`@a@b`)
--> ("aA" "aB" "aC" "bA" "bB" "bC" "cA" "cB" "cC")
(sum-each-prod ({(sym init-form)}*) body-form*)
(sum-each-prod* ({(sym init-form)}*) body-form*)
(mul-each-prod ({(sym init-form)}*) body-form*)
(mul-each-prod* ({(sym init-form)}*) body-form*)
The macros sum-each-prod and mul-each-prod have a similar syntax to sum-each and mul-each. However, instead of iterating over sequences in parallel, they iterate over the Cartesian product of the elements from the sequences.
The macros sum-each-prod* and mul-each-prod* variants perform sequential variable binding when establishing the initial values of the variables, similarly to the each* operator.
The body-forms are surrounded by an implicit anonymous block. If these forms execute a return from this block, then these macros terminate with the specified return value.
When no iterations are specified, including in the case when an empty list of variables is specified, the summing macros terminate, yielding 0, and the multiplicative macros terminate with 1. Note that this behavior differs from each, and its closely-related operators, which loop infinitely when no variables are specified.
;; Inefficiently calculate (+ (* 1 2 3) (* 4 3 2)).
;; Every value from (1 2 3) is paired with every value
;; from (4 3 2) to form a partial products, and
;; sum-each-prod adds these together implicitly:
(sum-each-prod ((x '(1 2 3))
(y '(4 3 2)))
(* x y))
-> 54
(block name body-form*)
(block* name-form body-form*)
The block operator introduces a named block around the execution of some forms. The name argument may be any object, though block names are usually symbols. Two block name objects are considered to be the same name according to eq equality. Since a block name is not a variable binding, keyword symbols are permitted, and so are the symbols t and nil. A block named by the symbol nil is slightly special: it is understood to be an anonymous block.
The block* operator differs from block in that it evaluates name-form, which is expected to produce a symbol. The resulting symbol is used for the name of the block.
A named or anonymous block establishes an exit point for the return-from or return operator, respectively. These operators can be invoked within a block to cause its immediate termination with a specified return value.
A block also establishes a prompt for a delimited continuation. Anywhere in a block, a continuation can be captured using the sys:capture-cont function. Delimited continuations are described in the section Delimited Continuations. A delimited continuation allows an apparently abandoned block to be restarted at the capture point, with the entire call chain and dynamic environment between the prompt and the capture point intact.
Blocks in TXR Lisp have dynamic scope. This means that the following situation is allowed:
(defun func () (return-from foo 42))
(block foo (func))
The function can return from the foo block even though the foo block does not lexically surround foo.
It is because blocks are dynamic that the block* variant exists; for lexically scoped blocks, it would make little sense to have support a dynamically computed name.
Thus blocks in TXR Lisp provide dynamic nonlocal returns, as well as returns out of lexical nesting.
It is permitted for blocks to be aggressively progn-converted by compilation. This means that a block form which meets certain criteria is converted to a progn form which surrounds the body-forms and thus no longer establishes an exit point.
A block form will be spared from progn-conversion by the compiler if it meets the following rules.
Additionally, the compiler may progn-convert blocks in contravention of the above rules, but only if doing so makes no difference to visible program behavior.
(defun helper ()
(return-from top 42))
;; defun implicitly defines a block named top
(defun top ()
(helper) ;; function returns 42
(prinl 'notreached)) ;; never printed
(defun top2 ()
(let ((h (fun helper)))
(block top (call h)) ;; may progn-convert
(block top (call 'helper)) ;; may progn-convert
(block top (helper)))) ;; not removed
In the above examples, the block containing
(call h)
may be converted to
progn
because it doesn't express a
direct
call to the
helper
function. The block which calls
helper
using
(call 'helper)
is also not considered to be making a direct call.
In Common Lisp, blocks are lexical. A separate mechanism consisting of catch and throw operators performs nonlocal transfer based on symbols. The TXR Lisp example:
(defun func () (return-from foo 42))
(block foo (func))
is not allowed in Common Lisp, but can be transliterated to:
(defun func () (throw 'foo 42))
(catch 'foo (func))
Note that foo is quoted in CL. This underscores the dynamic nature of the construct. throw itself is a function and not an operator. Also note that the CL example, in turn, is even more closely transcribed back into TXR Lisp simply by replacing its throw and catch with return* and block*:
(defun func () (return* 'foo 42))
(block* 'foo (func))
Common Lisp blocks also do not support delimited continuations.
(return [value])
(return-from name [value])
The return operator must be dynamically enclosed within an anonymous block (a block named by the symbol nil). It immediately terminates the evaluation of the innermost anonymous block which encloses it, causing it to return the specified value. If the value is omitted, the anonymous block returns nil.
The return-from operator must be dynamically enclosed within a named block whose name matches the name argument. It immediately terminates the evaluation of the innermost such block, causing it to return the specified value. If the value is omitted, that block returns nil.
(block foo
(let ((a "abc\n")
(b "def\n"))
(pprint a *stdout*)
(return-from foo 42)
(pprint b *stdout*)))
Here, the output produced is "abc". The value of b is not printed because. return-from terminates block foo, and so the second pprint form is not evaluated.
(return* name [value])
The return* function is similar to the return-from operator, except that name is an ordinary function parameter, and so when return* is used, an argument expression must be specified which evaluates to a symbol. Thus return* allows the target block of a return to be dynamically computed.
The following equivalence holds between the operator and function:
(return-from a b) <--> (return* 'a b)
Expressions used as name arguments to return* which do not simply quote a symbol have no equivalent in return-from.
(tagbody {form | label}*)
(go label)
The tagbody macro provides a form of the "go to" control construct. The arguments of a tagbody form are a mixture of zero or more forms and go labels. The latter consist of those arguments which are symbols, integers or characters. Labels are not considered by tagbody and go to be forms, and are not subject to macro expansion or evaluation.
The go macro is available inside tagbody. It is erroneous for a go form to occur outside of a tagbody. This situation is diagnosed by global macro called go, which unconditionally throws an error.
In the absence of invocations of go or other control transfers, the tagbody macro evaluates each form in left-to-right order. The go labels are ignored. After the last form is evaluated, the tagbody form terminates, and yields nil.
Any form itself, or else any of its subforms, may be the form (go label) where label matches one of the go labels of a surrounding tagbody. When this go form is evaluated, then the evaluation of form is immediately abandoned, and control transfers to the specified label. The forms are then evaluated in left-to-right order starting with the form immediately after that label. If the label is not followed by any forms, then the tagbody terminates. If label doesn't match to any label in any surrounding tagbody, the go form is erroneous.
The abandonment of a form by invocation of go is a dynamic transfer. All necessary unwinding inside form takes place.
The go labels are lexically scoped, but dynamically bound. Their scope being lexical means that the labels are not visible to forms which are not enclosed within the tagbody, even if their evaluation is invoked from that tagbody. The dynamic binding means that the labels of a tagbody form are established when it begins evaluating, and removed when that form terminates. Once a label is removed, it is not available to be the target of a go control transfer, even if that go form has the label in its lexical scope. Such an attempted transfer is erroneous.
It is permitted for tagbody forms to nest arbitrarily. The labels of an inner tagbody are not visible to an outer tagbody. However, the reverse is true: a go form in an inner tagbody may branch to a label in an outer tagbody, in which case the entire inner tagbody terminates.
In cases where the same objects are used as labels by an inner and outer tagbody, the inner labels shadow the outer labels.
There is no restriction on what kinds of symbols may be labels. Symbols in the keyword package as well as the symbols t and nil are valid tagbody labels.
ANSI Common Lisp tagbody supports only symbols and integers as labels (which are called "go tags"); characters are not supported.
;; print the numbers 1 to 10
(let ((i 0))
(tagbody
(go skip) ;; forward goto skips 0
again
(prinl i)
skip
(when (<= (inc i) 10)
(go again))))
;; Example of erroneous usage: by the time func is invoked
;; by (call func) the tagbody has already terminated. The
;; lambda body can still "see" the label, but it doesn't
;; have a binding.
(let (func)
(tagbody
(set func (lambda () (go label)))
(go out)
label
(prinl 'never-reached)
out)
(call func))
;; Example of unwinding when the unwind-protect
;; form is abandoned by (go out). Output is:
;; reached
;; cleanup
;; out
(tagbody
(unwind-protect
(progn
(prinl 'reached)
(go out)
(prinl 'notreached))
(prinl 'cleanup))
out
(prinl 'out))
(prog ({sym | (sym init-form)}*)
{body-form | label}*)
(prog* ({sym | (sym init-form)}*)
{body-form | label}*)
The prog and progn* macros combine the features of let and let*, respectively, anonymous block and tagbody.
The prog macro treats the sym and init-form expressions similarly to let, establishing variable bindings in parallel. The prog* macro treats these expressions in a similar way to let*.
The forms enclosed are treated like the argument forms of the tagbody macro: labels are permitted, along with use of go.
Finally, an anonymous block is established around all of the enclosed forms (both the init-forms and body-formss) allowing the use of return to terminate evaluation with a value.
The prog macro may be understood according to the following equivalence:
(prog vars forms ...) <--> (block nil
(let vars
(tagbody forms ...)))
Likewise, the prog* macro follows an analogous equivalence, with let replaced by let*.
(eval form [env [menv]])
The eval function treats the form object as a Lisp expression, which is expanded and evaluated. The side effects implied by the form are performed, and the value which it produces is returned.
The optional env argument specifies an environment for resolving the function and variable references encountered in form. If this argument is omitted, then evaluation takes place in the global environment.
The optional menv object specifies a macro environment for expanding macros encountered in form. If this argument is omitted, then form may refer to only global macros.
If both menv and env are specified, then env takes precedence over menv, behaving like a more nested scope. Definitions contained in env shadow same-named definitions in menv.
The form is not expanded all at once. Rather, it is treated by the following algorithm:
For instance, a form like (progn (defmacro foo ()) (foo)) may be processed with eval, because the above algorithm ensures that the (defmacro foo ()) expression is fully evaluated first, thereby providing the macro definition required by (foo).
This expansion and evaluation order is important because the semantics of eval forms the reference model for how the load function processes top-level forms. Moreover, file compilation perform a similar treatment of top-level forms and incremental macro compilation. The result is that the behavior is consistent between source files and compiled files. See the sections Top-Level Forms and File Compilation Model.
Note that, according to these rules, the constituent body forms of a macrolet or symacrolet top-level form are not individual top-level forms, even if the expansion of the construct combines the expanded versions of those forms with progn.
The form (macrolet () (defmacro foo ()) (foo)) will therefore not work correctly. However, the specific problem in this situation can be be resolved by rewriting foo as a macrolet macro: (macrolet ((foo ())) (foo)).
See also: the make-env function.
(constantp form [env])
The constantp function determines whether form is a constant form, with respect to environment env.
If env is absent, the global environment is used. The env argument is used for fully expanding form prior to analyzing.
Currently, constantp returns true for any form which, after macro-expansion, is any of the following: a compound form with the symbol quote in its first position; a non-symbolic atom; or one of the symbols which evaluate to themselves and cannot be bound as variables. These symbols are the keyword symbols, and the symbols t and nil.
Additionally, constantp returns true for a compound form, or a DWIM form, whose symbol is the member of a set a large number of constant-foldable library functions, and whose arguments are, recursively, constantp expressions for the same environment. The arithmetic functions are members of this set.
For all other inputs, constantp returns nil.
Note: some uses of constantp require manual expansion.
(constantp nil) -> t
(constantp t) -> t
(constantp :key) -> t
(constantp :) -> t
(constantp 'a) -> nil
(constantp 42) -> t
(constantp '(+ 2 2 [* 3 (/ 4 4)])) -> t
;; symacrolet form expands to 42, which is constant
(constantp '(symacrolet ((a 42)) a))
(defmacro cp (:env e arg)
(constantp arg e))
;; macro call (cp 'a) is replaced by t because
;; the symbol a expands to (+ 2 2) in the given environment,
;; and so (* a a) expands to (* (+ 2 2) (+ 2 2)) which is constantp.
(symacrolet ((a (+ 2 2)))
(cp '(* a a))) -> t
(make-env [var-bindings [fun-bindings [next-env]]])
The make-env function creates an environment object suitable as the env parameter.
The var-bindings and fun-bindings parameters, if specified, should be association lists, mapping symbols to objects. The objects in fun-bindings should be functions, or objects callable as functions.
The next-env argument, if specified, should be an environment.
Note: bindings can also be added to an environment using the env-vbind and env-fbind functions.
(env-vbind env symbol value)
(env-fbind env symbol value)
These functions bind a symbol to a value in either the function or variable space of environment env.
Values established in the function space should be functions or objects that can be used as functions such as lists, strings, arrays or hashes.
If symbol already exists in the environment, in the given space, then its value is updated with value.
If env is specified as nil, then the binding takes place in the global environment.
(env-vbindings env)
(env-fbindings env)
(env-next env)
These function retrieve the components of env, which must be an environment. The env-vbindings function retrieves the association list representing variable bindings. Similarly, the env-fbindings retrieves the association list of function bindings. The env-next function retrieves the next environment, if env has one, otherwise nil.
If e is an environment constructed by the expression (make-env v f n), then (env-vbindings e) retrieves v, (env-fbindings e) retrieves f and (env-next e) returns n.
(symbol-function {symbol | method-name | lambda-expr})
(symbol-macro symbol)
(symbol-value symbol)
(set (symbol-function {symbol | method-name}) new-value)
(set (symbol-macro symbol) new-value)
(set (symbol-value symbol) new-value)
If given a symbol argument, the symbol-function function retrieves the value of the global function binding of the given symbol if it has one: that is, the function object bound to the symbol. If symbol has no global function binding, then nil is returned.
The symbol-function function supports method names of the form (meth struct slot) where struct names a struct type, and slot is either a static slot or one of the keyword symbols :init or :postinit which refer to special functions associated with a structure type. Names in this format are returned by the func-get-name function. The symbol-function function also supports names of the form (macro name) which denote macros. Thus, symbol-function provides unified access to functions, methods and macros.
If a lambda expression is passed to symbol-function, then the expression is macro-expanded and if that is successful, the function implied by that expression is returned. It is unspecified whether this function is interpreted or compiled.
The symbol-macro function retrieves the value of the global macro binding of symbol if it has one.
Note: the name of this function has nothing to do with symbol macros; it is named for consistency with symbol-function and symbol-value, referring to the "macro-expander binding of the symbol cell".
The value of a macro binding is a function object. Intrinsic macros are C functions in the TXR kernel, which receive the entire macro call form and macro environment, performing their own destructuring. Currently, macros written in TXR Lisp are represented as curried C functions which carry the following list object in their environment cell:
(#<environment object> macro-parameter-list body-form*)
Local macros created by macrolet have nil in place of the environment object.
This representation is likely to change or expand to include other forms in future TXR versions.
The symbol-value function retrieves the value stored in the dynamic binding of symbol that is apparent in the current context. If the variable has no dynamic binding, then symbol-value retrieves its value in the global environment. If symbol has no variable binding, but is defined as a global symbol macro, then the value of that symbol macro binding is retrieved. The value of a symbol macro binding is simply the replacement form.
Rather than throwing an exception, each of these functions returns nil if the argument symbol doesn't have the binding in the respective namespace or namespaces which that function searches.
A symbol-function, symbol-macro, or symbol-value form denotes a place, if symbol has a binding of the respective kind. This place may be assigned to or deleted. Assignment to the place causes the denoted binding to have a new value. Deleting a place with the del macro removes the binding, and returns the previous contents of that binding. A binding denoted by a symbol-function form is removed using fmakunbound, one denoted by by symbol-macro is removed using mmakunbound and a binding denoted by symbol-value is removed using makunbound.
Deleting a method via symbol-function is not possible; an attempt to do so has no effect.
Storing a value, using any one of these three accessors, to a nonexistent variable, function or macro binding, is not erroneous. It has has the effect of creating that binding.
Using symbol-function accessor to assign to a lambda expression is erroneous.
Deleting a binding, using any of these three accessors, when the binding does not exist, also isn't erroneous. There is no effect and the del operator yields nil as the prior value, consistent with the behavior when accessors are used to retrieve a nonexistent value.
In ANSI Common Lisp, the symbol-function function retrieves a function, macro or special operator binding of a symbol. These are all in one space and may not coexist. In TXR Lisp, it retrieves a symbol's function binding only. Common Lisp has an accessor named macro-function similar to symbol-macro.
(boundp symbol)
(fboundp {symbol | method-name | lambda-expr})
(mboundp symbol)
boundp returns t if the symbol is bound as a variable or symbol macro in the global environment, otherwise nil.
fboundp returns t if the symbol has a function binding in the global environment, the method specified by method-name exists, or a lambda expression argument is given. Otherwise it returns nil.
mboundp returns t if the symbol has an operator macro binding in the global environment, otherwise nil.
The boundp function in ANSI Common Lisp doesn't report that global symbol macros have a binding. They are not considered bindings. In TXR Lisp, they are considered bindings.
The ANSI Common Lisp fboundp yields true if its argument has a function, macro or operator binding, whereas the TXR Lisp fboundp does not consider operators or macros. The ANSI CL fboundp does not yield true for lambda expressions. Behavior similar to the Common Lisp expression (fboundp x) in Common Lisp can be obtained in TXR Lisp using the
(or (fboundp x) (mboundp x) (special-operator-p x))
expression, except that this will also yield true when x is a lambda expression.
The mboundp function doesn't exist in ANSI Common Lisp.
(makunbound symbol)
The function makunbound removes the binding of symbol from either the dynamic environment or the global symbol macro environment. After the call to makunbound, symbol appears to be unbound.
If the makunbound call takes place in a scope in which there exists a dynamic rebinding of symbol, the information for restoring the previous binding is not affected by makunbound. When that scope terminates, the previous binding will be restored.
If the makunbound call takes place in a scope in which the dynamic binding for symbol is the global binding, then the global binding is removed. When the global binding is removed, then if symbol was previously marked as special (for instance by defvar) this marking is removed.
Otherwise if symbol has a global symbol macro binding, that binding is removed.
If symbol has no apparent dynamic binding, and no global symbol macro binding, makunbound does nothing.
In all cases, makunbound returns symbol.
The behavior of makunbound differs from its counterpart in ANSI Common Lisp.
The makunbound function in Common Lisp only removes a value from a dynamic variable. The dynamic variable does not cease to exist, it only ceases to have a value (because a binding is a value). In TXR Lisp, the variable ceases to exist. The binding of a variable isn't its value, it is the variable itself: the association between a name and an abstract storage location, in some environment. If the binding is undone, the variable disappears.
The makunbound function in Common Lisp does not remove global symbol macros, which are not considered to be bindings in the variable namespace. That is to say, the Common Lisp boundp does not report true for symbol macros.
The Common Lisp makunbound also doesn't remove the special attribute from a symbol. If a variable is introduced with defvar and then removed with makunbound, the symbol continues to exhibit dynamic binding rather than lexical in subsequent scopes. In TXR Lisp, if a global binding is removed, so is the special attribute.
(fmakunbound symbol)
(mmakunbound symbol)
The function fmakunbound removes any binding for symbol from the function namespace of the global environment. If symbol has no such binding, it does nothing. In either case, it returns symbol.
The function mmakunbound removes any binding for symbol from the operator macro namespace of the global environment. If symbol has no such binding, it does nothing. In either case, it returns symbol.
The behavior of fmakunbound differs from its counterpart in ANSI Common Lisp. The fmakunbound function in Common Lisp removes a function or macro binding, which do not coexist.
The mmakunbound function doesn't exist in Common Lisp.
(func-get-form func)
The func-get-form function retrieves a source code form of func, which must be an interpreted function. The source code form has the syntax (name arglist body-form*) .
(func-get-name func [env])
The func-get-name tries to resolve the function object func to a name. If that is not possible, it returns nil.
The resolution is performed by an exhaustive search through up to three spaces.
If an environment is specified by env, then this is searched first. If a binding is found in that environment which resolves to the function, then the search terminates and the binding's symbol is returned as the function's name.
If the search through environment env fails, or if that argument is not specified, then the global environment is searched for a function binding which resolves to func. If such a binding is found, then the search terminates, and the binding's symbol is returned. If two or more symbols in the global environment resolve to the function, it is not specified which one is returned.
If the global function environment search fails, then the function is considered as a possible macro. The global macro environment is searched for a macro binding whose expander function is func, similarly to the way the function environment was searched. If a binding is found, then the syntax (macro name) is returned, where name is the name of the global macro binding that was found which resolves to func. If two or more global macro bindings share func, it is not specified which of those bindings provides name.
If the global macro search fails, then func is considered as a possible method. The static slot space of all struct types is searched for a slot which contains func. If such a slot is found, then the method name is returned, consisting of the syntax (meth type name) where type is a symbol denoting the struct type and name is the static slot of the struct type which holds func.
A check is also performed whether func might be equal to one of the two special functions of a structure type: its initfun or postinitfun, in which case it is returned as either the (meth type :init) or the (meth type :postinit) syntax.
If func is an interpreted function not found under any name, then a lambda expression denoting that function is returned in the syntax (lambda args form*)
If func cannot be identified as a function, then nil is returned.
(func-get-env func)
The func-get-env function retrieves the environment object associated with function func. The environment object holds the captured bindings of a lexical closure.
(fun-fixparam-count func)
(fun-optparam-count func)
The fun-fixparam-count reports func's number of fixed parameters. The fixed parameters consist of the required parameters and the optional parameters. Variadic functions have a parameter which captures the remaining arguments which are in excess of the fixed parameters. That parameter is not considered a fixed parameter and therefore doesn't contribute to this count.
The fun-optparam-count reports func's number of optional parameters.
The func argument must be a function.
Note: if a function isn't variadic (see the fun-variadic function) then the value reported by fun-fixparam-count represents the maximum number of arguments which can be passed to the function. The minimum number of required arguments can be calculated for any function by subtracting the value from fun-optparam-count from the value from fun-fixparam-count.
(fun-variadic func)
The fun-variadic function returns t if func is a variadic function, otherwise nil.
The func argument must be a function.
(interp-fun-p obj)
The interp-fun-p function returns t if obj is an interpreted function, otherwise it returns nil.
(vm-fun-p obj)
The vm-fun-p function returns t if obj a function compiled for the virtual machine: a function representation produced by means of the functions compile-file, compile-toplevel or compile. If obj is of any other type, the function returns nil.
(special-var-p obj)
The special-var-p function returns t if obj is a symbol marked for special variable binding, otherwise it returns nil. Symbols are marked special by defvar and defparm.
(special-operator-p obj)
The special-operator-p function returns t if obj is a symbol which names a special operator, otherwise it returns nil.
The symbol macro %fun% indicates the current function name, There is a global %fun% symbol macro which expands to nil. Around certain kinds of named functions, a local binding for %fun% is established which provides the function name. The purpose of this name is for use in diagnostic messages; therefore it is an abbreviated name.
The %fun% macro is established for defun, defmacro and defmeth forms. It is also established for methods defined inside a defstruct form including the methods :init, :postinit, :fini and :postfini.
The %fun% macro is visible not only to the its function's body, but also to the expressions inside the parameter list which compute the default values for optional parameters.
The name provided by %fun% is intended for use in diagnostic messages and is therefore an informal name, and not the formal name which can be passed to symbol-function to retrieve the function.
In the case of a defun function named x, the %fun% name is that symbol, x. Thus, in this case, the name is the same as the formal name. In the case of a defmacro named x, %fun% also expands to the symbol x x, but that is the formal name of the macro, which is (macro x). In the case of a method x of a structure type s, %fun% is the two-element list (s x), rather than the formal name (meth s x).
;; log a message naming the function
(defun connect-to-host (addr)
(format t "~s: connecting to host ~s" %fun% addr))
In TXR Lisp, objects obey the following type hierarchy. In this type hierarchy, the internal nodes denote abstract types: no object is an instance of an abstract type. Nodes in square brackets indicate an internal structure in the type graph, invisible to programs, and angle brackets indicate a plurality of types which are not listed by name:
t ----+--- [cobj types] ---+--- hash
| |
| +--- hash-iter
| |
| +--- stream
| |
| +--- random-state
| |
| +--- regex
| |
| +--- buf
| |
| +--- tree
| |
| +--- tree-iter
| |
| +--- seq-iter
| |
| +--- cptr
| |
| +--- dir
| |
| +--- struct-type
| |
| +--- <all structures>
| |
| +--- ... others
|
|
+--- sequence ---+--- string ---+--- str
| | |
| | +--- lstr
| | |
| | +--- lit
| |
| +--- list ---+--- null
| | |
| | +--- cons
| | |
| | +--- lcons
| |
| +--- vec
| |
| +--- <structures with car or length methods>
|
+--- number ---+--- float
| |
| +--- integer ---+--- fixnum
| |
| +--- bignum
|
+--- chr
|
+--- sym
|
+--- env
|
+--- range
|
+--- tnode
|
+--- pkg
|
+--- fun
|
+--- args
In addition to the above hierarchy, the following relationships also exist:
t ---+--- atom --- <any type other than cons> --- nil
|
+--- cons ---+--- lcons --- nil
|
+--- nil
sym --- null
struct ---- <all structures>
That is to say, the types are exhaustively partitioned into atoms and conses; an object is either a cons or else it isn't, in which case it is the abstract type atom.
The cons type is odd in that it is both an abstract type, serving as a supertype for the type lcons and it is also a concrete type in that regular conses are of this type.
The type nil is an abstract type which is empty. That is to say, no object is of type nil. This type is considered the abstract subtype of every other type, including itself.
The type nil is not to be confused with the type null which is the type of the nil symbol.
Because the type of nil is the type null and nil is also a symbol, the null type is a subtype of sym.
Lastly, the symbol struct serves as the supertype of all structures.
(typeof value)
The typeof function returns a symbol representing the type of value.
The core types are identified by the following symbols:
There are more kinds of objects, such as user-defined structures.
(subtypep left-type right-type)
The subtypep function tests whether left-type and right-type name a pair of types, such that the left type is a subtype of the right type.
The arguments are either type symbols, or structure type objects, as returned by the find-struct-type function. Thus, the symbol time, which is the name of a predefined struct type, and the object returned by (find-struct-type 'time) are considered equivalent argument values.
If either argument doesn't name a type, the behavior is unspecified.
Each type is a subtype of itself. Most other type relationships can be inferred from the type hierarchy diagrams given in the introduction to this section.
In addition, there are inheritance relationships among structures. If left-type and right-type are both structure types, then subtypep yields true if the types are the same struct type, or if the right type is a direct or indirect supertype of the left.
The type symbol struct is a supertype of all structure types.
(typep object type-symbol)
The typep function tests whether the type of object is a subtype of the type named by type-symbol.
The following equivalence holds:
(typep a b) --> (subtypep (typeof a) b)
(typecase test-form {(type-sym clause-form*)}*)
The typecase macro evaluates test-form and then successively tests its type against each clause.
Each clause consists of a type symbol type-sym and zero or more clause-forms.
The first clause whose type-sym is a supertype of the type of test-form's value is considered to be the matching clause. That clause's clause-forms are evaluated, and the value of the last form is returned.
If there is no matching clause, or there are no clauses present, or the matching clause has no clause-forms, then nil is returned.
Note: since t is the supertype of every type, a clause whose type-sym is the symbol t always matches. If such a clause is placed as the last clause of a typecase, it provides a fallback case, whose forms are evaluated if none of the previous clauses match.
(etypecase test-form {(type-sym clause-form*)}*)
The etypecase macro is the error-catching variant of typecase, similar to the relationship between the ecaseq and caseq families of macros.
If one of the clauses has a type-sym which is the symbol t, then etypecase is precisely equivalent to typecase. Otherwise, a clause with a type-sym of t and which throws an exception of type case-error, derived from error, is appended to the existing clauses, after which the semantics follows that of typecase.
(built-in-type-p object)
The built-in-type-p function returns t if object is a symbol which is the name of a built-in type. For all other objects it returns nil.
(identity value)
(identity* value*)
(use value)
The identity function returns its argument.
If the identity* function is given at least one argument, then it returns its leftmost argument, otherwise it returns nil.
The use function is a synonym of identity.
The
identity
function is useful as a functional argument, when a transformation
function is required, but no transformation is actually desired.
In this role, the
use
synonym leads to readable code. For instance:
;; construct a function which returns its integer argument
;; if it is odd, otherwise it returns its successor.
;; "If it's odd, use it, otherwise take its successor".
[iff oddp use succ]
;; Applications of the function:
[[iff oddp use succ] 3] -> 3 ;; use applied to 3
[[iff oddp use succ] 2] -> 3 ;; succ applied to 2
(null value)
(not value)
(false value)
The null, not and false functions are synonyms. They tests whether value is the object nil. They return t if this is the case, nil otherwise.
(null '()) -> t
(null nil) -> t
(null ()) -> t
(false t) -> nil
(if (null x) (format t "x is nil!"))
(let ((list '(b c d)))
(if (not (memq 'a list))
(format t "list ~s does not contain the symbol a\n")))
(true value)
(have value)
The true function is the complement of the null, not and false functions. The have function is a synonym for true.
It return t if the value is any object other than nil. If value is nil, it returns nil.
Note: programs should avoid explicitly testing values with true. For instance (if x ...) should be favored over (if (true x) ...). However, the latter is useful with the ifa macro because (ifa (true expr) ...) binds the it variable to the value of expr, no matter what kind of form expr is, which is not true in the (ifa expr ...) form.
;; Compute indices where the list '(1 nil 2 nil 3)
;; has true values:
[where '(1 nil 2 nil 3) true] -> (1 3)
(eq left-obj right-obj)
(eql left-obj right-obj)
(equal left-obj right-obj)
The principal equality test functions eq, eql and equal test whether two objects are equivalent, using different criteria. They return t if the objects are equivalent, and nil otherwise.
The eq function uses the strictest equivalence test, called implementation equality. The eq function returns t if and only if, left-obj and right-obj are the same object.
Two character values are eq if they are the same character, and two fixnum integers are eq if they have the same value.
Whether two identical floating-point values are always eq depends on how TXR has been built.
On 64 bit systems, TXR is usually built with to support unboxed floating-point numbers, which may be reliably compared with eq. On 32 bit targets, floating-point values are pointers to heap-allocated values, and so two identical values might not be eq. Note that even in 64 bit build of TXR, the build configuration can override the selection so that floating-point values are heap-allocated.
All other object representations are pointers to heap-allocated objects. Two such values are eq if and only if they point to the same object in memory. So, for instance, two bignum integers might not be eq even if they have the same numeric value, two lists might not be eq even if all their corresponding elements are eq and two strings might not be eq even if they hold identical text.
The eql function is less strict than eq. The difference between eql and eq is that if left-obj and right-obj are numbers which are of the same kind and have the same numeric value, eql returns t, even if they are different objects. Note that an integers and a floating-point number are not eql even if one has a value which converts to the other: thus, (eql 0.0 0) yields nil; a comparison expression which finds these numbers equal is (= 0.0 0). The eql function also specially treats range objects. Two distinct range objects are eql if their corresponding from and to fields are eql. For all other object types, eql behaves like eq.
The equal function is less strict still than eql. In general, it recurses into some kinds of aggregate objects to perform a structural equivalence check. For struct types, it also supports customization via equality substitution. See the Equality Substitution section under Structures.
Firstly, if left-obj and right-obj are eql then they are also equal, though the converse isn't necessarily the case.
If two objects are both cons cells, then they are equal if their car fields are equal and their cdr fields are equal.
If two objects are vectors, they are equal if they have the same length, and their corresponding elements are equal.
If two objects are strings, they are equal if they are textually identical.
If two objects are functions, they are equal if they have equal environments, and if they have the same code. Two compiled functions are considered to have the same code if and only if they are pointers to the same function. Two interpreted functions are considered to have the same code if their list structure is equal.
Two hashes are equal if they use the same equality (both are :equal-based, or both are :eql-based or else both are :eq-based), if their associated user data elements are equal (see the function hash-userdata), if their sets of keys are identical, and if the data items associated with corresponding keys from each respective hash are equal objects.
Two ranges are equal if their corresponding to and from fields are equal.
For some aggregate objects, there is no special semantics. Two arguments which are symbols, packages, or streams are equal if and only if they are the same object.
Certain object types have a custom equal function.
(neq left-obj right-obj)
(neql left-obj right-obj)
(nequal left-obj right-obj)
The functions neq, neql and nequal are logically negated counterparts of, respectively, eq, eql and equal.
If eq returns t for a given pair of arguments left-obj and right-obj, then neq returns nil. Vice versa, if eq returns nil, neq returns t.
The same relationship exits between eql and neql, and between equal and nequal.
(meq left-obj right-obj*)
(meql left-obj right-obj*)
(mequal left-obj right-obj*)
The functions meq, meql and mequal ("member equal" or "multi-equal") provide a particular kind of a generalization of the binary equality functions eq, eql and equal to multiple arguments.
The left-obj value is compared to each right-obj value using the corresponding binary equality function. If a match occurs, then t is returned, otherwise nil.
The traversal of the right-obj argument values proceeds from left to right, and stops when a match is found.
(less left-obj right-obj)
(less obj obj*)
The less function, when called with two arguments, determines whether left-obj compares less than right-obj in a generic way which handles arguments of various types.
The argument syntax of less is generalized. It can accept one argument, in which case it unconditionally returns t regardless of that argument's value. If more than two arguments are given, then less generalizes in a way which can be described by the following equivalence pattern, with the understanding that each argument expression is evaluated exactly once:
(less a b c) <--> (and (less a b) (less b c))
(less a b c d) <--> (and (less a b) (less b c) (less c d))
The less function is used as the default for the lessfun argument of the functions sort and merge, as well as the testfun argument of the pos-min and find-min.
The less function is capable of comparing numbers, characters, symbols, strings, as well as lists and vectors of these. It can also compare buffers.
If both arguments are the same object so that (eq left-obj right-obj) holds true, then the function returns nil regardless of the type of left-obj, even if the function doesn't handle comparing different instances of that type. In other words, no object is less than itself, no matter what it is.
The less function pairs with the equal function. If values a and b are objects which are of suitable types to the less function, then exactly one of the following three expressions must be true: (equal a b), (less a b) or (less b a).
The less relation is: antisymmetric, such that if (less a b) is true, then then (less b a) is false; irreflexive, such that (less a a) is false; and transitive, such that (less a b) and (less b c) imply (less a c).
The following are detailed criteria that less applies to arguments of different types and combinations thereof.
If both arguments are numbers or characters, they are compared as if using the < function.
If both arguments are strings, they are compared as if using the string-lt function.
If both arguments are symbols, the following rules apply. If the symbols have names which are different, then the result is that of their names being compared by the string-lt function. If less is passed symbols which have the same name, and neither of these symbols has a home package, then the raw bit patterns of their values are compared as integers: effectively, the object with the lower machine address is considered lesser than the other. If only one of the two same-named symbols has no home package, then if that symbol is the left argument, less returns t, otherwise nil. If both same-named symbols have home packages, then the result of less is that of string-lt applied to the names of their respective packages. Thus a:foo is less than z:foo.
If both arguments are conses, then they are compared as follows:
This logic performs a lexicographic comparison on ordinary lists such that for instance (1 1) is less than (1 1 1) but not less than (1 0) or (1).
Note that the empty nil list nil compared to a cons is handled by type-based precedence, described below.
Two vectors are compared by less lexicographically, similarly to strings. Corresponding elements, starting with element 0, of the vectors are compared until an index position is found where corresponding elements of the two vectors are not equal. If this differing position is beyond the end of one of the two vectors, then the shorter vector is considered to be lesser. Otherwise, the result of less is the outcome of comparing those differing elements themselves with less.
Two buffers are also compared by less lexicographically, as if they were vectors of integer byte values.
Two ranges are compared by less using lexicographic logic similar to conses and vectors. The from fields of the ranges are first compared. If they are not equal, equal then less is applied to those fields and the result is returned. If the from fields are equal, then less is applied to the to fields and that result is returned.
If the two arguments are of the above types, but of different types from each other, then less resolves the situation based on the following precedence: numbers and characters are less than ranges, which are less than strings, which are less than symbols, which are less than conses, which are less than vectors, which are less than buffers.
Note that since nil is a symbol, it is ranked lower than a cons. This interpretation ensures correct behavior when nil is regarded as an empty list, since the empty list is lexicographically prior to a nonempty list.
If either argument is a structure for which the equal method is defined, the method is invoked on that argument, and the value returned is used in place of that argument for performing the comparison. Structures with no equal method cannot participate in a comparison, resulting in an error. See the Equality Substitution section under Structures.
Finally, if either of the arguments has a type other than the above types, the situation is an error.
(greater left-obj right-obj)
(greater obj obj*)
The greater function is equivalent to less with the arguments reversed. That is to say, the following equivalences hold:
(greater a) <--> (less a) <--> t
(greater a b) <--> (less b a)
(greater a b c ...) <--> (less ... c b a)
The greater function is used as the default for the testfun argument of the pos-max and find-max functions.
(lequal obj obj*)
(gequal obj obj*)
The functions lequal and gequal are similar to less and greater respectively, but differ in the following respect: when called with two arguments which compare true under the equal function, the lequal and gequal functions return t.
When called with only one argument, both functions return t and both functions generalize to three or more arguments in the same way as do less and greater.
(copy object)
The copy function duplicates objects of various supported types: sequences, hashes, structures and random states. If object is nil, it returns nil. Otherwise, copy is equivalent to invoking a more specific copying function according to the type of the argument, as follows:
For all other types of object, the invocation is erroneous.
Except in the case when sequence is nil, copy returns a value that is distinct from (not eq to) sequence. This is different from the behavior of [sequence 0..t] or (sub sequence 0 t) which recognize that they need not make a copy of sequence, and just return it.
Note however, that the elements of the returned sequence may be eq to elements of the original sequence. In other words, copy is a deeper copy than just duplicating the sequence value itself, but it is not a deep copy.
(cons car-value cdr-value)
The cons function allocates, initializes and returns a single cons cell. A cons cell has two fields called car and cdr, which are accessed by functions of the same name, or by the functions first and rest, which are synonyms for these.
Lists are made up of conses. A (proper) list is either the symbol nil denoting an empty list, or a cons cell which holds the first item of the list in its car, and the list of the remaining items in cdr. The expression (cons 1 nil) allocates and returns a single cons cell which denotes the one-element list (1). The cdr is nil, so there are no additional items.
A cons cell whose cdr is an atom other than nil is printed with the dotted pair notation. For example the cell produced by (cons 1 2) is denoted (1 . 2). The notation (1 . nil) is perfectly valid as input, but the cell which it denotes will print back as (1). The notations are equivalent.
The dotted pair notation can be used regardless of what type of object is the cons cell's cdr. so that for instance (a . (b c)) denotes the cons cell whose car is the symbol a a and whose cdr is the list (b c). This is exactly the same thing as (a b c). In other words (a b ... l m . (n o ... w . (x y z))) is exactly the same as (a b ... l m n o ... w x y z).
Every list, and more generally cons-cell tree structure, can be written in a "fully dotted" notation, such that there are as many dots as there are cells. For instance the cons structure of the nested list (1 (2) (3 4 (5))) can be made more explicit using (1 . ((2 . nil) . ((3 . (4 . ((5 . nil) . nil))) . nil)))). The structure contains eight conses, and so there are eight dots in the fully dotted notation.
The number of conses in a linear list like (1 2 3) is simply the number of items, so that list in particular is made of three conses. Additional nestings require additional conses, so for instance (1 2 (3)) requires four conses. A visual way to count the conses from the printed representation is to count the atoms, then add the count of open parentheses, and finally subtract one.
A list terminated by an atom other than nil is called an improper list, and the dot notation is extended to cover improper lists. For instance (1 2 . 3) is an improper list of two elements, terminated by 3, and can be constructed using (cons 1 (cons 2 3)). The fully dotted notation for this list is (1 . (2 . 3)).
(atom value)
The atom function tests whether value is an atom. It returns t if this is the case, nil otherwise. All values which are not cons cells are atoms.
(atom x) is equivalent to (not (consp x)).
(atom 3) -> t
(atom (cons 1 2)) -> nil
(atom "abc") -> t
(atom '(3)) -> nil
(consp value)
The consp function tests whether value is a cons. It returns t if this is the case, nil otherwise.
(consp x) is equivalent to (not (atom x)).
Nonempty lists test positive under consp because a list is represented as a reference to the first cons in a chain of one or more conses.
Note that a lazy cons is a cons and satisfies the consp test. See the function make-lazy-cons and the macro lcons.
(consp 3) -> nil
(consp (cons 1 2)) -> t
(consp "abc") -> nil
(consp '(3)) -> t
(car object)
(first object)
(set (car object) new-value)
(set (first object) new-value)
The functions car and first are synonyms.
If object is a cons cell, these functions retrieve the car field of that cons cell. (car (cons 1 2)) yields 1.
For programming convenience, object may be of several other kinds in addition to conses.
(car nil) is allowed, and returns nil.
object may also be a vector or a string. If it is an empty vector or string, then nil is returned. Otherwise the first character of the string or first element of the vector is returned.
object may be a structure. The car operation is possible if the object has a car method. If so, car invokes that method and returns whatever the method returns. If the structure has no car method, but has a lambda method, then the car function calls that method with one argument, that being the integer zero. Whatever the method returns, car returns. If neither method is defined, an error exception is thrown.
A car form denotes a valid place whenever object is a valid argument for the rplaca function. Modifying the place denoted by the form is equivalent to invoking rplaca with object as the left argument, and the replacement value as the right argument. It takes place in the manner given under the description rplaca function, and obeys the same restrictions.
A car form supports deletion. The following equivalence then applies:
(del (car place)) <--> (pop place)
This implies that deletion requires the argument of the car form to be a place, rather than the whole form itself. In this situation, the argument place may have a value which is nil, because pop is defined on an empty list.
The abstract concept behind deleting a car is that physically deleting this field from a cons, thereby breaking it in half, would result in just the cdr remaining. Though fragmenting a cons in this manner is impossible, deletion simulates it by replacing the place which previously held the cons, with that cons' cdr field. This semantics happens to coincide with deleting the first element of a list by a pop operation.
(cdr object)
(rest object)
(set (cdr object) new-value)
(set (rest object) new-value)
The functions cdr and rest are synonyms.
If object is a cons cell, these functions retrieve the cdr field of that cons cell. (cdr (cons 1 2)) yields 2.
For programming convenience, object may be of several other kinds in addition to conses.
(cdr nil) is allowed, and returns nil.
object may also be a vector or a string. If it is a nonempty string or vector containing at least two items, then the remaining part of the object is returned, with the first element removed. For example (cdr "abc") yields "bc". If object is a one-element vector or string, or an empty vector or string, then nil is returned. Thus (cdr "a") and (cdr "") both result in nil.
If object is a structure, then cdr requires it to support either the cdr method or the lambda method. If both are present, cdr is used. When the cdr function uses the cdr method, it invokes it with no arguments. Whatever value the method returns becomes the return value of cdr. When cdr invokes a structure's lambda method, it passes as the argument the range object #R(1 t). Whatever the lambda method returns becomes the return value of cdr.
The invocation syntax of a cdr or rest form is a syntactic place. The place is semantically correct if object is a valid argument for the rplacd function. Modifying the place denoted by the form is equivalent to invoking rplacd with object as the left argument, and the replacement value as the right argument. It takes place in the manner given under the description rplacd function, and obeys the same restrictions.
A cdr place supports deletion, according to the following near equivalence:
(del (cdr place)) <--> (prog1 (cdr place)
(set place (car place)))
The place expression is evaluated only once.
Note that this is symmetric with the delete semantics of car in that the cons stored in place goes away, as does the cdr field, leaving just the car, which takes the place of the original cons.
Example:
Walk every element of the list (1 2 3) using a for loop:
(for ((i '(1 2 3))) (i) ((set i (cdr i)))
(print (car i) *stdout*)
(print #\newline *stdout*))
The variable i marches over the cons cells which make up the "backbone" of the list. The elements are retrieved using the car function. Advancing to the next cell is achieved using (cdr i). If i is the last cell in a (proper) list, (cdr i) yields nil and so i becomes nil, the loop guard expression i fails and the loop terminates.
(rplaca object new-car-value)
(rplacd object new-cdr-value)
If object is a cons cell or lazy cons cell, then rplaca and rplacd functions assign new values into the car and cdr fields of the object. In addition, these functions are meaningful for other kinds of objects also.
Note that, except for the difference in return value, (rplaca x y) is the same as the more generic (set (car x) y), and likewise (rplacd x y) can be written as (set (cdr x) y).
The rplaca and rplacd functions return cons. Note: In TXR versions 89 and earlier, these functions returned the new value. The behavior was undocumented.
The cons argument does not have to be a cons cell. Both functions support meaningful semantics for vectors and strings. If cons is a string, it must be modifiable.
The rplaca function replaces the first element of a vector or first character of a string. The vector or string must be at least one element long.
The rplacd function replaces the suffix of a vector or string after the first element with a new suffix. The new-cdr-value must be a sequence, and if the suffix of a string is being replaced, it must be a sequence of characters. The suffix here refers to the portion of the vector or string after the first element.
It is permissible to use rplacd on an empty string or vector. In this case, new-cdr-value specifies the contents of the entire string or vector, as if the operation were done on a nonempty vector or string, followed by the deletion of the first element.
The object argument may be a structure. In the case of rplaca, the structure must have a defined rplaca method or else, failing that, a lambda-set method. The first of these methods which is available, in the given order, is used to perform the operation. Whatever the respective method returns, If the lambda-set method is used, it is called with two arguments (in addition to object): the integer zero, and new-car-value.
In the case of rplacd, the structure must have a defined rplacd method or else, failing that, a lambda-set method. The first of these methods which is available, in the given order, is used to perform the operation. Whatever the respective method returns, If the lambda-set method is used, it is called with two arguments (in addition to object): the range value #R(1 t) and new-car-value.
(first object)
(second object)
(third object)
(fourth object)
(fifth object)
(sixth object)
(seventh object)
(eighth object)
(ninth object)
(tenth object)
(set (first object) new-value)
(set (second object) new-value)
...
(set (tenth object) new-value)
Used as functions, these accessors retrieve the elements of a sequence by position. If the sequence is shorter than implied by the position, these functions return nil.
When used as syntactic places, these accessors denote the storage locations by position. The location must exist, otherwise an error exception results. The places support deletion.
(third '(1 2)) -> nil
(second "ab") -> #\b
(third '(1 2 . 3)) -> **error, improper list*
(let ((x (copy "abcd")))
(inc (third x))
x) -> "abce"
(append [sequence*])
(nconc [sequence*])
The append function creates a new object which is a catenation of the list arguments. All arguments are optional; append produces the empty list, and if a single argument is specified, that argument is returned.
If two or more arguments are present, then the situation is identified as one or more sequence arguments followed by last-arg. The sequence arguments must be sequences; last-arg may be a sequence or atom.
The append operation over three or more arguments is left-associative, such that (append x y z) is equivalent to both (append (append x y) z) and (append x (append z y)).
This allows the catenation of an arbitrary number of arguments to be understood in terms of a repeated application of the two-argument case, whose semantics is given by these rules:
(append nil nil) -> nil
(append nil '(1 2)) -> (1 2) (append nil '(1 2 . 3)) -> (1 2 . 3)
(append '(1 2) nil) -> (1 2)
(append '(1 2) #(3)) -> (1 2 . #(3)) (append '(1 2) 3) -> (1 2 . 3)
(append #(1 2) #(3 4)) -> #(1 2 3 4) (append "ab" "cd") -> "abcd" (append "ab" #(#\c #\d)) -> "abcd" (append "ab" #(3 4)) -> ;; error
(append #(1 2) 3) -> #(1 2 3) (append "ab" #\c) -> "abc" (append "ab" 3) -> ;; error
(append '(1 2 . "ab") "c") -> (1 2 . "abc") (append '(1 2 . "ab") '(2 3)) -> ;; error
(append 1 2) -> ;; error (append '(1 . 2) 3) -> ;; error
If N arguments are specified, where N > 1, then the first N-1 arguments must be proper lists. Copies of these lists are catenated together. The last argument N, shown in the above syntax as last-arg, may be any kind of object. It is installed into the cdr field of the last cons cell of the resulting list. Thus, if argument N is also a list, it is catenated onto the resulting list, but without being copied. Argument N may be an atom other than nil; in that case append produces an improper list.
The nconc function works like append, but may destructively manipulate any of the input objects.
;; An atom is returned.
(append 3) -> 3
;; A list is also just returned: no copying takes place.
;; The eq function can verify that the same object emerges
;; from append that went in.
(let ((list '(1 2 3)))
(eq (append list) list)) -> t
(append '(1 2 3) '(4 5 6) 7) -> '(1 2 3 4 5 6 . 7))
;; the (4 5 6) tail of the resulting list is the original
;; (4 5 6) object, shared with that list.
(append '(1 2 3) '(4 5 6)) -> '(1 2 3 4 5 6)
(append nil) -> nil
;; (1 2 3) is copied: it is not the last argument
(append '(1 2 3) nil) -> (1 2 3)
;; empty lists disappear
(append nil '(1 2 3) nil '(4 5 6)) -> (1 2 3 4 5 6)
(append nil nil nil) -> nil
;; atoms and improper lists other than in the last position
;; are erroneous
(append '(a . b) 3 '(1 2 3)) -> **error**
;; sequences other than lists can be catenated.
(append "abc" "def" "g" #\h) -> "abcdefgh"
;; lists followed by non-list sequences end with non-list
;; sequences catenated in the terminating atom:
(append '(1 2) '(3 4) "abc" "def") -> (1 2 3 4 . "abcdef")
(append* [list*])
The append* function lazily catenates lists.
If invoked with no arguments, it returns nil. If invoked with a single argument, it returns that argument.
Otherwise, it returns a lazy list consisting of the elements of every list argument from left to right.
Arguments other than the last are treated as lists, and traversed using car and cdr functions to visit their elements.
The last argument isn't traversed: rather, that object itself becomes the cdr field of the last cons cell of the lazy list constructed from the previous arguments.
(revappend list1 list2)
(nreconc list1 list2)
The revappend function returns a list consisting of list2 appended to a reversed copy of list1. The returned object shares structure with list2, which is unmodified.
The nreconc function behaves similarly, except that the returned object may share structure with not only list2 but also list1, which is modified.
(list value*)
The list function creates a new list, whose elements are the argument values.
(list) -> nil
(list 1) -> (1)
(list 'a 'b) -> (a b)
(list* value*)
The list* function is a generalization of cons. If called with exactly two arguments, it behaves exactly like cons: (list* x y) is identical to (cons x y). If three or more arguments are specified, the leading arguments specify additional atoms to be consed to the front of the list. So for instance (list* 1 2 3) is the same as (cons 1 (cons 2 3)) and produces the improper list (1 2 . 3). Generalizing in the other direction, list* can be called with just one argument, in which case it returns that argument, and can also be called with no arguments in which case it returns nil.
(list*) -> nil
(list* 1) -> 1
(list* 'a 'b) -> (a . b)
(list* 'a 'b 'c) -> (a b . c)
Note that unlike in some other Lisp dialects, the effect of (list* 1 2 x) can also be obtained using (list 1 2 . x). However, (list* 1 2 (func 3)) cannot be rewritten as (list 1 2 . (func 3)) because the latter is equivalent to (list 1 2 func 3).
(sub-list list [from [to]])
(set (sub-list list [from [to]]) new-value)
The sub-list function has the same parameters and semantics as the sub function, except that it operates on its list argument using list operations, and assumes that list is terminated by nil.
If a sub-list form is used as a place, then the list argument form must also be a place.
The sub-list place denotes a subrange of list as if it were a storage location. The previous value of this location, if needed, is fetched by a call to sub-list. Storing new-value to the place is performed by a call to replace-list. The return value of replace-list is stored into list. In an update operation which accesses the prior value and stores a new value, the arguments list, from, to and new-value are evaluated once.
(replace-list list item-sequence [from [to]])
The replace-list function is like the replace function, except that it operates on its list argument using list operations. It assumes that list is terminated by nil, and that it is made of cells which can be mutated using rplaca.
(listp value)
(proper-list-p value)
The listp and proper-list-p functions test, respectively, whether value is a list, or a proper list, and return t or nil accordingly.
The listp test is weaker, and executes without having to traverse the object. The value produced by the expression (listp x) is the same as that of (or (null x) (consp x)), except that x is evaluated only once. The empty list nil is a list, and a cons cell is a list.
The proper-list-p function returns t only for proper lists. A proper list is either nil, or a cons whose cdr is a proper list. proper-list-p traverses the list, and its execution will not terminate if the list is circular.
These functions return nil for list-like sequences that are not made of actual cons cells.
Dialect Note: in TXR 137 and older, proper-list-p is called proper-listp. The name was changed for adherence to conventions and compatibility with other Lisp dialects, like Common Lisp. However, the function continues to be available under the old name. Code that must run on TXR 137 and older installations should use proper-listp, but its use going forward is deprecated.
(endp object)
The endp function returns t if object is the object nil.
If object is a cons cell, then endp returns t.
Otherwise, endp function throws an exception.
(length-list list)
The length-list function returns the length of list, which may be a proper or improper list. The length of a list is the number of conses in that list.
(copy-list list)
The copy-list function which returns a list similar to list, but with a newly allocated cons-cell structure.
If list is an atom, it is simply returned.
Otherwise, list is a cons cell, and copy-list returns the same object as the expression (cons (car list) (copy-list (cdr list))).
Note that the object (car list) is not deeply copied, but only propagated by reference into the new list. copy-list produces a new list structure out of the same items that are in list.
Common Lisp does not allow the argument to be an atom, except for the empty list nil.
(length-list-< list len)
The length-list-< function determines whether the length of list, is less than the integer len.
The expression
(length-list-< x y)
is similar to, but usefully different from
(< (length-list x) y)
because length-list-< is required to only traverses list far enough to be able to determine the return value. If the end of the list is reached before len conses are encountered, the function returns t, otherwise if len conses are encountered, the function terminates immediately and returns nil.
The length-list-< function is therefore safe to use with infinite lazy lists and circular lists, for which length would not terminate.
Note: there is more generic function length-< which works with efficiently with different kinds of sequences.
Note: the length-list-< is useful in situations when a decision must be made between two algorithms based on the length of one or more input lists. The decision can be made without wastefully performing a full pass over the input lists to measure their length.
(copy-cons cons)
The copy-cons function creates and returns a new object that is a replica of cons.
The cons argument must be either a cons cell, or else a lazy cons: an object of type lcons.
A new cell of the same type as cons is created, and all of its fields are initialized by copying the corresponding fields from cons.
If cons is lazy, the newly created object is in the same state as the original. If the original has not yet been updated and thus has an update function, the copy also has not yet been updated and has the same update function.
(copy-tree obj)
The copy-tree function returns a copy of obj which represents an arbitrary cons-cell-based structure.
The cell structure of obj is traversed and a similar structure is constructed, but without regard for substructure sharing or circularity.
More precisely, if obj is an atom, then it is returned. If it is an ordinary cons cell, then copy-tree is recursively applied to the car and cdr fields to produce their individual replicas. A new cons cell is then produced from the replicated car and cdr. If obj is a lazy cons, then just like in the ordinary cons case, the car and cdr fields are duplicated with a recursive call to copy-tree. Then, a lazy cons is created from these replicated fields. If cell has an update function, then the newly created lazy cons has the same update function; the function isn't copied.
Like copy-cons, the copy-tree function doesn't trigger the update of lazy conses. The copies of lazy conses which have not been updated are also conses which have not been updated.
(reverse list)
(nreverse list)
Description:
The functions reverse and nreverse produce an object which contains the same items as proper list list, but in reverse order. If list is nil, then both functions return nil.
The reverse function is non-destructive: it creates a new list.
The nreverse function creates the structure of the reversed list out of the cons cells of the input list, thereby destructively altering it (if it contains more than one element). How nreverse uses the material from the original list is unspecified. It may rearrange the cons cells into a reverse order, or it may keep the structure intact, but transfer the car values among cons cells into reverse order. Other approaches are possible.
(nthlast index list)
(set (nthlast index list) new-value)
The nthlast function retrieves the n-th last cons cell of a list, indexed from one. The index parameter must be a an integer. If index is positive and so large that it specifies a nonexistent cons beyond the beginning of the list, nthlast returns list. Effectively, values of index larger than the length of the list are clamped to the length. If index is negative, then nthlast yields nil. An index value of zero retrieves the terminating atom of list or else the value list itself, if list is an atom.
The following equivalences hold:
(nthlast 1 list) <--> (last list)
An nthlast place designates the storage location which holds the n-th cell, as indicated by the value of index.
A negative index doesn't denote a place.
A positive index greater than the length of the list is treated as if it were equal to the length of the list.
If list is itself a syntactic place, then the index value n is permitted for a list of length n. This index value denotes the list place itself. Storing to this value overwrites list. If list isn't a syntactic place, then storing to position n isn't permitted.
If list is of length zero, or an atom (in which case its length is considered to be zero) then the above remarks about position n apply to an index value of zero: if list is a syntactic place, then the position denotes list itself, otherwise the position doesn't exist as a place.
If list contains one or more elements, then index value of zero denotes the cdr field of its last cons cell. Storing a value to this place overwrites the terminating atom.
(butlastn num list)
(set (butlastn num list) new-value )
The butlastn function calculates that initial portion of list which excludes the last num elements.
Note: the butlastn function doesn't support non-list sequences as sequences; it treats them as the terminating atom of a zero-length improper list. The butlast sequence function supports non-list sequences. If x is a list, then the following equivalence holds:
(butlastn n x) <--> (butlast x n)
If num is zero, or negative, then butlastn returns list.
If num is positive, and meets or exceeds the length of list, then butlastn returns nil.
If a butlastn form is used as a syntactic place, then list must be a place. Assigning to the form causes list to be replaced with a new list which is a catenation of the new value and the last num elements of the original list, according to the following equivalence:
(set (butlastn n x) v)
<-->
(progn (set x (append v (nthlast n x))) v)
except that n, x and v are evaluated only once, in left-to-right order.
(nth index object)
(set (nth index object) new-value)
The nth function performs random access on a list, retrieving the n-th element indicated by the zero-based index value given by index. The index argument must be a nonnegative integer.
If index indicates an element beyond the end of the list, then the function returns nil.
The following equivalences hold:
(nth 0 list) <--> (car 0) <--> (first list)
(nth 1 list) <--> (cadr list) <--> (second list)
(nth 2 list) <--> (caddr list) <--> (third list)
(nth x y) <--> (car (nthcdr x y))
(nthcdr index list)
(set (nthcdr index list) new-value)
The nthcdr function retrieves the n-th cons cell of a list, indexed from zero. The index parameter must be a nonnegative integer. If index specifies a nonexistent cons beyond the end of the list, then nthcdr yields nil.
The following equivalences hold:
(nthcdr 0 list) <--> list
(nthcdr 1 list) <--> (cdr list)
(nthcdr 2 list) <--> (cddr list)
(car (nthcdr x y)) <--> (nth x y)
An nthcdr place designates the storage location which holds the n-th cell, as indicated by the value of index. Indices beyond the last cell of list do not designate a valid place. If list is itself a place, then the zeroth index is permitted and the resulting place denotes list. Storing a value to (nthcdr 0 list) overwrites list. Otherwise if list isn't a syntactic place, then the zeroth index does not designate a valid place; index must have a positive value. A nthcdr place does not support deletion.
In Common Lisp, nthcdr is only a function, not an accessor; nthcdr forms do not denote places.
(tailp object list)
The tailp function tests whether object is a tail of list. This means that object is either list itself, or else one of the cons cells of list or else the terminating atom of list.
More formally, a recursive definition follows. If object and list are the same object (thus equal under the eq function) then tailp returns t. If list is an atom, and is not object, then the function returns nil. Otherwise, list is a cons that is not object and tailp yields the same value as the (tailp object (cdr list)) expression.
(caar object)
(cadr object)
(cdar object)
(cddr object)
...
(cdddr object)
(set (caar object) new-value)
(set (cadr object) new-value)
...
The
a-d accessors
provide a shorthand notation for accessing two to five
levels deep into a cons-cell-based tree structure. For instance, the
the equivalent of the nested function call expression
(car (car (cdr object)))
can be achieved using the single function call
(caadr object).
The symbol names of the a-d accessors are a generalization of the words
"car" and "cdr". They encode the pattern of
car
and
cdr
traversal of the structure using a sequence of the letters
a
and
d
placed between
c
and
r.
The traversal is encoded in right-to-left order, so that
cadr
indicates a traversal of the
cdr
link, followed by the
car.
This order corresponds to the nested function call notation, which also
encodes the traversal right-to-left. The following diagram illustrates
the straightforward relationship:
(cdr (car (cdr x)))
^ ^ ^
| / |
| / /
| / ____/
|| /
(cdadr x)
TXR Lisp provides all possible a-d accessors up to five levels deep, from caar all the way through cdddddr.
Expressions involving a-d accessors are places. For example, (caddr x) denotes the same place as (car (cddr x)), and (cdadr x) denotes the same place as (cdr (cadr x)).
The a-d accessor places support deletion, with semantics derived from the deletion semantics of the car and cdr places. For example, (del (caddr x)) means the same as (del (car (cddr x))).
(cyr address object)
(cxr address object)
The cyr and cxr functions provide car/cdr navigation of tree structure driven by numeric address given by the address argument.
The address argument can express any combination of the application of car and cdr functions, including none at all.
The difference between cyr and cxr is the bit order of the encoding. Under cyr, the most significant bit of the encoding given in address indicates the initial car/cdr navigation, and the least significant bit gives the final one. Under cxr, it is opposite.
Both functions require address to be a positive integer. Any other argument raises an error.
Under both functions, the address value 1 encodes the identity operation: no car/cdr
(flatten {list | atom})
(flatten* {list | atom})
The flatten function recursively traverses a nested list, returning a list whose elements are all of the non-nil atoms contained in list, at any level of nesting. If the argument is an atom rather than a list, then it is returned. Otherwise, the list argument must be a proper list, as must all lists nested within it.
The flatten* function calculates the same result as flatten, except that it produces a lazy list. It can be used to lazily flatten an infinite lazy list.
(flatten 42) -> 42
(flatten '(1 2 () (3 4))) -> (1 2 3 4)
;; equivalent to previous, since
;; nil is the same thing as ()
(flatten '(1 2 nil (3 4))) -> (1 2 3 4)
(flatten nil) -> nil
(flatten '(((()) ()))) -> nil
(flatten '(a (b . c))) -> ;; error
(flatcar tree)
(flatcar* tree)
The flatcar function produces a list of all the atoms contained in the tree structure tree, in the order in which they appear, when the structure is traversed left to right.
This list includes those nil atoms which appear in car fields.
The list excludes nil atoms which appear in cdr fields.
If the tree argument is an atom, it is returned.
The flatcar* function works like flatcar except that it produces a lazy list. It can be used to lazily flatten an infinite lazy structure.
(flatcar '(1 2 () (3 4))) -> (1 2 nil 3 4)
(flatcar '(a (b . c) d (e) (((f)) . g) (nil . z) nil . h))
--> (a b c d e f g nil z nil h)
(tree-find obj tree [test-function])
(cons-find obj tree [test-function])
The tree-find and cons-find function search tree for an occurrence of obj. Tree can be any atom, or a cons. If tree it is a cons, it is understood to be a proper list whose elements are also trees.
The equivalence test is performed by test-function which must take two arguments, and has conventions similar to eq, eql or equal. If an argument is omitted, the default function is equal.
Under both tree-find and cons-find, if tree is equivalent to obj under test-function, then t is returned to announce a successful finding. Next, if the mismatched obj is an atom, both functions return nil to indicate that the search failed.
If none of the above cases occur, the semantics of the functions diverge, as follows.
In the case of tree-find, tree is taken to be a proper list, and tree-find is recursively applied to each element of the list in turn, using the same obj and test-function arguments, stopping at the first element which returns a non-nil value.
In the case of cons-find, tree is taken to be cons-cell-based tree structure. The cons-find function is recursively applied to the car and cdr fields of tree. Thus a match may be found in any position in the structure, including the dotted position of a list.
(memq object list)
(memql object list)
(memqual object list)
The memq, memql and memqual functions search list for a member which is, respectively, eq, eql or equal to object. (See the eq, eql and equal functions above.)
If no such element found, nil is returned.
Otherwise, that suffix of list is returned whose first element is the matching object.
(member key sequence [testfun [keyfun]])
(member-if predfun sequence [keyfun])
The member and member-if functions search through sequence for an item which matches a key, or satisfies a predicate function, respectively.
The keyfun argument specifies a function which is applied to the elements of the sequence to produce the comparison key. If this argument is omitted, then the untransformed elements of the sequence themselves are examined.
The member function's testfun argument specifies the test function which is used to compare the comparison keys taken from the sequence to the search key. If this argument is omitted, then the equal function is used. If member does not find a matching element, it returns nil. Otherwise it returns the suffix of sequence which begins with the matching element.
The member-if function's predfun argument specifies a predicate function which is applied to the successive comparison keys pulled from the sequence by applying the key function to successive elements. If no match is found, then nil is returned, otherwise what is returned is the suffix of sequence which begins with the matching element.
(rmemq object list)
(rmemql object list)
(rmemqual object list)
(rmember key sequence [testfun [keyfun]])
(rmember-if predfun sequence [keyfun])
These functions are counterparts to memq, memql, memqual, member and member-if which look for the rightmost element which matches object, rather than for the leftmost element.
(conses list)
(conses* list)
These functions return a list whose elements are the conses which make up list. The conses* function does this in a lazy way, avoiding the computation of the entire list: it returns a lazy list of the conses of list. The conses function computes the entire list before returning.
The input list may be proper or improper.
The first cons of list is that list itself. The second cons is the rest of the list, or (cdr list). The third cons is (cdr (cdr list)) and so on.
(conses '(1 2 3)) -> ((1 2 3) (2 3) (3))
These functions are useful for simulating the maplist function found in other dialects like Common Lisp.
TXR Lisp's (conses x) can be expressed in Common Lisp as (maplist #'identity x).
Conversely, the Common Lisp operation (maplist function list) can be computed in TXR Lisp as (mapcar function (conses list)).
More generally, the Common Lisp operation
(maplist function list0 list1 ... listn)
can be expressed as:
(mapcar function (conses list0)
(conses list1) ... (conses listn))
(delcons cons list)
The delcons function destructively removes a cons cell from a list. The list is searched to see whether one of its cons cells is the same object as cons. If so, that cell is removed from the list.
The list argument may be a proper or improper list, possibly empty. It may also be an atom other than nil, which is regarded as being, effectively, an empty improper list terminated by that atom.
The operation of delcons is divided into the following three cases. If cons is the first cons cell of list, then the cdr of list is returned. If cons is the second or subsequent cons of list, then list is destructively altered to remove cons and then returned. This means that the cdr field of the predecessor of cons is altered from referencing cons to referencing (cdr cons) instead. The returned value is the same cons cell as list. The third case occurs when cons is not found in list. In this situation, list is returned unchanged.
(let ((x (list 1 2 3)))
(delcons x x))
-> (2 3)
(let ((x (list 1 2 . 3)))
(delcons (cdr x) x))
-> (1 . 3)
Association lists are ordinary lists formed according to a special convention. Firstly, any empty list is a valid association list. A nonempty association list contains only cons cells as the key elements. These cons cells are understood to represent key/value associations, hence the name "association list".
(assoc key alist)
The assoc function searches an association list alist for a cons cell whose car field is equivalent to key under the equal function. The first such cons is returned. If no such cons is found, nil is returned.
(assq key alist)
(assql key alist)
The assq and assql functions are very similar to assoc, with the only difference being that they determine equality using, respectively, the eq and eql functions rather than equal.
(rassq value alist)
(rassql value alist)
(rassoc value alist)
The rassq, rassql and rassoc functions are reverse lookup counterparts to assql and assoc. When searching, they examine the cdr field of the pairs of alist rather than the car field.
The rassoc function searches association list alist for a cons whose cdr field equivalent to value according to the equal function. If such a cons is found, it is returned. Otherwise nil is returned.
The rassq and rassql functions search in the same way as rassoc but compares values using, respectively, eq and eql.
(acons car cdr alist)
The acons function constructs a new alist by consing a new cons to the front of alist. The following equivalence holds:
(acons car cdr alist) <--> (cons (cons car cdr) alist)
(acons-new car cdr alist)
The acons-new function searches alist, as if using the assoc function, for an existing cell which matches the key provided by the car argument. If such a cell exists, then its cdr field is overwritten with the cdr argument, and then the alist is returned. If no such cell exists, then a new list is returned by adding a new cell to the input list consisting of the car and cdr values, as if by the acons function.
(aconsql-new car cdr alist)
The aconsql-new function has similar same parameters and semantics as acons-new, except that the eql function is used for equality testing. Thus, the list is searched for an existing cell as if using the assql function rather than assoc.
(alist-remove alist key*)
The alist-remove function takes association list alist and produces a duplicate from which cells matching any of the specified keys have been removed.
(alist-nremove alist key*)
The alist-nremove function is like alist-remove, but potentially destructive. The input list alist may be destroyed and its structural material reused to form the output list. The application should not retain references to the input list.
(copy-alist alist)
The copy-alist function duplicates alist. Unlike copy-list, which only duplicates list structure, copy-alist also duplicates each cons cell of the input alist. That is to say, each element of the output list is produced as if by the copy-cons function applied to the corresponding element of the input list.
(pairlis keys values [alist])
The pairlis function returns an association list consisting of pairs formed from the elements of keys and values prepended to the existing alist.
If an alist argument is omitted, it defaults to nil.
Pairs of elements are formed by taking successive elements from the keys and values sequences in parallel.
If the sequences are not of equal length, the excess elements from the longer sequence are ignored.
The pairs appear in the resulting list in the original order in which their constituents appeared in keys and values.
The ANSI CL pairlis requires key and data to be lists, not sequences. The behavior of the ANSI CL pairlis is undefined of those lists are of different lengths. Finally, the elements are permitted to appear in either the original order or reverse order.
(pairlis nil nil) -> nil
(pairlis "abc" #(1 2 3 4)) -> ((#\a . 1) (#\b . 2) (#\c . 3))
(pairlis '(1 2 3) '(a b c) '((x . y) (z . w)))
-> ((1 . a) (2 . b) (3 . c) (x . y) (z . w))
A property list, also referred to as a plist, is a flat list of even length consisting of interleaved pairs of property names (usually symbols) and their values (arbitrary objects). An example property list is (:a 1 :b "two") which contains two properties, :a having value 1, and :b having value "two".
An improper plist represents Boolean properties in a condensed way, as property indicators which are not followed by a value. Such properties only indicate their presence or absence, which is useful for encoding a Boolean value. If it is absent, then the property is false. Correctly using an improper plist requires that the exact set of Boolean keys is established by convention.
In this document, the unqualified terms property list and plist refer strictly to an ordinary plist, not to an improper plist.
Unlike in some other Lisp dialects, including ANSI Common Lisp, symbols do not have property lists in TXR Lisp. Improper plists aren't a concept in ANSI CL.
(prop plist key)
The prop function searches property list plist for key key. If the key is found, then the value next to it is returned. Otherwise nil is returned.
It is ambiguous whether nil is returned due to the property not being found, or due to the property being present with a nil value.
The indicators in plist are compared with key using eq equality, allowing them to be symbols, characters or fixnum integers.
(memp key plist)
The memp function searches property list plist for key key, using eq equality.
If the key is found, then the entire suffix of plist beginning with the indicator is returned, such that the first element of the returned list is key and the second element is the property value.
Note the reversed argument convention relative to the prop function, harmonizing with functions in the member family.
(plist-to-alist plist)
(improper-plist-to-alist imp-plist bool-keys)
The functions plist-to-alist and improper-plist-to-alist convert, respectively, a property list and improper property list to an association list.
The plist-to-alist function scans plist and returns the indicator-property pairs as a list of cons cells, such that each car is the indicator, and each cdr is the value.
The improper-plist-to-alist is similar, except that it handles the Boolean properties which, by convention, aren't followed by a value. The list of all such indicators is specified by the bool-keys argument.
(plist-to-alist '(a 1 b 2)) --> ((a . 1) (b . 2))
(improper-plist-to-alist '(:x 1 :blue :y 2) '(:blue))
--> ((:x . 1) (:blue) (:y . 2))
Note: these functions operate on lists. The principal sorting function in TXR Lisp is sort, described under Sequence Manipulation.
The merge function described here provides access to an elementary step of the algorithm used internally by sort when operating on lists.
The multi-sort operation sorts multiple lists in parallel. It is implemented using sort.
(merge seq1 seq2 [lessfun [keyfun]])
The merge function merges two sorted sequences seq1 and seq2 into a single sorted sequence. The semantics and defaulting behavior of the lessfun and keyfun arguments are the same as those of the sort function.
The sequence which is returned is of the same kind as seq1.
This function is destructive of any inputs that are lists. If the output is a list, it is formed out of the structure of the input lists.
(multi-sort columns less-funcs [key-funcs])
The multi-sort function regards a list of lists to be the columns of a database. The corresponding elements from each list constitute a record. These records are to be sorted, producing a new list of lists.
The columns argument supplies the list of lists which comprise the columns of the database. The lists should ideally be of the same length. If the lists are of different lengths, then the shortest list is taken to be the length of the database. Excess elements in the longer lists are ignored, and do not appear in the sorted output.
The less-funcs argument supplies a list of comparison functions which are applied to the columns. Successive functions correspond to successive columns. If less-funcs is an empty list, then the sorted database will emerge in the original order. If less-funcs contains exactly one function, then the rows of the database are sorted according to the first column. The remaining columns simply follow their row. If less-funcs contains more than one function, then additional columns are taken into consideration if the items in the previous columns compare equal. For instance if two elements from column one compare equal, then the corresponding second column elements are compared using the second column comparison function. The less-funcs argument may be a function object, in which case it is treated as if it were a one-element list containing that function object.
The optional key-funcs argument supplies transformation functions through which column entries are converted to comparison keys, similarly to the single key function used in the sort function and others. If there are more key functions than less functions, the excess key functions are ignored.
(make-lazy-cons function [car [cdr]])
The function make-lazy-cons makes a special kind of cons cell called a lazy cons, whose type is lcons. Lazy conses are useful for implementing lazy lists.
Lazy lists are lists which are not allocated all at once. Rather, the elements of its structure materialize just before they are accessed.
A lazy cons has car and cdr fields like a regular cons, and those fields are initialized to the values of the car and cdr arguments of make-lazy-cons when the lazy cons is created. These arguments default to nil if omitted. A lazy cons also has an update function, which is specified by the function argument to make-lazy-cons.
The function argument must be a function that may be called with exactly one parameter.
When either the car and cdr fields of a cons are accessed for the first time to retrieve their value, function is automatically invoked first, and is given the lazy cons as a parameter. That function has the opportunity to store new values into the car and cdr fields. Once the function is called, it is removed from the lazy cons: the lazy cons no longer has an update function. If the update function itself attempts to retrieve the value of the lazy cons cell's car or cdr field, it will be recursively invoked.
The functions lcons-car and lcons-cdr may be used to access the fields of a lazy cons without triggering the update function.
Storing a value into either the car or cdr field does not have the effect of invoking the update function.
If the function terminates by returning normally, the access to the value of the field then proceeds in the ordinary manner, retrieving whatever value has most recently been stored.
The return value of the function is ignored.
To perpetuate the growth of a lazy list, the function can make another call to make-lazy-cons and install the resulting cons as the cdr of the lazy cons.
;;; lazy list of integers between min and max
(defun integer-range (min max)
(let ((counter min))
;; min is greater than max; just return empty list,
;; otherwise return a lazy list
(if (> min max)
nil
(make-lazy-cons
(lambda (lcons)
;; install next number into car
(rplaca lcons counter)
;; now deal wit cdr field
(cond
;; max reached, terminate list with nil!
((eql counter max)
(rplacd lcons nil))
;; max not reached: increment counter
;; and extend with another lazy cons
(t
(inc counter)
(rplacd lcons
(make-lazy-cons
(lcons-fun lcons))))))))))
(lconsp value)
The lconsp function returns t if value is a lazy cons cell. Otherwise it returns nil, even if value is an ordinary cons cell.
(lcons-fun lazy-cons)
The lcons-fun function retrieves the update function of a lazy cons. Once a lazy cons has been accessed, it no longer has an update function and lcons-fun returns nil. While the update function of a lazy cons is executing, it is still accessible. This allows the update function to retrieve a reference to itself and propagate itself into another lazy cons (as in the example under make-lazy-cons).
(lcons-car lazy-cons)
(lcons-cdr lazy-cons)
The functions lcons-car and lcons-cdr retrieve the car and cdr fields of lazy-cons, without triggering the invocation of its associated update function.
The lazy-cons argument must be an object of type lcons. Unlike the functions car and cdr, These functions cannot be applied to any other type of object.
Note: these functions may be used by the update function to retrieve the values which were stored into lazy-cons by the make-lazy-cons constructor, without triggering recursion. The function may then overwrite either or both of these values. This allows the fields of the lazy cons to store state information necessary for the propagation of a lazy list. If that state information consists of no more than two values, then no additional context object need be allocated.
(lcons-force object)
The lcons-function recursively forces a lazy cons.
If the argument object is of lcons type, and has not been previously forced, then it is forced. The associated lazy function is invoked. Then, lcons-force is recursively invoked on the car and cdr fields of the lazy cons.
The lcons-force function returns its argument.
(lcons car-expression cdr-expression)
The lcons macro simplifies the construction of structures based on lazy conses. Syntactically, it resembles the cons function. However, the arguments are expressions rather than values. The macro generates code which, when evaluated, immediately produces a lazy cons. The expressions car-expression and cdr-expression are not immediately evaluated. Rather, when either the car or cdr field of the lazy cons cell is accessed, these expressions are both evaluated at that time, in the order that they appear in the lcons expression, and in the original lexical scope in which that expression was evaluated. The return values of these expressions are used, respectively, to initialize the corresponding fields of the lazy cons.
Note: the lcons macro may be understood in terms of the following reference implementation, as a syntactic sugar combining the make-lazy-cons constructor with a lexical closure provided by a lambda function:
(defmacro lcons (car-form cdr-form)
(let ((lc (gensym)))
^(make-lazy-cons (lambda (,lc)
(rplaca ,lc ,car-form)
(rplacd ,lc ,cdr-form)))))
;; Given the following function ...
(defun fib-generator (a b)
(lcons a (fib-generator b (+ a b))))
;; ... the following function call generates the Fibonacci
;; sequence as an infinite lazy list.
(fib-generator 1 1) -> (1 1 2 3 5 8 13 ...)
(lazy-stream-cons stream [no-throw-close-p])
(get-lines [stream [no-throw-close-p]])
The lazy-stream-cons and get-lines functions are synonyms, except that the stream argument is optional in get-lines and defaults to *stdin*. Thus, the following description of lazy-stream-cons also applies to get-lines.
The lazy-stream-cons returns a lazy cons which generates a lazy list based on reading lines of text from input stream stream, which form the elements of the list. The get-line function is called on demand to add elements to the list.
The lazy-stream-cons function itself makes the first call to get-line on the stream. If this returns nil, then the stream is closed and nil is returned. Otherwise, a lazy cons is returned whose update function will install that line into the car field of the lazy cons, and continue the lazy list by making another call to lazy-stream-cons, installing the result into the cdr field. When this lazy list obtains an end-of-file indication from the stream, it closes the stream.
lazy-stream-cons inspects the real-time property of a stream as if by the real-time-stream-p function. This determines which of two styles of lazy list are returned. For an ordinary (non-real-time) stream, the lazy list treats the end-of-file condition accurately: an empty file turns into the empty list nil, a one line file into a one-element list which contains that line and so on. This accuracy requires one line of lookahead which is not acceptable in real-time streams, and so a different type of lazy list is used, which generates an extra nil item after the last line. Under this type of lazy list, an empty input stream translates to the list (nil); a one-line stream translates to ("line" nil) and so forth.
If and when stream is closed by the function directly, or else by the returned lazy list, the no-throw-close-p Boolean argument, defaulting to nil, controls the throw-on-error-p argument of the call to the close-stream function. These arguments have opposite polarity: if no-throw-close-p is true, then throw-on-error-p shall be false, and vice versa.
Note: the lcons-force function may be used on the return value of get-lines to force the lazy list.
(close-lazy-streams body-form*)
The close-lazy-streams macro establishes a dynamic environment in which zero or more body-forms are evaluated, yielding the value of the last body-form, or else nil if there are no body-form arguments. In this regard, the macro operator resembles progn.
The environment established by close-lazy-streams sets up special monitoring of the the functions lazy-stream-cons and get-lines. Whenever these functions register an I/O stream with a lazy list, in the dynamic scope of this environment, that stream is recorded in a hidden list associated with the innermost enclosing close-lazy-streams form. When the form terminates, it invokes close-stream on each stream in the hidden list.
Note: the close-lazy-streams macro provides a possible solution for situations in which a body of code, possibly consisting of nested functions, manipulates lazy lists of lines coming from from I/O streams, such that these lists are not completely forced. Incompletely processed lazy lists will not close their associated streams until they are reclaimed by garbage collection, which could cause the application to run out of file descriptors. The close-lazy-streams macro allows the application to delineate a dynamic contour of code upon whose termination all such stream associations generated within that contour will be duly cleaned up.
Collect list of names of .tl files which contain the string "(cons ":
;; Incorrect version: could run out of open files if there are many
;; files which contain a match processed, because find-if will stop
;; traversing the list of lines when it finds a match:
(build
(each ((file (glob "*.tl")))
(if (find-if #/\(cons / (file-get-lines file))
(add file))))
;; Addressed with close-lazy-streams: after each iteration, the
;; stream created by file-get-lines is closed.
(build
(each ((file (glob "*.tl")))
(close-lazy-streams
(if (find-if #/\(cons / (file-get-lines file))
(add file)))))
(delay expression)
The delay operator arranges for the delayed (or "lazy") evaluation of expression. This means that the expression is not evaluated immediately. Rather, the delay expression produces a promise object.
The promise object can later be passed to the force function (described later in this document). The force function will trigger the evaluation of the expression and retrieve the value.
The expression is evaluated in the original scope, no matter where the force takes place.
The expression is evaluated at most once, by the first call to force. Additional calls to force only retrieve a cached value.
;; list is popped only once: the value is computed
;; just once when force is called on a given promise
;; for the first time.
(defun get-it (promise)
(format t "*list* is ~s\n" *list*)
(format t "item is ~s\n" (force promise))
(format t "item is ~s\n" (force promise))
(format t "*list* is ~s\n" *list*))
(defvar *list* '(1 2 3))
(get-it (delay (pop *list*)))
Output:
*list* is (1 2 3)
item is 1
item is 1
*list* is (2 3)
(force promise)
(set (force promise) new-value)
The force function accepts a promise object produced by the delay macro. The first time force is invoked, the expression which was wrapped inside promise by the delay macro is evaluated (in its original lexical environment, regardless of where in the program the force call takes place). The value of expression is cached inside promise and returned, becoming the return value of the force function call. If the force function is invoked additional times on the same promise, the cached value is retrieved.
A force form is a syntactic place, denoting the value cache location within promise.
Storing a value in a force place causes future accesses to the promise to return that value.
If the promise had not yet been forced, then storing a value into it prevents that from ever happening. The delayed expression will never be evaluated.
If, while a promise is being forced, the evaluation of expression itself causes an assignment to the promise, it is not specified whether the promise will take on the value of expression or the assigned value.
(promisep object)
The promisep function returns t if object is a promise object: an object created by the delay macro. Otherwise it returns nil.
Note: promise objects are conses. The typeof function applied to a promise returns cons.
(mlet ({sym | (sym init-form)}*) body-form*)
The mlet macro ("magic let" or "mutual let") implements a variable binding construct similar to let and let*.
Under mlet, the scope of the bindings of the sym variables extends over the init-forms, as well as the body-forms.
Unlike the let* construct, each init-form has each sym in scope. That is to say, an init-form can refer not only to previous variables, but also to later variables as well as to its own variable.
The variables are not initialized until their values are accessed for the first time. Any sym whose value is not accessed is not initialized.
Furthermore, the evaluation of each init-form does not take place until the time when its value is needed to initialize the associated sym. This evaluation takes place once. If a given sym is not accessed during the evaluation of the mlet construct, then its init-form is never evaluated.
The bound variables may be assigned. If, before initialization, a variable is updated in such a way that its prior value is not needed, it is unspecified whether initialization takes place, and thus whether its init-form is evaluated.
Direct circular references are erroneous and are diagnosed. This takes place when the macro-expanded form is evaluated, not during the expansion of mlet.
;; Dependent calculations in arbitrary order
(mlet ((x (+ y 3))
(z (+ x 1))
(y 4))
(+ z 4)) --> 12
;; Error: circular reference:
;; x depends on y, y on z, but z on x again.
(mlet ((x (+ y 1))
(y (+ z 1))
(z (+ x 1)))
z)
;; Okay: lazy circular reference because lcons is used
(mlet ((list (lcons 1 list)))
list) --> (1 1 1 1 1 ...) ;; circular list
In the last example, the list variable is accessed for the first time in the body of the mlet form. This causes the evaluation of the lcons form. This form evaluates its arguments lazily, which means that it is not a problem that list is not yet initialized. The form produces a lazy cons, which is then used to initialize list. When the car or cdr fields of the lazy cons are accessed, the list expression in the lcons argument is accessed. By that time, the variable is initialized and holds the lazy cons itself, which creates the circular reference, and a circular list.
(generate while-fun gen-fun)
(giterate while-fun gen-fun [value])
(ginterate while-fun gen-fun [value])
The generate function produces a lazy list which dynamically produces items according to the following logic.
The arguments to generate are functions which do not take any arguments. The return value of generate is a lazy list.
When the lazy list is accessed, for instance with the functions car and cdr, it produces items on demand. Prior to producing each item, while-fun is called. If it returns a true Boolean value (any value other than nil), then the gen-fun function is called, and its return value is incorporated as the next item of the lazy list. But if while-fun yields nil, then the lazy list immediately terminates.
Prior to returning the lazy list, generate invokes the while-fun one time. If while-fun yields nil, then generate returns the empty list nil instead of a lazy list. Otherwise, it instantiates a lazy list, and invokes the gen-fun to populate it with the first item.
The giterate function is similar to generate, except that while-fun and gen-fun are functions of one argument rather than functions of no arguments. The optional value argument defaults to nil and is threaded through the function calls. That is to say, the lazy list returned is (value [gen-fun value] [gen-fun [gen-fun value]] ...).
The lazy list terminates when a value fails to satisfy while-fun. That is to say, prior to generating each value, the lazy list tests the value using while-fun. If that function returns nil, then the item is not added, and the sequence terminates.
Note: giterate could be written in terms of generate like this:
(defun giterate (w g v)
(generate (lambda () [w v])
(lambda () (prog1 v (set v [g v])))))
The ginterate function is a variant of giterate which includes the test-failing item in the generated sequence. That is to say ginterate generates the next value and adds it to the lazy list. The value is then tested using while-fun. If that function returns nil, then the list is terminated, and no more items are produced.
(giterate (op > 5) (op + 1) 0) -> (0 1 2 3 4)
(ginterate (op > 5) (op + 1) 0) -> (0 1 2 3 4 5)
(expand-right gen-fun value)
The expand-right function is a complement to reduce-right, with lazy semantics.
The gen-fun parameter is a function, which must accept a single argument, and return either a cons pair or nil.
The value parameter is any value.
The first call to gen-fun receives value.
The return value is interpreted as follows. If gen-fun returns a cons-cell pair (elem . next) then elem specifies the element to be added to the lazy list, and next specifies the value to be passed to the next call to gen-fun. If gen-fun returns nil then the lazy list ends.
;; Count down from 5 to 1 using explicit lambda
;; for gen-fun:
(expand-right
(lambda (item)
(if (zerop item) nil
(cons item (pred item))))
5)
--> (5 4 3 2 1)
;; Using functional combinators:
[expand-right [iff zerop nilf [callf cons identity pred]] 5]
--> (5 4 3 2 1)
;; Include zero:
[expand-right
[iff null
nilf
[callf cons identity [iff zerop nilf pred]]] 5]
--> (5 4 3 2 1 0)
(expand-left gen-fun value)
(nexpand-left gen-fun value)
The expand-left function is a companion to expand-right.
Unlike expand-right, it has eager semantics: it calls gen-fun repeatedly and accumulates an output list, not returning until gen-fun returns nil.
The semantics is as follows. expand-left initializes an empty accumulation list. Then gen-fun is called, with value as its argument.
If gen-fun it returns a cons cell, then the car of that cons cell is pushed onto the accumulation list, and the procedure is repeated: gen-fun is called again, with cdr taking the place of value.
If gen-fun returns nil, then the accumulation list is returned.
If the expression (expand-right f v) produces a terminating list, then the following equivalence holds:
(expand-left f v) <--> (reverse (expand-right f v))
The equivalence cannot hold for arguments to expand-left which produce an infinite list.
The nexpand-left function is a destructive version of expand-left.
The list returned by nexpand-left is composed of the cons cells returned by gen-fun whereas the list returned by expand-left is composed of freshly allocated cons cells.
(repeat list [count])
If list is empty, then repeat returns an empty list.
If count is omitted, the repeat function produces an infinite lazy list formed by catenating together copies of list.
If count is specified and is zero or negative, then an empty list is returned.
Otherwise a list is returned consisting of count repetitions of list catenated together.
(pad sequence [object [count]])
The pad function produces a lazy list which consists of all of the elements of sequence followed by repetitions of object.
If object is omitted, it defaults to nil.
If count is omitted, then the repetition of object is infinite. Otherwise the specified number of repetitions occur.
Note that sequence may be a lazy list which is infinite. In that case, the repetitions of object will never occur.
(weave {sequence}*)
The weave function interleaves elements from the sequences given as arguments.
If called with no arguments, it returns the empty list.
If called with a single sequence, it returns the elements of that sequence as a new lazy list.
When called with two or more sequences, weave returns a lazy list which draws elements from the sequences in a round-robin fashion, repeatedly scanning the sequences from left to right, and taking an item from each one, removing it from the sequence. Whenever a sequence runs out of items, it is deleted; the weaving then continues with the remaining sequences. The weaved sequence terminates when all sequences are eliminated. (If at least one of the sequences is an infinite lazy list, then the weaved sequence is infinite.)
;; Weave negative integers with positive ones:
(weave (range 1) (range -1 : -1)) -> (1 -1 2 -2 3 -3 ...)
(weave "abcd" (range 1 3) '(x x x x x x x))
--> (#\a 1 x #\b 2 x #\c 3 x #\d x x x x)
(gen while-expression produce-item-expression)
(gun produce-item-expression)
The gen macro operator produces a lazy list, in a manner similar to the generate function. Whereas the generate function takes functional arguments, the gen operator takes two expressions, which is often more convenient.
The return value of gen is a lazy list. When the lazy list is accessed, for instance with the functions car and cdr, it produces items on demand. Prior to producing each item, the while-expression is evaluated, in its original lexical scope. If the expression yields a non-nil value, then produce-item-expression is evaluated, and its return value is incorporated as the next item of the lazy list. If the expression yields nil, then the lazy list immediately terminates.
The gen operator itself immediately evaluates while-expression before producing the lazy list. If the expression yields nil, then the operator returns the empty list nil. Otherwise, it instantiates the lazy list and invokes the produce-item-expression to force the first item.
The gun macro similarly creates a lazy list according to the following rules. Each successive item of the lazy list is obtained as a result of evaluating produce-item-expression. However, when produce-item-expression yields nil, then the list terminates (without adding that nil as an item).
Note 1: the form gun can be implemented as a macro-expanding to an instance of the gen operator, like this:
(defmacro gun (expr)
(let ((var (gensym)))
^(let (,var)
(gen (set ,var ,expr)
,var))))
This exploits the fact that the set operator returns the value that is assigned, so the set expression is tested as a condition by gen, while having the side effect of storing the next item temporarily in a hidden variable.
In turn, gen can be implemented as a macro expanding to some lambda functions which are passed to the generate function:
(defmacro gen (while-expr produce-expr)
^(generate (lambda () ,while-expr)
(lambda () ,produce-expr)))
Note 2: gen can be considered as an acronym for Generate, testing Expression before Next item, whereas gun stands for Generate Until Null.
;; Make a lazy list of integers up to 1000
;; access and print the first three.
(let* ((counter 0)
(list (gen (< counter 1000) (inc counter))))
(format t "~s ~s ~s\n" (pop list) (pop list) (pop list)))
Output:
1 2 3
(range [from [to [step]]])
(range* [from [to [step]]])
The range and range* functions generate a lazy, potentially infinite list, according to several disciplines.
There is a major division in behavior depending on whether or not the from argument, which specifies the initial item, is an arithmetic type according to the arithp function. The following remarks describe the arithmetic case. A description of the non-arithmetic behavior follows.
The difference between and range* is that range* excludes the endpoint. For instance (range 0 3) generates the list (0 1 2 3), whereas (range* 0 3) generates (0 1 2).
All arguments are optional. If the step argument is omitted, then it defaults to 1 if the to argument is omitted, or else if it is greater than or equal to from according to the > function. If to is given, and is less than from, then a missing step argument defaults to -1.
Each value in the list is obtained from the previous by adding the step value. Positive or negative step values are allowed. There is no check for a step size of zero, or for a step direction which cannot meet the endpoint.
The step argument may be a function. The function must accept one argument. That argument is an element of the list, from which the function calculates the next element.
The to argument specifies the endpoint value, which, if it occurs in the list, is excluded from it by the range* function, but included by the range function. If to is missing, or specified as nil, then there is no endpoint, and the list which is generated is infinite, regardless of step.
If from is omitted, then the list begins at zero, otherwise from must be an arithmetic object which specifies the initial value.
The list stops if it reaches the endpoint value (which is included in the case of range, and excluded in the case of range*). However, depending on the arguments, it is possible that the generated list doesn't contain the endpoint value, yet steps over it. This occurs when the previous value of the list is less than the endpoint value, but the next value is greater, or vice versa. In this situation, the list also stops, and the excess value which surpasses the endpoint is excluded from the list.
The rest of the description applies to the case when the from argument is a non-arithmetic type.
In the non-arithmetic case, the step argument unconditionally defaults to 1. If it is given, it must either be a function, or else a positive integer.
If step is a function, that function is used to determine each successive value from the previous similarly to the arithmetic case. If the to value is omitted, an infinite list is generated this way. If the to argument is present, the list stops if it attains the endpoint value. No provision is made for the endpoint value being skipped, like in the arithmetic case. When the endpoint value is reached, range* function omits that value from the list.
If step is a positive integer, then range iteration is used. A range value is constructed from the from and to arguments as if by the (rcons* from to) expression. Here, the to argument defaults to nil if it is missing. An iterator is created for the resulting range object as if by iter-begin and this iterator is then used to obtain values for the lazy list returned by range or range*. The list ends when the iterator indicates that no more items are available. In the case of the range* function, the last value produced by the iterator is omitted from the list. The step size is used to skip items from the iterator. For instance, if the value is 3, then the sequence begins with the from value. The next two values from the sequence are omitted, The fourth item from the sequence is included in the list, (unless there either is no such item, or the function is range*, and that item is the last one).
(range 1 1) -> (1)
(range 0 4) -> (0 1 2 3 4)
(range 4 0) -> (4 3 2 1 0)
(range 0.0 2.0 0.5) (0.0 0.5 1.0 1.5 2.0)
(range #R(0 1) #R(3 4)) (#R(0 1) #R(1 2) #R(3 4))
(range 0 4 2) -> (0 2 4)
(range #\a #\e 2) (#\a #\c #\e)
(range 1 32 (op * 2)) -> (1 2 4 8 16 32))
(range* 1 1) -> nil
(range* 0 4) -> (0 1 2 3)
(range* 4 0 -2) -> (4 2)
(range 0 1.25 0.5) -> (0 0.5 1.0)
(range* 0 1.25 0.5) -> (0 0.5 1.0))
(range "A" "A") -> nil
(range "AA" "BC") -> ("AA" "AB" "AC" "BA" "BB" "BC")
(range "AA" "BC" 2) -> ("AA" "AC" "BB")
[range* "ABCD" nil rest] -> ("ABCD" "BCD" "CD" "D")
(rlist item*)
(rlist* item*)
The rlist ("range list") function is useful for producing a list consisting of a mixture of discontinuous numeric or character ranges and individual items.
The function returns a lazy list of elements. The items are produced by converting the function's successive item arguments into lists, which are lazily catenated together to form the output list.
Each item is transformed into a list as follows. Any item which is not a range object is trivially turned into a one-element list as if by the (list item*) expression.
Any item which is a range object, whose to field isn't a range is turned into a lazy list as if by evaluating the (range (from item)(to item)) expression. Thus for instance the argument 1..10 turns into the (lazy) list (1 2 3 4 5 6 7 8 9 10).
Any item which is a range object such that its to field is also a range is turned into a lazy list as if by evaluating the (range (from item)(from (to item))(to (to item))) expression. Thus for instance the argument expression 1..10..2 produces an item which rlist turns into the lazy list (1 3 5 7 9) as if by the call (range 1 10 2). Note that the expression 1..10..2 stands for the expression (rcons 1 (rcons 10 2)) which evaluates to #R(1 #R(10 2)).
The #R(1 #R(10 2)) range literal syntax can be passed as an argument to rlist with the same result as 1..10..2.
The rlist* function differs from rlist in one regard: under rlist*, the ranges denoted by the range notation exclude the endpoint. That is, the ranges are generated as if by the range* function rather than range.
Note: it is permissible for item objects to specify infinite ranges. It is also permissible to apply rlist to an infinite argument list.
(rlist 1 "two" :three) -> (1 "two" :three)
(rlist 10 15..16 #\a..#\d 2) -> (10 15 16 #\a #\b #\c #\d 2)
(take 7 (rlist 1 2 5..:)) -> (1 2 5 6 7 8 9)
Ranges are objects that aggregate two values, not unlike cons cells. However, they are atoms, and are primarily intended to hold numeric or character values in their two fields. These fields are called from and to which are the names of the functions which access them. These fields are not mutable; a new value cannot be stored into either field of a range.
The printed notation for a range object consists of the prefix #R (hash R) followed by the two values expressed as a two-element list. Ranges can be constructed using the rcons function. The notation x..y corresponds to (rcons x y).
Ranges behave as a numeric type and support a subset of the numeric operations. Two ranges can be added or subtracted, which obeys these equivalences:
(+ a..b c..d) <--> (+ a c)..(+ b d)
(- a..b c..d) <--> (- a c)..(- b d)
A range a..b can be combined with a character or number n using addition or subtractions, which obeys these equivalences:
(+ a..b n) <--> (+ n a..b) <--> (+ a n)..(+ b n)
(- a..b n) <--> (- a n)..(- b n)
(- n a..b) <--> (- n a)..(- n b)
A range can be multiplied by a number:
(* a..b n) <--> (* n a..b) <--> (* a n)..(* b n)
A range can be divided by a number using the / or trunc functions, but a number cannot be divided by a range:
(trunc a..b n) <--> (trunc a n)..(trunc b n)
(/ a..b n) <--> (/ a n)..(/ b n)
Ranges can be compared using the equality and inequality functions =, <, >, <= and >=. Equality obeys this equivalence:
(= a..b c..d) <--> (and (= a c) (= b d))
Inequality comparisons treat the from component with precedence over to such that only if the from components of the two ranges are not equal under the = function, then the inequality is based solely on them. If they are equal, then the inequality is based on the to components. This gives rise to the following equivalences:
(< a..b c..d) <--> (if (= a c) (< b d) (< a c))
(> a..b c..d) <--> (if (= a c) (> b d) (> a c))
(>= a..b c..d) <--> (if (= a c) (>= b d) (> a c))
(<= a..b c..d) <--> (if (= a c) (<= b d) (< a c))
Ranges can be negated with the one-argument form of the - function, which is equivalent to subtraction from zero: the negation distributes over the two range components.
The abs function also applies to ranges and distributes into their components.
The succ and pred family of functions also operate on ranges.
The length of a range may be obtained with the length function;
The length of the range a..b is defined as (- b a), and may be obtained using the length function. The empty function accepts ranges and tests them for zero length.
(rcons from to)
The rcons function constructs a range object which holds the values from and to.
Though range objects are effectively binary cells like conses, they are atoms. They also aren't considered sequences, nor are they structures.
Range objects are used for indicating numeric ranges, such as substrings of lists, arrays and strings. The dotdot notation serves as a syntactic sugar for rcons. The syntax a..b denotes the expression (rcons a b).
Note that ranges are immutable, meaning that it is not possible to replace the values in a range.
(rangep value)
The rangep function returns t if value is a range. Otherwise it returns nil.
(from range)
(to range)
The from and to functions retrieve, respectively, the from and to fields of a range.
Note that these functions are not accessors, which is because ranges are immutable.
(in-range range value)
(in-range* range value)
The in-range and in-range* functions test whether the value argument lies in the range represented by the range argument, indicating the Boolean result using one of the values t or nil.
The range argument must be a range object.
It is expected that the range object's from value does not exceed the to value; a reversed range is considered empty.
The in-range* function differs from in-range in that it excludes the upper endpoint.
The implicit comparison against the range endpoints is performed using the less and lequal functions, as appropriate.
The following equivalences hold:
(in-range r x) <--> (and (lequal (from r) x)
(lequal x (to r)))
(in-range* r x) <--> (and (lequal (from r) x)
(less x (to r)))
(rangeref range [idx | seq])
The rangeref function requires its range argument to be a range object.
It supports two semantics, based on the type of the second argument.
If the second argument is an integer, then it is interpreted as idx. The function then treats the range as if it were a sequence. The range must be a numeric or character range. The from field of range is added to idx to form the tentative return value.
If the to field is a value other than t or the : (colon) symbol, then the tentative value must be less than the value of this field, or an exception is thrown. In other words, ind must indicate a point within the range.
After the above range check is performed, if applicable, the tentative value is returned.
If the second argument isn't an integer, it is interpreted as a sequence seq. The range object's values are used to extract a subrange of seq, according to the following equivalence:
(rangeref r s) <--> (sub s (from r) (to r))
except that r and s are evaluated only once, in that order.
(mkstring length [char])
(str length [char | string])
The mkstring function constructs a string object of a length specified by the length parameter. The length parameter must be non-negative. Every position in the string is initialized with char, which must be a character value.
If the optional argument char is not specified, it defaults to the space character.
The str function resembles mkstring, and behaves the same way when the second argument is omitted, and when it is a character value. The second argument of str may be a string, in which case the newly created string is filled by taking successive characters from string. If string is longer than length, its excess characters are ignored. If string is shorter, then characters are taken from the beginning again; string is effectively taken as a fill pattern to be repeated as many times as necessary to provide the required number of characters. If string is empty, str fills the new string with spaces.
(copy-str string)
The copy-str function constructs a new string whose contents are identical to string.
If string is a lazy string, then a lazy string is constructed with the same attributes as string. The new lazy string has its own copy of the prefix portion of string which has been forced so far. The unforced list and separator string are shared between string and the newly constructed lazy string.
(upcase-str string)
The upcase-str function produces a copy of string such that all lowercase characters of the English alphabet are mapped to their uppercase counterparts.
(downcase-str string)
The downcase-str function produces a copy of string such that all uppercase characters of the English alphabet are mapped to their lowercase counterparts.
(string-extend string tail [final])
The string-extend function destructively increases the length of string, which must be an ordinary dynamic string. It is an error to invoke this function on a literal string or a lazy string.
The tail argument can be a character, string or integer. If it is a string or character, it specifies material which is to be added to the end of the string: either a single character or a sequence of characters. If it is an integer, it specifies the number of characters to be added to the string.
If tail is an integer, the newly added characters have indeterminate contents. The string appears to be the original one because of an internal terminating null character remains in place, but the characters beyond the terminating zero are indeterminate.
The optional Boolean argument final, defaulting to nil, is a hint which indicates whether this string-extend call is expected to be the last time that the function is invoked on the given string. If final is true, then the string object's underlying memory allocation is trimmed to fit the actual string data. If the argument is false, the object may be given a larger allocation intended to improves the performance of subsequent string-extend calls.
(string-finish string)
The string-finish function removes excess allocation from string that may have been produced by previous calls to string-extend.
Note: if the most recent call to string string-extend specified a true value for the final parameter, then calling string-finish is unnecessary and does nothing.
(stringp obj)
The stringp function returns t if obj is one of the several kinds of strings. Otherwise it returns nil.
(length-str string)
The length-str function returns the length string in characters. The argument must be a string.
(coded-length string)
The coded-length function returns the number of bytes required to encode string in UTF-8.
The argument must be a character string.
If the string contains only characters in the ASCII range U+0001 to U+007F range, then the value returned shall be the same as that returned by the length-str function.
(search-str haystack needle [start [from-end]])
The search-str function finds an occurrence of the string needle inside the haystack string and returns its position. If no such occurrence exists, it returns nil.
If a start argument is not specified, it defaults to zero. If it is a nonnegative integer, it specifies the starting character position for the search. Negative values of start indicate positions from the end of the string, such that -1 is the last character of the string.
If the from-end argument is specified and is not nil, it means that the search is conducted right-to-left. If multiple matches are possible, it will find the rightmost one rather than the leftmost one.
(search-str-tree haystack tree [start [from-end]])
The search-str-tree function is similar to search-str, except that instead of searching haystack for the occurrence of a single needle string, it searches for the occurrence of numerous strings at the same time. These search strings are specified, via the tree argument, as an arbitrarily structured tree whose leaves are strings.
The function finds the earliest possible match, in the given search direction, from among all of the needle strings.
If tree is a single string, the semantics is equivalent to search-str.
(match-str bigstring littlestring [start])
Without the start argument, the match-str function determines whether littlestring is a prefix of bigstring.
If the start argument is specified, and is a nonnegative integer, then the function tests whether littlestring matches a prefix of that portion of bigstring which starts at the given position.
If the start argument is a negative integer, then match-str determines whether littlestring is a suffix of bigstring, ending on that position of bigstring, where -1 denotes the last character of bigstring, -2 the second last one and so on.
If start is -1, then this corresponds to testing whether littlestring is a suffix of bigstring.
The match-str function returns nil if there is no match.
If a prefix match is successful, then an integer value is returned indicating the position, inside bigstring, one character past the matching prefix. If the entire string is matched, then this value corresponds to the length of bigstring.
If a suffix match is successful, the return value is the position within bigstring where the leftmost character of littlestring matched.
(match-str-tree bigstring tree [start])
The match-str-tree function is a generalization of match-str which matches multiple test strings against bigstring at the same time. The value reported is the longest match from among any of the strings.
The strings are specified as an arbitrarily shaped tree structure which has strings at the leaves.
If tree is a single string atom, then the function behaves exactly like match-str.
(sub-str str [from [to]])
(set (sub-str str [from [to]]) new-value)
The sub-str function has the same parameters and semantics as the sub function, function, except that the first argument is operated upon using string operations.
If a sub-str form is used as a place, it denotes a subrange of list as if it were a storage location. The previous value of this location, if needed, is fetched by a call to sub-str. Storing new-value to the place is performed by a call to replace-str. In an update operation which accesses the prior value and stores a new value, the arguments str, from, to and new-value are evaluated once.
The str argument is not itself required to be a place; it is not updated when a value is written to the sub-str storage location.
(replace-str string item-sequence [from [to]])
The replace-str function has the same parameters and semantics as the replace function, except that the first argument is operated upon using string operations.
(cat-str item-seq [sep])
(join-with sep item*)
(join item*)
The cat-str, join-with and join functions combine items into a single string, which is returned.
Every item argument must be a character, string or else a possibly empty sequence of items. This rule applies recursively.
If a sep argument is present, it must be a character or string.
The item-seq argument must be a sequence of any mixture of items which are characters, strings or sequences of items. Note that this means that if item-seq is a character string, it is a valid argument, since it is a sequence of characters.
If item-seq is empty, or no item arguments are present, then all three functions return an empty string.
All three functions operate on an abstract sequence of character and string items, produced by a left-to-right recursive traversal of their item-seq or item arguments.
Under the join-with function, as well as the cat-str function a sep argument is given to it, the items are catenated together such that sep is interposed between them. If there are n character or string items, then n - 1 copies of sep occur in the resulting string, which is returned.
Under the join function, or cat-str function invoked without a sep argument, the items are catenated together directly, without any separator. The resulting string is returned.
(split-str string sep [keep-between [count]])
The split-str function breaks the string into pieces, returning a list thereof. The sep argument must be one of three types: a string, a character or a regular expression. It determines the separator character sequences within string.
The following describes the behavior of split-str in the case when the integer parameter count is omitted. The semantics of count are then given.
All non-overlapping matches for sep within string are identified in left-to-right order, and are removed from string. The string is broken into pieces according to the gaps left behind by the removed separators, and a list of the remaining pieces is returned.
If sep is the empty string, then the separator pieces removed from the string are considered to be the empty strings between its characters. In this case, if string is of length one or zero, then it is considered to have no such pieces, and a list of one element is returned containing the original string. These remarks also apply to the situation when sep is a regular expression which matches only an empty substring of string.
If a match for sep is not found in the string at all (not even an empty match), then the string is not split at all: a list of one element is returned containing the original string.
If sep matches the entire string, then a list of two empty strings is returned, except in the case that the original string is empty, in which case a list of one element is returned, containing the empty string.
Whenever two adjacent matches for sep occur, they are considered separate cuts with an empty piece between them.
This operation is nondestructive: string is not modified in any way.
If the optional keep-between argument is specified and is not nil, If an argument is given and is true, then split-str incorporates the matching separating pieces of string into the resulting list, such that if the resulting list is catenated, a string equivalent to the original string will be produced.
Note: to split a string into pieces of length one such that an empty string produces nil rather than (), use the (tok-str string #/./) pattern.
Note: the function call (split-str s r t) produces a resulting list identical to (tok-str s r t), for all values of r and s, provided that r does not match empty strings. If r matches empty strings, then the tok-str call returns extra elements compared to split-str, because tok-str allows empty matches to take place and extract empty tokens before the first character of the string, and after the last character, whereas split-str does not recognize empty separators at these outer limits of the string.
If the count parameter is present, it must be a non-negative integer. This value specifies the maximum number of pieces of the input string which are extracted by the splitting process. The returned list consists of these pieces, followed by the remainder of the string, if the remainder is nonempty. If keep-sep is true, then separators appear between the pieces, and if the remainder piece is present, the separator between the last piece and the remainder is included. If count is zero, then split-str returns a list of one element, which is string.
(spl sep [keep-between] string)
(spln count sep [keep-between] string)
The spl function performs the same computation as split-str. The same-named parameters of spl and split-str have the same semantics. The difference is the argument order. The spl function takes the sep argument first. The last argument is always string whether or not there are two arguments or three. If there are three arguments, then keep-between is the middle one.
Note: the argument conventions of spl facilitate less verbose partial application, such as with macros in the op family, in the common situation when string is the unbound argument.
The spln function is similar to spl, taking a required argument count, which behaves exactly like the same-named argument of spl-str.
(split-str-set string set)
(sspl set string)
The split-str-set function breaks the string into pieces, returning a list thereof. The set argument must be a string. It specifies a set of characters. All occurrences of any of these characters within string are identified, and are removed from string. The string is broken into pieces according to the gaps left behind by the removed separators.
Adjacent occurrences of characters from set within string are considered to be separate gaps which come between empty strings.
This operation is nondestructive: string is not modified in any way.
The sspl function performs the same operation; the only difference between sspl and split-str-set is argument order.
(tok-str string regex [keep-between [count]])
(tok-where string regex)
The tok-str function searches string for tokens, which are defined as substrings of string which match the regular expression regex in the longest possible way, and do not overlap. These tokens are extracted from the string and returned as a list.
Whenever regex matches an empty string, then an empty token is returned, and the search for another token within string resumes after advancing by one character position. However, if an empty match occurs immediately after a nonempty token, that empty match is not turned into a token.
So for instance, (tok-str "abc" #/a?/) returns ("a" "" ""). After the token "a" is extracted from a nonempty match for the regex, an empty match for the regex occurs just before the character b. This match is discarded because it is an empty match which immediately follows the nonempty match. The character b is skipped. The next match is an empty match between the b and c characters. This match causes an empty token to be extracted. The character c is skipped, and one more empty match occurs after that character and is extracted.
If the keep-between argument is true, then the behavior of tok-str changes in the following way. The pieces of string which are skipped by the search for tokens are included in the output. If no token is found in string, then a list of one element is returned, containing string. Generally, if N tokens are found, then the returned list consists of 2N + 1 elements. The first element of the list is the (possibly empty) substring which had to be skipped to find the first token. Then the token follows. The next element is the next skipped substring and so on. The last element is the substring of string between the last token and the end.
If count is specified, it must be a nonnegative integer. The value limits the number of tokens which are extracted. The returned list then includes one more item: the remainder of the string after the last extracted token. This item is omitted if the rest of the string is empty, unless keep-between is true.
The tok-where function works similarly to tok-str, but instead of returning the extracted tokens themselves, it returns a list of the character position ranges within string where matches for regex occur. The ranges are pairs of numbers, represented as cons cells, where the first number of the pair gives the starting character position, and the second number is one position past the end of the match. If a match is empty, then the two numbers are equal.
The tok-where function does not support the keep-between parameter.
(tok regex [keep-between] string)
(tokn count regex [keep-between] string)
The tok function performs the same computation as tok-str. The same-named parameters of tok and tok-str have the same semantics. The difference is the argument order. The tok function takes the regex argument first. The last argument is always string whether or not there are two arguments or three. If there are three arguments, then keep-between is the middle one.
Note: the argument conventions of tok facilitate less verbose partial application, such as with macros in the op family, in the common situation when string is the unbound argument.
The tokn function is similar to tok, taking a required argument count, which behaves exactly like the same-named argument of tok-str.
(list-str string)
The list-str function converts a string into a list of characters.
(trim-str string)
The trim-str function produces a copy of string from which leading and trailing tabs, spaces and newlines are removed.
(str-esc esc-set esc-tok str)
The str-esc function performs a character escaping transformation on the input string str.
The argument esc-set is a string containing zero or more characters.
The esc-tok argument is a character or string.
The function returns a transformed version of str in which every character of str which occurs in esc-set is preceded by esc-tok.
(str-esc "$@#" "$" "$foo @abc #1") -> "$$foo $@abc $#1"
(str-esc "'" "'\\'" "foo 'bar' baz") -> "foo '\\''bar'\\'' baz"
(string-set-code string value)
(string-get-code string)
The string-set-code and string-get-code functions provide a mechanism for associating an integer code with a string.
Note: this mechanism is the basis for associating system error messages passed in exceptions with the errno values of the failed system library calls which precipitated these error exceptions.
Not all string types can have an integer code: lazy strings and literal strings do not have this capability. The string argument must be of type str.
The value argument must be an integer or character. It is recommended that its value be confined to the non-negative range of the platform's int C type. Otherwise it is unspecified whether the same value shall be observed by string-get-code as what was stored with string-set-code.
The string-set-code function associates the integer value with the given string, and returns string. Any previously associated value is overwritten.
The string-get-code function retrieves the value most recently associated with string. If string has no associated value, then nil is returned.
If the string-extend is invoked on a string then it is unspecified whether or not string has an associated value and, if so, what value that is, except in the following case: if string-extend is invoked with a final argument which is true, then string is caused not to have an associated value.
If the string-finish function is invoked on a string, that string is caused not to have an associated value.
(chrp obj)
Returns t if obj is a character, otherwise nil.
(chr-isalnum char)
Returns t if char is an alphanumeric character, otherwise nil. Alphanumeric means one of the uppercase or lowercase letters of the English alphabet found in ASCII, or an ASCII digit. This function is not affected by locale.
(chr-isalpha char)
Returns t if char is an alphabetic character, otherwise nil. Alphabetic means one of the uppercase or lowercase letters of the English alphabet found in ASCII. This function is not affected by locale.
(chr-isascii char)
The chr-isascii function returns t if the code of character char is in the range 0 to 127 inclusive. For characters outside of this range, it returns nil.
(chr-iscntrl char)
The chr-iscntrl function returns t if the character char is a control character. For all other character, it returns nil.
A control character is one which belongs to the Unicode C0 or C1 block. C0 consists of the characters U+0000 through U+001F, plus the character U+007F. These are the original ASCII control characters. Block C1 consists of U+0080 through U+009F.
(chr-isdigit char)
(chr-digit char)
If char is an ASCII decimal digit character, chr-isdigit returns the value t and chr-digit returns the integer value corresponding to that digit character, a value in the range 0 to 9. Otherwise, both functions return nil.
(chr-isgraph char)
The chr-isgraph function returns t if char is a non-space printable ASCII character. It returns nil if it is a space or control character.
It also returns nil for non-ASCII characters: Unicode characters with a code above 127.
(chr-islower char)
The chr-islower function returns t if char is an ASCII lowercase letter. Otherwise it returns nil.
(chr-isprint char)
The chr-isprint function returns t if char is an ASCII character which is not a control character. It also returns nil for all non-ASCII characters: Unicode characters with a code above 127.
(chr-ispunct char)
The chr-ispunct function returns t if char is an ASCII character which is not a control character. It also returns nil for all non-ASCII characters: Unicode characters with a code above 127.
(chr-isspace char)
The chr-isspace function returns t if char is an ASCII whitespace character: any of the characters in the set #\space, #\tab, #\linefeed, #\newline, #\return, #\vtab and #\page. For all other characters, it returns nil.
(chr-isblank char)
The chr-isblank function returns t if char is a space or tab: the character #\space or #\tab. For all other characters, it returns nil.
(chr-isunisp char)
The chr-isunisp function returns t if char is a Unicode whitespace character. This the case for all the characters for which chr-isspace returns t. It also returns t for these additional characters: #\xa0, #\x1680, #\x180e, #\x2000, #\x2001, #\x2002, #\x2003, #\x2004, #\x2005, #\x2006, #\x2007, #\x2008, #\x2009, #\x200a, #\x2028, #\x2029, #\x205f, and #\x3000. For all other characters, it returns nil.
(chr-isupper char)
The chr-isupper function returns t if char is an ASCII uppercase letter. Otherwise it returns nil.
(chr-isxdigit char)
(chr-xdigit char)
If char is a hexadecimal digit character, chr-isxdigit returns the value t and chr-xdigit returns the integer value corresponding to that digit character, a value in the range 0 to 15. Otherwise, both functions returns nil.
A hexadecimal digit is one of the ASCII digit characters 0 through 9, or else one of the letters A through F or their lowercase equivalents a through f denoting the values 10 to 15.
(chr-toupper char)
If character char is a lowercase ASCII letter character, this function returns the uppercase equivalent character. If it is some other character, then it just returns char.
(chr-tolower char)
If character char is an uppercase ASCII letter character, this function returns the lowercase equivalent character. If it is some other character, then it just returns char.
(int-chr char)
(chr-int num)
The char argument must be a character. The int-chr function returns that character's Unicode code point value as an integer.
The num argument must be a fixnum integer in the range 0 to #\x10FFFF. The chr-int function interprets num as a Unicode code point value and returns the corresponding character object.
Note: these functions are also known by the obsolescent names num-chr and chr-num.
(chr-str str idx)
(set (chr-str str idx) new-value)
The chr-str function performs random access on string str to retrieve the character whose position is given by integer idx, which must be within range of the string.
The index value 0 corresponds to the first (leftmost) character of the string and so nonnegative values up to one less than the length are possible.
Negative index values are also allowed, such that -1 corresponds to the last (rightmost) character of the string, and so negative values down to the additive inverse of the string length are possible.
An empty string cannot be indexed. A string of length one supports index 0 and index -1. A string of length two is indexed left to right by the values 0 and 1, and from right to left by -1 and -2.
If the element idx of string str exists, and the string is modifiable, then the chr-str form denotes a place.
A chr-str place supports deletion. When a deletion takes place, then the character at idx is removed from the string. Any characters after that position move by one position to close the gap, and the length of the string decreases by one.
Direct use of chr-str is equivalent to the DWIM bracket notation except that str must be a string. The following relation holds:
(chr-str s i) --> [s i]
since [s i] <--> (ref s i), this also holds:
However, note the following difference. When the expression [s i] is used as a place, then the subexpression s must be a place. When (chr-str s i) is used as a place, s need not be a place.
(chr-str-set str idx char)
The chr-str function performs random access on string str to overwrite the character whose position is given by integer idx, which must be within range of the string. The character at idx is overwritten with character char.
The idx argument works exactly as in chr-str.
The str argument must be a modifiable string.
Direct use of chr-str is equivalent to the DWIM bracket notation provided that str is a string and idx an integer. The following relation holds:
(chr-str-set s i c) --> (set [s i] c)
Since (set [s i] c) <--> (refset s i c) for an integer index i, this also holds:
(chr-str s i) --> (refset s i c)
(span-str str set)
The span-str function determines the longest prefix of string str which consists only of the characters in string set, in any combination.
If both arguments are strings, the function returns an integer between 0 and the length of str.
(span-str "abcde" "ab") -> 2
(span-str "abcde" "z") -> 0
(span-str "abcde" "") -> 0
(span-str "abcde" "edcba") -> 5
(compl-span-str str set)
The compl-span-str function determines the longest prefix of string str which consists only of the characters which do not appear in set, in any combination.
If both arguments are strings, the function returns an integer between 0 and the length of str.
(compl-span-str "abc,def" ",") -> 3
(compl-span-str "abc," ",") -> 3
(compl-span-str "abc" ",") -> 3
(compl-span-str "abc3" "0123456789") -> 3
(compl-span-str "3" "0123456789") -> 0
(break-str str set)
The break-str function returns an integer which represents the position of the first character in string str which appears in string set.
If there is no such character, then nil is returned.
(break-str "abc,def.ghi" ",.:") -> 3
(break-str "abc,def.ghi" ".:") -> 6
(break-str "abc,def.ghi" ":") -> nil
Lazy strings are objects that were developed for the TXR pattern-matching language, and are exposed via TXR Lisp. Lazy strings behave much like strings, and can be substituted for strings. However, unlike regular strings, which exist in their entirety, first to last character, from the moment they are created, lazy strings do not exist all at once, but are created on demand. If character at index N of a lazy string is accessed, then characters 0 through N of that string are forced into existence. However, characters at indices beyond N need not necessarily exist.
A lazy string dynamically grows by acquiring new text from a list of strings which is attached to that lazy string object. When the lazy string is accessed beyond the end of its hitherto materialized prefix, it takes enough strings from the list in order to materialize the index. If the list doesn't have enough material, then the access fails, just like an access beyond the end of a regular string. A lazy string always takes whole strings from the attached list.
Lazy string growth is achieved via the lazy-str-force-upto function which forces a string to exist up to a given character position. This function is used internally to handle various situations.
The lazy-str-force function forces the entire string to materialize. If the string is connected to an infinite lazy list, this will exhaust all memory.
Lazy strings are specially recognized in many of the regular string functions, which do the right thing with lazy strings. For instance when sub-str is invoked on a lazy string, a special version of the sub-str logic is used which handles various lazy string cases, and can potentially return another lazy string. Taking a sub-str of a lazy string from a given character position to the end does not force the entire lazy string to exist, and in fact the operation will work on a lazy string that is infinite.
Furthermore, special lazy string functions are provided which allow programs to be written carefully to take better advantage of lazy strings. What carefully means is code that avoids unnecessarily forcing the lazy string. For instance, in many situations it is necessary to obtain the length of a string, only to test it for equality or inequality with some number. But it is not necessary to compute the length of a string in order to know that it is greater than some value.
(lazy-str string-list [terminator [limit-count]])
The lazy-str function constructs a lazy string which draws material from string-list which is a list of strings.
If the optional terminator argument is given, then it specifies a string which is appended to every string from string-list, before that string is incorporated into the lazy string. If terminator is not given, then it defaults to the string "\n", and so the strings from string-list are effectively treated as lines which get terminated by newlines as they accumulate into the growing prefix of the lazy string. To avoid the use of a terminator string, a null string terminator argument must be explicitly passed. In that case, the lazy string grows simply by catenating elements from string-list.
If the limit-count argument is specified, it must be a positive integer. It expresses a maximum limit on how many elements will be consumed from string-list in order to feed the lazy string. Once that many elements are drawn, the string ends, even if the list has not been exhausted. However, that remaining list, though not contributing to the string, is still incorporated into the value returned by lazy-str-get-trailing-list.
(lazy-stringp obj)
The lazy-stringp function returns t if obj is a lazy string. Otherwise it returns nil.
(lazy-str-force-upto lazy-str index)
The lazy-str-force-upto function tries to instantiate the lazy string such that the position given by index materializes. The index is a character position, exactly as used in the chr-str function.
It is an error if the lazy-str argument isn't a lazy string.
Some positions beyond index may also materialize, as a side effect, because the operation takes only whole strings from the internal list, according to the algorithm described below.
If the string is already materialized through to at least index, or if it is possible to materialize the string that far, then the value t is returned to indicate success.
If there is sufficient material to force the lazy string through to the index position, then t is returned, otherwise nil.
The lazy-str object's limit-count is observed: a total of no more than limit-count elements are taken from the object's list.
The algorithm is as follows:
(lazy-str-force lazy-str)
The lazy-str argument must be a lazy string. The lazy string is forced to fully materialize.
The return value is an ordinary, non-lazy string equivalent to the fully materialized lazy string.
The lazy-str object's limit-count is observed: a total of no more than limit-count elements are taken from the object's list.
The algorithm that is followed by lazy-str-force is similar to the one followed by lazy-str-force-upto, with only the following modification. The test in step 1 isn't concerned with the length of the materialized prefix, since the goal is to materialize all available characters. Steps 2 and 3 are performed while elements are available in the list, subject to observance of the limit-count.
(lazy-str-get-trailing-list string index)
The lazy-str-get-trailing-list function can be considered, in some way, an inverse operation to the production of the lazy string from its associated list.
Note: the behavior of this function changed in TXR 274. This is subject to a note in the COMPATIBILITY section.
First, the lazy string string is forced up through the position index, as if by a call to lazy-str-force-upto.
If string consists of index or more characters, then after the forcing operation, it is guaranteed that at least index characters of the string have been materialized into a single string, called the materialized prefix of the lazy string. If fewer than index characters are available, taking into account the contribution of the terminator string, then the number of characters in the materialized prefix fall short of index. The materialized prefix never takes fractional strings from the lazy string's list, and is always terminated by the terminator string.
Next, the materialized prefix is split into pieces on occurrences of string's terminator string, as if by using spl function. If the terminator string is empty, it is split into individual characters, in accordance with the semantics of that function.
Then, if the last piece of the split prefix is an empty string, it is removed. This situation occurs in two cases: the materialized prefix is empty, or else it ends in the terminating string. For example, if the terminating string is a single newline, and the prefix is "foo\n". In this case, (spl "\n" "foo\n") produces ("foo" "") from which the trailing empty string is removed, leaving ("foo").
Finally, a list is formed by appending the split piece of the materialized prefix, calculated as described above, with string's remaining list of strings which have not been pulled into the materialized prefix. This list is returned.
(length-str-> string len)
(length-str->= string len)
(length-str-< string len)
(length-str-<= string len)
These functions compare the lengths of two strings. The following equivalences hold, as far as the resulting value is concerned:
(length-str-> s l) <--> (> (length-str s) l)
(length-str->= s l) <--> (>= (length-str s) l)
(length-str-< s l) <--> (< (length-str s) l)
(length-str-<= s l) <--> (<= (length-str s) l)
The difference between the functions and the equivalent forms is that if the string is lazy, the length-str function will fully force it in order to calculate and return its length.
These functions only force a string up to position len, so they are not only more efficient, but on infinitely long lazy strings they are usable.
length-str cannot compute the length of a lazy string with an unbounded length; it will exhaust all memory trying to force the string.
These functions can be used to test such as string whether it is longer or shorter than a given length, without forcing the string beyond that length.
(cmp-str left-string right-string)
The cmp-str function returns -1 if left-string is lexicographically prior to right-string. If the reverse relationship holds, it returns 1. Otherwise the strings are equal and zero is returned.
If either or both of the strings are lazy, then they are only forced to the minimum extent necessary for the function to reach a conclusion and return the appropriate value, since there is no need to look beyond the first character position in which they differ.
The lexicographic ordering is naive, based on the character code point values in Unicode taken as integers, without regard for locale-specific collation orders.
Note: in TXR 232 and earlier versions, cmp-str conforms to a weaker requirements: any negative integer value may be returned rather than -1, and any positive integer value can be returned instead of 1.
(str= left-string right-string)
(str< left-string right-string)
(str> left-string right-string)
(str<= left-string right-string)
(str>= left-string right-string)
These functions compare left-string and right-string lexicographically, as if by the cmp-str function.
The str= function returns t if the two strings are exactly the same, character for character, otherwise it returns nil.
The str< function returns t if left-string is lexicographically before right-string, otherwise nil.
The str> function returns t if left-string is lexicographically after right-string, otherwise nil.
The str< function returns t if left-string is lexicographically before right-string, or if they are exactly the same, otherwise nil.
The str< function returns t if left-string is lexicographically after right-string, or if they are exactly the same, otherwise nil.
(vector length [initval])
The vector function creates and returns a vector object of the specified length. The elements of the vector are initialized to initval, or to nil if initval is omitted.
(vec arg*)
The vec function creates a vector out of its arguments.
(vectorp obj)
The vectorp function returns t if obj is a vector, otherwise it returns nil.
(vec-set-length vec len)
The vec-set-length modifies the length of vec, making it longer or shorter. If the vector is made longer, then the newly added elements are initialized to nil. The len argument must be nonnegative.
The return value is vec.
(vecref vec idx)
(set (vecref vec idx) new-value)
The vecref function performs indexing into a vector. It retrieves an element of vec at position idx, counted from zero. The idx value must range from 0 to one less than the length of the vector. The specified element is returned.
If the element idx of vector vec exists, then the vecref form denotes a place.
A vecref place supports deletion. When a deletion takes place, then if idx denotes the last element in the vector, the vector's length is decreased by one, so that the vector no longer has that element. Otherwise, if idx isn't the last element, then each elements values at a higher index than idx shifts by one one element position to the adjacent lower index. Then, the length of the vector is decreased by one, so that the last element position disappears.
(vec-push vec elem)
The vec-push function extends the length of a vector vec by one element, and sets the new element to the value elem.
The previous length of the vector (which is also the position of elem) is returned.
(length-vec vec)
The length-vec function returns the length of vector vec. It performs similarly to the generic length function, except that the argument must be a vector.
(size-vec vec)
The size-vec function returns the number of elements for which storage is reserved in the vector vec.
The length of the vector can be extended up to this size without any memory allocation operations having to be performed.
(vec-list list)
The vec-list function returns a vector which contains all of the same elements and in the same order as list list.
Note: this function is also known by the obsolescent name vector-list.
(list-vec vec)
The list-vec function returns a list of the elements of vector vec.
Note: this function is also known by the obsolescent name list-vector.
(copy-vec vec)
The copy-vec function returns a new vector object of the same length as vec and containing the same elements in the same order.
(sub-vec vec [from [to]])
(set (sub-vec vec [from [to]]) new-value)
The sub-vec function has the same parameters and semantics as the function sub, except that the vec argument must be a vector.
If a sub-vec form is used as a place, it denotes a subrange of list as if it were a storage location. The previous value of this location, if needed, is fetched by a call to sub-vec. Storing new-value to the place is performed by a call to replace-vec. In an update operation which accesses the prior value and stores a new value, the arguments vec, from, to and new-value are evaluated once.
The vec argument is not itself required to be a place; it is not updated when a value is written to the sub-vec storage location.
(replace-vec vec item-sequence [from [to]])
The replace-vec is like the replace function except that the vec argument must be a vector.
(fill-vec vec elem [from [to]])
The fill-vec function overwrites a range of the vector with copies of the elem value.
The from and to index arguments follow the same range indexing conventions as the replace and sub functions. If from is omitted, it defaults to zero. If to is omitted, it defaults to the length of vec. Negative values of from and to are adjusted by adding the length of the vector to them, once.
If the adjusted value of either from or to is negative, or exceeds the length of vec, an error exception is thrown.
The adjusted values of to and from specify a range of vec starting at the from index, and ending at the to index, which is excluded from the range.
If the adjusted to is less than or equal to the adjusted from, then vec is unaltered.
Otherwise, copies of element are stored into vec starting at the from index, ending just before the to index is reached.
The fill-vec function returns vec.
(defvarl v (vec 1 2 3))
v --> #(1 2 3)
(fill-vec v 0) --> #(0 0 0)
(fill-vec v 3 1) --> #(0 3 3)
(fill-vec v 4 -1) --> #(0 3 4)
(fill-vec v 5 -3 -1) --> #(5 5 4)
(cat-vec vec-list)
The vec-list argument is a list of vectors. The cat-vec function produces a catenation of the vectors listed in vec-list. It returns a single large vector formed by catenating those vectors together in order.
(nested-vec dimension*)
(nested-vec-of object dimension*)
The nested-vec-of function constructs a nested vector according to the dimension arguments, described in detail below.
The nested-vec function is equivalent to nested-vec-of with an object argument of nil.
When there are no dimension arguments, nested-vec-of returns nil.
If there is exactly one dimension argument, it must be a nonnegative integer. A newly created having that many elements is returned, with each element of the vector being object.
If there are two or more dimension arguments, nested vector is returned. The first dimension argument specifies the outermost dimension: a vector of that many elements are returned. Each element of that vector is a vector whose length is given by the second dimension. This nesting pattern continues through the remaining dimensions. The last dimension specifies the length of vectors which are filled with object.
From the above it follows that if a zero-valued dimension is encountered, every vector corresponding to that level of nesting shall be empty, and that shall be the last dimension regardless of the presence of additional dimension arguments.
(nested-vec) -> nil
(nested-vec-of 0 4) -> #(0 0 0 0)
(nested-vec-of 0 4 3) -> #(#(0 0 0)
#(0 0 0)
#(0 0 0)
#(0 0 0))
(nested-vec-of 'a 4 3 2) -> #(#(#(a a) #(a a) #(a a))
#(#(a a) #(a a) #(a a))
#(#(a a) #(a a) #(a a))
#(#(a a) #(a a) #(a a)))
(nested-vec-of 'a 1 1 1) -> #(#(#(a)))
(nested-vec-of 'a 1 1 0) -> #(#(#()))
(nested-vec-of 'a 1 0 1) -> #(#())
(nested-vec-of 'a 1 0) -> #(#())
(nested-vec-of 'a 0 1) -> #()
(nested-vec-of 'a 0) -> #()
(nested-vec-of 'a 4 0 1) #(#() #() #() #())
(nested-vec-of 'a 4 0) #(#() #() #() #()))
Object of the type buf are buffers: vector-like objects specialized for holding binary data represented as a sequence of 8-bit bytes. Buffers support operations specialized toward the encoding of Lisp values into machine-oriented data types, and decoding such data types into Lisp values.
Buffers are particularly useful in conjunction with the Foreign Function Interface (FFI), since they can be used to prepare arbitrary data which can be passed into and out of a function by pointer. They are also useful for binary I/O.
Buffers support a number of similar functions for converting Lisp numeric values into common data types, which are placed into the buffer. These functions are named starting with the buf-put- prefix, followed by an abbreviated type name.
Each of these functions takes three arguments: buf specifies the buffer, pos specifies the byte offset position into the buffer which receives the low-order byte of the data transfer, and val indicates the value.
If pos has a value such that any portion of the data transfer would like outside of the buffer, the buffer is automatically extended in length to contain the data transfer. If this extension causes any padding bytes to appear between the previous length of the buffer and pos, those bytes are initialized to zero.
The argument val giving the value to be stored must be an integer or character, except in the case of the types float and double(the functions buf-put-float and buf-put-double) for which it is required to be of type float, and in case of the function buf-put-cptr which expects the val argument to be a cptr object.
The val argument must be in range for the data type, or an exception results.
Unless otherwise indicated, the stored datum is in the local format used by the machine with regard to byte order and other representational details.
Buffers support a number of similar functions for extracting common data types, and converting them into Lisp values. These functions are named starting with the buf-get- prefix, followed by an abbreviated type name.
Each of these functions takes two arguments: buf specifies the buffer and pos specifies the byte offset position into the buffer which holds the low-order byte of the datum to be extracted.
If any portion of requested datum lies outside of the boundaries of the buffer, an error exception is thrown.
The extracted value is converted to a Lisp datum. For the majority of these functions, the returned value is of type integer. The buf-get-float and buf-get-double return a floating-point value. The buf-get-cptr function returns a value of type cptr.
(make-buf len [init-val [alloc-size]])
The make-buf function creates a new buffer object which holds len bytes. This argument may be zero.
If init-val is present, it specifies the value with which the first len bytes of the buffer are initialized. If omitted, it defaults to zero. The value of init-val must lie in the range 0 to 255.
The alloc-size parameter indicates how much memory to actually allocate for the buffer. If an argument is not given, the parameter takes on the same value as len. If an argument is given, its value must not be less than len.
(bufp object)
The bufp function returns t if object is a buf, otherwise it returns nil.
(length-buf buf)
The length-buf function retrieves the buffer length: how many bytes are stored in the buffer.
Note: the generic length function is also applicable to buffers.
(buf-alloc-size buf)
The buf-alloc-size function retrieves the allocation size of the buffer.
(buf-trim buf)
The buf-trim function reduces the amount of memory allocated to the buffer to the minimum required to hold it contents, effectively setting the allocation size to the current length.
The previous allocation size is returned.
(buf-set-length buf len [init-val])
The buf-set-length function changes the length of the buffer. If the buffer is made longer, the newly added bytes appear at the end, and are initialized to the value given by init-val. If init-val is specified, its value must be in the range 0 to 255. It defaults to zero.
(copy-buf buf)
The copy-buf function returns a duplicate of buf: an object distinct from buf which has the same length and contents, and compares equal to buf.
(sub-buf buf [from [to]])
(set (sub-buf buf [from [to]]) new-val)
The sub-buf function has the same semantics as the sub function, except that the first argument must be a buffer.
The extracted sub-range of a buffer is itself a buffer object.
If sub-buf is used as a syntactic place, the argument expressions buf, from, to and new-val are evaluated just once. The prior value, if required, is accessed by calling sub-buf and new-val is then stored via replace-buf.
(replace-buf buf item-sequence [from [to]])
The replace-buf function has the same semantics as the replace function, except that the first argument must be a buffer.
The elements of item-sequence are stored into buf as if using the buf-put-u8 function and therefore must be suitable val arguments for that function.
The of the arguments, semantics and return value given for replace apply to replace-buf.
(buf-list list)
The buf-list function creates and returns a new buffer, whose contents are derived from the elements of list, which may be any kind of sequence.
The elements of list must be integers whose values lie in the range 0 to 255, or else characters whose code point values lie in that range. These values are placed into the newly created buffer, which therefore has the same length as list.
(buf-put-buf dst-buf pos src-buf)
The buf-put-buf function stores a copy of buffer src-buf into dst-buf at the offset indicated by pos.
The source and destination memory regions may overlap.
The return value is src-buf.
Note: the effect of a buf-put-buf operation may also be performed by a suitable call to replace-buf; however, buf-put-buf is less general: it doesn't insert or delete by replacing destination ranges with data of differing length, and requires a source operand of buffer type.
(buf-put-i8 buf pos val)
The buf-put-i8 converts val into an 8-bit signed integer, and stores it into the buffer at the offset indicated by pos.
The return value is val.
(buf-put-u8 buf pos val)
The buf-put-u8 converts val into an 8-bit unsigned integer, and stores it into the buffer at the offset indicated by pos.
The return value is val.
(buf-put-i16 buf pos val)
The buf-put-i16 converts val into a sixteen bit signed integer, and stores it into the buffer at the offset indicated by pos.
The return value is val.
(buf-put-u16 buf pos val)
The buf-put-u16 converts val into a sixteen bit unsigned integer, and stores it into the buffer at the offset indicated by pos.
The return value is val.
(buf-put-i32 buf pos val)
The buf-put-i32 converts val into a 32-bit signed integer, and stores it into the buffer at the offset indicated by pos.
The return value is val.
(buf-put-u32 buf pos val)
The buf-put-u32 converts val into a 32-bit unsigned integer, and stores it into the buffer at the offset indicated by pos.
The return value is val.
(buf-put-i64 buf pos val)
The buf-put-i64 converts val into a 64-bit signed integer, and stores it into the buffer at the offset indicated by pos.
The return value is val.
(buf-put-u64 buf pos val)
The buf-put-u64 converts the value val into a 64-bit unsigned integer, and stores it into the buffer at the offset indicated by pos.
The return value is val.
(buf-put-char buf pos val)
The buf-put-char converts val into a value of the C type char and stores it into the buffer at the offset indicated by pos.
The return value is val.
Note that the char type may be signed or unsigned.
(buf-put-uchar buf pos val)
The buf-put-uchar converts val into a value of the C type unsigned char and stores it into the buffer at the offset indicated by pos.
(buf-put-short buf pos val)
The buf-put-short converts val into a value of the C type short and stores it into the buffer at the offset indicated by pos.
(buf-put-ushort buf pos val)
The buf-put-ushort converts val into a value of the C type unsigned short and stores it into the buffer at the offset indicated by pos.
(buf-put-int buf pos val)
The buf-put-int converts val into a value of the C type int and stores it into the buffer at the offset indicated by pos.
(buf-put-uint buf pos val)
The buf-put-uint converts val into a value of the C type unsigned int and stores it into the buffer at the offset indicated by pos.
(buf-put-long buf pos val)
The buf-put-long converts val into a value of the C type long and stores it into the buffer at the offset indicated by pos.
(buf-put-ulong buf pos val)
The buf-put-ulong converts val into a value of the C type unsigned long and stores it into the buffer at the offset indicated by pos.
(buf-put-float buf pos val)
The buf-put-float converts val into a value of the C type float and stores it into the buffer at the offset indicated by pos.
Note: the conversion of a TXR Lisp floating-point value to the C type float may be inexact, reducing the numeric precision.
(buf-put-double buf pos val)
The buf-put-double converts val into a value of the C type double and stores it into the buffer at the offset indicated by pos.
(buf-put-cptr buf pos val)
The buf-put-cptr expects val to be of type cptr. It stores the object's pointer value into the buffer at the offset indicated by pos.
(buf-get-i8 buf pos)
The buf-get-i8 function extracts and returns signed 8-bit integer from buf at the offset given by pos.
(buf-get-u8 buf pos)
The buf-get-u8 function extracts and returns an unsigned 8-bit integer from buf at the offset given by pos.
(buf-get-i16 buf pos)
The buf-get-i16 function extracts and returns a signed 16-bit integer from buf at the offset given by pos.
(buf-get-u16 buf pos)
The buf-get-u16 function extracts and returns an unsigned 16-bit integer from buf at the offset given by pos.
(buf-get-i32 buf pos)
The buf-get-i32 function extracts and returns a signed 32-bit integer from buf at the offset given by pos.
(buf-get-u32 buf pos)
The buf-get-u32 function extracts and returns an unsigned 32-bit integer from buf at the offset given by pos.
(buf-get-i64 buf pos)
The buf-get-i64 function extracts and returns a signed 64-bit integer from buf at the offset given by pos.
(buf-get-u64 buf pos)
The buf-get-u64 function extracts and returns an unsigned 64-bit integer from buf at the offset given by pos.
(buf-get-char buf pos)
The buf-get-char function extracts and returns a value of the C type char from buf at the offset given by pos. Note that char may be signed or unsigned.
(buf-get-uchar buf pos)
The buf-get-uchar function extracts and returns a value of the C type unsigned char from buf at the offset given by pos.
(buf-get-short buf pos)
The buf-get-short function extracts and returns a value of the C type short from buf at the offset given by pos.
(buf-get-ushort buf pos)
The buf-get-ushort function extracts and returns a value of the C type unsigned short from buf at the offset given by pos.
(buf-get-int buf pos)
The buf-get-int function extracts and returns a value of the C type int from buf at the offset given by pos.
(buf-get-uint buf pos)
The buf-get-uint function extracts and returns a value of the C type unsigned int from buf at the offset given by pos.
(buf-get-long buf pos)
The buf-get-long function extracts and returns a value of the C type long from buf at the offset given by pos.
(buf-get-ulong buf pos)
The buf-get-ulong function extracts and returns a value of the C type unsigned long from buf at the offset given by pos.
(buf-get-float buf pos)
The buf-get-float function extracts and returns a value of the C type float from buf at the offset given by pos, returning that value as a Lisp floating-point number.
(buf-get-double buf pos)
The buf-get-double function extracts and returns a value of the C type double from buf at the offset given by pos, returning that value as a Lisp floating-point number.
(buf-get-cptr buf pos)
The buf-get-cptr function extracts a C pointer from buf at the offset given by pos, returning that value as a Lisp object of type cnum.
(put-buf buf [pos [stream]])
The put-buf function writes the contents of buffer buf, starting at position pos to a stream, through to the last byte, if possible. Successive bytes from the buffer are written to the stream as if by a put-byte operation.
If stream is omitted, it defaults to *stdout*.
If pos is omitted, it defaults to zero. It indicates the starting position within the buffer.
The stream must support the put-byte operation. Streams which support put-byte can be expected to support put-buf and, conversely, streams which do not support put-byte do not support put-buf.
The put-buf function returns the position of the last byte that was successfully written. If the buffer was written through to the end, then this value corresponds to the length of the buffer.
If an error occurs before any bytes are written, the function throws an error.
(fill-buf buf [pos [stream]])
(fill-buf-adjust buf [pos [stream]])
The fill-buf reads bytes from stream and writes them into consecutive locations in buffer buf starting at position pos. The bytes are read as if using the get-byte function.
If the stream argument is omitted, it defaults to *stdin*.
If pos is omitted, it defaults to zero. It indicates the starting position within the buffer.
The stream must support the get-byte operation. Buffers which support get-byte can be expected to support fill-buf and, conversely, streams which do not support get-byte do not support fill-buf.
The fill-buf function returns the position that is one byte past the last byte that was successfully read. If an end-of-file or other error condition occurs before the buffer is filled through to the end, then the value returned is smaller than the buffer length. In this case, the area of the buffer beyond the read size retains its previous content.
If an error situation occurs other than a premature end-of-file before any bytes are read, then an exception is thrown.
If an end-of-file condition occurs before any bytes are read, then zero is returned.
The fill-buf-adjust differs usefully from fill-buf as follows. Whereas fill-buf doesn't manipulate the length of the buffer at any stage of the operation, the fill-buf-adjust begins by adjusting the length of the buffer to the underlying allocated size. Then it performs the fill operation in exactly the same manner as fill-buf. Finally, if the operation succeeds, then fill-buf-adjust adjusts the length of the buffer to match the position that is returned.
(get-line-as-buf [stream])
The get-line-as-buf reads bytes from stream as if using the get-byte function, until either a the newline character is encountered, or else the end of input is encountered. The bytes which are read, exclusive of the newline character, are returned in a new buffer object. The newline character, if it occurs, is consumed.
If stream is omitted, it defaults to *stdin*.
The stream is required to support byte input.
(file-get-buf name [max-bytes
[skip-bytes [mode-opts]]])
(command-get-buf cmd [max-bytes [skip-bytes]])
The file-get-buf function opens a binary stream over the file indicated by the string argument name for reading. By default, the entire file is read and its contents are returned as a buffer object. The buffer's length corresponds to the number of bytes read from the file.
The command-get-buf function opens a binary stream over an input command pipe created for the command string cmd, as if by the open-command function. It read bytes from the pipe until the indication that no more input is available. The bytes are returned aggregated into a buffer object.
If the max-bytes parameter is given an argument, it must be a nonnegative integer. That value specifies a limit on the number of bytes to read. A buffer no longer than max-bytes shall be returned.
If the skip-bytes parameter is given an argument, it must be a nonnegative integer. That value specifies how many initial bytes of the input should be discarded before accumulation of the buffer begins. If possible, the semantics of this parameter is achieved by performing a seek-stream operation, falling back on reading and discarding bytes if the stream doesn't support seeking.
If max-bytes is specified, then the stream is opened in unbuffered mode, so that bytes beyond the specified range shall not be requested from the underlying file, device or process.
The file-get-buf function opens the file as if using the open-file function, using a mode-string of "r". If the mode-opts is present, it specifies options to be added to the string. These must be compatible with the implicit "r" mode.
(file-put-buf name buf skip-bytes [mode-opts])
(file-place-buf name buf skip-bytes [mode-opts])
(file-append-buf name buf [mode-opts])
(command-put-buf cmd buf)
The file-put-buf function opens a text stream over the file indicated by the string argument name, writes the contents of the buffer object buf into the file, and then closes the file. If the file doesn't exist, it is created. If it exists, it is truncated to zero length and overwritten. The default value of the optional skip-bytes parameter is zero. If an argument is given, it must be a nonnegative integer. If it is nonzero, then after opening the file, before writing the buffer, the function will seek to an offset of that many bytes from the start of the file. The contents of buf will be written at that offset.
The file-place-buf function does not truncate an existing file to zero length. In all other regards, it is equivalent to file-put-buf.
The file-append-buf function is similar to file-put-buf except that if the file exists, it isn't overwritten. Rather, the buffer is appended to the file.
The command-put-buf function opens an output text stream over an output command pipe created for the command specified in the string argument cmd, as if by the open-command function. It then writes the contents of buffer buf into the stream and closes the stream.
The file-put-buf, file-place-buf and file-append-buf functions open a file as if using the open-file function using, respectively, mode-string values of "wb", "mb", and "ab".
The mode-opts argument, if present, specifies additional options to be added to these modes.
The return value of all three functions is that of the put-buf operation which is implicitly performed.
(buf-str str [null-term-p])
(str-buf buf [null-term-p])
The buf-str and str-buf functions perform UTF-8 conversion between the character string and buffer data types.
The buf-str function UTF-8-encodes str and returns a buffer containing the converted representation. If a true argument is given to the null-term-p parameter, then a null terminating byte is added to the buffer. This byte is added even if the previous byte is already a null byte from the conversion of a pseudo-null character occurring in str.
The str-buf function takes the contents of buffer buf to be UTF-8 data, which is converted to a character string and returned. Null bytes in the buffer are mapped to the pseudo-null character #\xDC00. If a true argument is given to the null-term-p parameter, then if the contents of buf end in a null byte, that byte is not included in the conversion.
(buf-int integer)
(buf-uint integer)
The buf-int and buf-uint functions convert a signed and unsigned integer, respectively, or else a character, into a binary representation, which is returned as a buffer object.
Under both functions, the representation uses big endian byte order: most significant byte first.
The buf-uint function requires a nonnegative integer argument, which may be a character. The representation stored in the buffer is a pure binary representation of the value using the smallest number of bytes required for the given integer value.
The buf-int function requires an integer or character argument. The representation stored in the buffer is a two's complement representation of integer using the smallest number of bytes which can represent that value. If integer is nonnegative, then the first byte of the buffer lies in the range 0 to 127. If integer is negative, then the first byte of the buffer lies in the range 128 to 255. The integer 255 therefore doesn't convert to the buffer #b'ff' but rather #b'00ff'. The buffer #b'ff' represents -1.
If the integer argument is a character object, it is taken to be its Unicode code point value, as returned by the int-chr function.
(int-buf buf)
(uint-buf buf)
The int-buf and uint-buf functions recover an integer value from its binary form which appears inside buf, which must be a buffer object. These functions expect buf to contain the representation produced by, respectively, the functions buf-int and buf-uint.
If buf holds the representation of an integer value n, as produced by (buf-int n) then (int-buf buf) returns n.
The same relationship holds between buf-uint and uint-buf.
Thus, these equalities hold:
(= (int-buf (buf-int
n)) n)
(= (uint-buf (buf-uint n)) n)
provided that n is of integer type and, in the case of buf-uint, nonnegative.
(buf-compress buf [level])
(buf-decompress buf)
The buf-compress and buf-decompress functions perform compression using the Deflate algorithm, via Zlib. These functions are only available if TXR is built with Zlib support. More specifically, buf-compress uses Zlib's compress2 function; therefore it can be expected to interoperate with other software which uses the same function.
The buf-compress function compresses the entire contents of buf and returns new buffer with the compressed contents. The optional level argument specifies the compression level as an integer. Valid values range from 0 (no compression) to 9 (maximum compression). The value -1 selects a default compression determined internally by Zlib.
The buf-decompress function reverses the buf-compress operation: it takes a compressed buf and returns a buffer containing the original uncompressed data.
The buf-compress function throws an error exception if the level value is unacceptable to Zlib. The buf-decompress function throws an error exception if buf doesn't contain a compressed image.
TXR supports user-defined types in the form of structures. Structures are objects which hold multiple storage locations called slots, which are named by symbols. Structures can be related to each other by inheritance. Multiple inheritance is permitted.
The type of a structure is itself an object, of type struct-type.
When the program defines a new structure type, it does so by creating a new struct-type instance, with properties which describe the new structure type: its name, its list of slots, its initialization and "boa constructor" functions, and the structures type it inherits from (the supertypes).
The struct-type object is then used to generate instances.
Structures instances are not only containers which hold named slots, but they also indicate their struct type. Two structures which have the same number of slots having the same names are not necessarily of the same type.
Structure types and structures may be created and manipulated using a programming interface based on functions.
For more convenient and clutter-free expression of structure-based program code, macros are also provided.
Furthermore, concise and expressive slot access syntax is provided courtesy of the referencing dot and unbound referencing dot syntax, a syntactic sugar for the qref and uref macros.
Structure types have a name, which is a symbol. The typeof function, when applied to any struct type, returns the symbol struct-type. When typeof is applied to a struct instance, it returns the name of the struct type. Effectively, struct names are types.
The consequences are unspecified if an existing struct name is reused for a different struct type, or an existing type name is used for a struct type.
Structure slots can be of two kinds: they can be the ordinary instance slots or they can be static slots. The instances of a given structure type have their own instance of a given instance slot. However, they all share a single instance of a static slot.
Static slots are allocated in a global area associated with a structure type and are initialized when the structure type is created. They are useful for efficiently representing properties which have the same value for all instances of a struct. These properties don't have to occupy space in each instance, and time doesn't have to be wasted initializing them each time a new instance is created. Static slots are also useful for struct-specific global variables. Lastly, static slots are also useful for holding methods and functions. Although structures can have methods and functions in their instances, usually, all structures of the same type share the same functions. The defstruct macro supports a special syntax for defining methods and struct-specific functions at the same time when a new structure type is defined. The defmeth macro can be used for adding new methods and functions to an existing structure and its descendants.
Static slots may be assigned just like instance slots. Changing a static slot changes that slot in every structure of the same type.
Static slots are not listed in the #S(...) notation when a structure is printed. When the structure notation is read from a stream, if static slots are present, they will be processed and their values stored in the static locations they represent, thus changing their values for all instances.
Static slots are inherited just like instance slots. The following simplified discussion is restricted to single inheritance. A detailed description of multiple inheritance is given in the Multiple Inheritance section below. If a given structure B has some static slot s, and a new structure D is derived from B, using defstruct, and does not define a slot s, then D inherits s. This means that D shares the static slot with B: both types share a single instance of that slot.
On the other hand if D defines a static slot s then that slot will have its own instance in the D structure type; D will not inherit the B instance of slot s. Moreover, if the definition of D omits the init-form for slot s, then that slot will be initialized with a copy of the current value of slot s of the B base type, which allows derived types to obtain the value of base type's static slot, yet have that in their own instance.
The slot type can be overridden. A structure type deriving from another type can introduce slots which have the same names as the supertype, but are of a different kind: an instance slot in the supertype can be replaced by a static slot in the derived type or vice versa.
Note that, in light of the above type overriding possibility, the static slot value propagation happens only from the immediate supertype. If D is derived from G which has a static slot s, whereas D specifies s as an instance slot, but then B again specifies a static slot s, then B's slot s will not inherit the value from G's s slot. Simply, B's supertype is D and that supertype is not considered to have a static slot s.
A structure type is associated with a static initialization function which may be used to store initial values into static slots. This function is invoked once in a type's life time, when the type is created. The function is also inherited by derived struct types and invoked when they are created.
In the make-struct-type function and defstruct macro, a list of supertypes can be given instead of just one. The type then inherits slots from all of the specified types. If any conflicts arise among the supertypes due to slots having the same name, the leftmost supertype dominates: that type's slot will be inherited. If the leftmost slot is static, then that static slot will be inherited. Otherwise, the instance slot will be inherited.
Of course, any slot which is specified in the newly defined type itself dominates over any same-named slots among the supertypes.
The new structure type inherits all of the slot initializing expressions, as well as :init and :postinit methods of all of its supertypes.
Each time the structure is instantiated, the :init initializing expressions inherited from the supertypes, together with the slot initializing expressions, are all evaluated, in right-to-left order: the initializations contributed by each supertype are performed before considering the next supertype to the left. The :postinit methods are similarly invoked in right-to-left order, before the :postinit methods of the new type itself. Thus the order is: supertype inits, own inits, supertype post-inits, own post-inits.
Until TXR 242, the situation of duplicate supertypes was ignored for the purposes of object initialization. It was documented that if a supertype is referenced by inheritance, directly or indirectly, two or more times, then its initializing expressions are evaluated that many times.
Starting in TXR 243, duplicate supertypes no longer give rise to duplicate initialization. When an object is instantiated, only one initialization of a duplicated supertype occurs. The subsequent initializations that would take place in the absence of duplicate detection are suppressed.
Note also that the :fini mechanism is tied to initialization. Initialization of an object registers the finalizers, and so in TXR 242, :fini finalizers are also executed multiple times, if :init initializers are.
Consider following program:
(defstruct base ()
(:init (me) (put-line "base init"))
(:fini (me) (put-line "base fini")))
(defstruct d1 (base)
(:init (me) (put-line "d1 init"))
(:fini (me) (put-line "d1 fini")))
(defstruct d2 (base)
(:init (me) (put-line "d2 init"))
(:fini (me) (put-line "d2 fini")))
(defstruct s (d1 d2))
(call-finalizers (new s))
Under TXR 242, and earlier versions that support multiple inheritance, it produces the output:
base init
d2 init
base init
d1 init
d1 fini
base fini
d2 fini
base fini
The supertypes are initialized in a right-to-left traversal of the type lattice, without regard for base being duplicated.
Starting with TXR 243, the output is:
base init
d2 init
d1 init
d1 fini
d2 fini
base fini
The rightmost duplicate of the base is initialized, so that the initialization is complete prior to the initializations of any dependent types. Likewise, the same rightmost duplicate of the base is finalized, so that finalization takes place after that of any dependent struct types.
Note, however, that the derived function function mechanism is not required to detect duplicated direct supertypes. If a supertype implements the derived function to detect situations when it is the target of inheritance, and some subtype inherits that type more than once, that function may be called more than once. The behavior is unspecified.
Newly constructed objects come into existence dirty. The dirty flag state can be tested with the function test-dirty. An object can be marked as clean by clearing its dirty flag with clear-dirty. A combined operation test-clear-dirty is provided which clears the dirty flag, and returns its previous value.
The dirty flag is set whenever a new value is stored into the instance slot of an object.
Note: the dirty flag can be used to support support the caching of values derived from an object's slots. The derived values don't have to be recomputed while an object remains clean.
In object-based or object-oriented programming, sometimes it is necessary for a new data type to provide its own notion of equality: its own requirements for when two distinct instances of the type are considered equal. Furthermore, types sometimes have to implement their own notion, also, of inequality: the requirements for the manner in which one instance is considered lesser or greater than another.
TXR Lisp structures implement a concept called equality substitution which provides a simple, unified way for the implementor of an object to encode the requirements for both equality and inequality. Equality substitution allows for objects to be used as keys in a hash table according to the custom equality, without the programmer being burdened with the responsibility of developing a custom hashing function.
An object participates in equality substitution by implementing the equal method. The equal method takes no arguments other than the object itself. It returns a representative value which is used in place of that object for the purposes of equal comparison.
Whenever an object which supports equality substitution is used as an argument of any of the functions equal, nequal, greater, less, gequal, lequal or hash-equal, the equal method of that object is invoked, and the return value of that method is taken in place of that object.
The same is true if an object which supports equality substitution is used as a key in an :equal-based hash table.
The substitution is applied repeatedly: if the return value of the object's equal method is an object which itself supports equality substitution, than that returned object's method is invoked on that object to fetch its equality substitute. This repeats as many times as necessary until an object is determined which isn't a structure that supports equality substitution.
Once the equality substitute is determined, then the given function proceeds with the replacement object. Thus for example equal compares the replacement object in place of the original, and an :equal-based hash table uses the replacement object as the key for the purposes of hashing and comparison.
The defstruct macro has a provision for for application-defined clauses, which may be defined using the define-struct-clause macro. This macro associates new clause keywords with custom expansion. The :delegate clause of defstruct is in fact implemented externally to defstruct using define-struct-clause.
The defstruct macro has a provision for implicit inclusion of application-defined clauses called preludes, which are previously defined via the define-struct-prelude macro. During macro-expansion, defstruct checks whether the structure being defined is the target of one or more preludes. If so, it includes the clauses from those preludes as if they were written directly in the defstruct syntax.
(defstruct {name | (name arg*)} super
slot-specifier*)
The defstruct macro defines a new structure type and registers it under name, which must be a bindable symbol, according to the bindable function. Likewise, the name of every slot must also be a bindable symbol.
The super argument must either be nil, or a symbol which names an existing struct type, or else a list of such symbols. The newly defined struct type will inherit all slots, as well as initialization behaviors from the specified struct types.
The defstruct macro is implemented using the make-struct-type function, which is more general. The macro analyzes the defstruct argument syntax, and synthesizes arguments which are then used to call the function. Some remarks in the description of defstruct only apply to structure types defined using that macro.
Slots are specified using zero or more slot-specifier clauses.
Application-defined clauses are possible via define-struct-clause. The defstruct macro may bring in prelude clauses which are not specified in its syntax, but that have been specified using define-struct-prelude.
The following built-in clauses are supported:
The definition of a static slot in a defstruct causes the new type to have its own instance that slot, even if a same-named static slot occurs in the super base type, or its bases.
Due to the semantics of static slots, methods are naturally inherited from a base structure to a derived one, and defining a method in a derived class which also exists in a base class performs OOP-style overriding.
The remarks about inheritance and overriding in the description of :method also apply to :function.
Multiple :init specifiers may appear in the same defstruct form. They are executed in their order of appearance, left to right.
When an object with one or more levels of inheritance is instantiated, the :init code of a base structure type, if any, is executed before any initializations specific to a derived structure type. Under multiple inheritance, the :init code of the rightmost base type is executed first, then that of the remaining bases in right-to-left order.
The :init initializations are executed before any other slot initializations. The argument values passed to the new or lnew operator or the make-struct function are not yet stored in the object's slots, and are not accessible. Initialization code which needs these values to be stable can be defined with :postinit.
Initializers in base structures must be careful about assumptions about slot kinds, because derived structures can alter static slots to instance slots or vice versa. To avoid an unwanted initialization being applied to the wrong kind of slot, initialization code can be made conditional on the outcome of static-slot-p applied to the slot. (Code generated by defstruct for initializing instance slots performs this kind of check).
The body-forms of an :init specifier are not surrounded by an implicit block.
The body-forms of a :fini specifier are not surrounded by an implicit block.
Multiple :fini clauses may be specified in the same defstruct, in which case they are invoked in reverse, right-to-left order.
Note that an object's finalizers can be called explicitly with call-finalizers. Note: the with-objects macro arranges for finalizers to be called on objects when the execution of a scope terminates by any means.
When both :fini and :postfini clauses are specified in the same defstruct form, all the :postfini finalizers execute after all the :fini finalizers regardless of the order in which they appear.
A given structure type can have only one slot under a given symbolic name. If a newly specified slot matches the name of an existing slot in the super type or that type's chain of ancestors, it is called a repeated slot.
The kind of the repeated slot (static or instance) is not inherited; it is established by the defstruct and may be different from the type of the same-named slot in the supertype or its ancestors.
If a repeated slot is introduced as a static slot, and has no init-form then it receives the current of the a static of the same name from the nearest supertype which has such a slot.
If a repeated slot is an instance slot, no such inheritance of value takes place; only the local init-form applies to it; if it is absent, the slot it initialized to nil in each newly created instance of the new type.
However, :init and :postinit initializations are inherited from a base type and they apply to the repeated slots, regardless of their kind. These initializations take place on the instantiated object, and the slot references resolve accordingly.
The initialization for slots which are specified using the :method or :function specifiers is reordered with regard to :static slots. Regardless of their placement in the defstruct form, :method and :function slots are initialized before :static slots. This ordering is useful, because it means that when the initialization expression for a given static slot constructs an instance of the struct type, any instance initialization code executing for that instance can use all functions and methods of the struct type. However, note the static slots which follow that slot in the defstruct syntax are not yet initialized. If it is necessary for a structure's initialization code to have access to all static slots, even when the structure is instantiated during the initialization of a static slot, a possible solution may be to use lazy instantiation using the lnew operator, rather than ordinary eager instantiation via new. It is also necessary to ensure that that the instance isn't accessed until all static initializations are complete, since access to the instance slots of a lazily instantiated structure triggers its initialization.
The structure name is specified using two forms, plain name or the syntax (name arg*) If the second form is used, then the structure type will support "boa construction", where "boa" stands for "by order of arguments". The args specify the list of slot names which are to be initialized in the by-order-of-arguments style. For instance, if three slot names are given, then those slots can be optionally initialized by giving three arguments in the new macro or the make-struct function.
Slots are first initialized according to their init-forms, regardless of whether they are involved in boa construction.
A slot initialized in this style still has a init-form which is processed independently of the existence of, and prior to, boa construction.
The boa constructor syntax can specify optional parameters, delimited by a colon, similarly to the lambda syntax. However, the optional parameters may not be arbitrary symbols; they must be symbols which name slots. Moreover, the (name init-form [present-p]) optional parameter syntax isn't supported.
When boa construction is invoked with optional arguments missing, the default values for those arguments come from the init-forms in the remaining defstruct syntax.
(defvar *counter* 0)
;; New struct type foo with no super type:
;; Slots a and b initialize to nil.
;; Slot c is initialized by value of (inc *counter*).
(defstruct foo nil (a b (c (inc *counter*))))
(new foo) -> #S(foo a nil b nil c 1)
(new foo) -> #S(foo a nil b nil c 2)
;; New struct bar inheriting from foo.
(defstruct bar foo (c 0) (d 100))
(new bar) -> #S(bar a nil b nil c 0 d 100)
(new bar) -> #S(bar a nil b nil c 0 d 100)
;; counter was still incremented during
;; construction of d:
*counter* -> 4
;; override slots with new arguments
(new foo a "str" c 17) -> #S(foo a "str" b nil c 17)
*counter* -> 5
;; boa initialization
(defstruct (point x : y) nil (x 0) (y 0))
(new point) -> #S(point x 0 y 0)
(new (point 1 1)) -> #S(point x 1 y 1)
;; property list style initialization
;; can always be used:
(new point x 4 y 5) -> #S(point x 4 y 5)
;; boa applies last:
(new (point 1 1) x 4 y 5) -> #S(point x 1 y 1)
;; boa with optional argument omitted:
(new (point 1)) -> #S(point x 1 y 0)
;; boa with optional argument omitted and
;; with property list style initialization:
(new (point 1) x 5 y 5) -> #S(point x 1 y 5)
(defmeth type-name name param-list body-form*)
Unless name is one of the two symbols :init or :postinit, the defmeth macro installs a function into the static slot named by the symbol name in the struct type indicated by type-name.
If the structure type doesn't already have such a static slot, it is first added, as if by the static-slot-ensure function, subject to the same checks.
If the function has at least one argument, it can be used as a method. In that situation, the leftmost argument passes the structure instance on which the method is being invoked.
The function takes the arguments specified by the param-list symbols, and its body consists of the body-forms.
The body-forms are placed into a block named name.
A method named lambda allows a structure to be used as if it were a function. When a structure is applied to arguments, as if it were a function, the lambda method is invoked with those arguments, with the object itself inserted into the leftmost argument position.
If defmeth is used to redefine an existing method, the semantics can be inferred from that of static-slot-ensure. In particular, the method will be imposed into all subtypes which inherit (do not override) the method.
If name is the keyword symbol :init, then instead of operating on a static slot, the macro redefines the initfun of the given structure type, as if by a call to the function struct-set-initfun.
Similarly, if name is the keyword symbol :postinit, then the macro redefines the postinitfun of the given structure type, as if by a call to the function struct-set-postinitfun.
When redefining :initfun the admonishments given in the description of struct-set-initfun apply: if the type has an initfun generated by the defstruct macro, then that initfun is what implements all of the slot initializations given in the slot specifier syntax. These initializations are lost if the initfun is overwritten.
The defmeth macro returns a method name: a unit of syntax of the form (meth type-name name) which can be used as an argument to the accessor symbol-function and other situations.
(new {name | (name arg*)} {slot init-form}*)
(lnew {name | (name arg*)} {slot init-form}*)
The new macro creates a new instance of the structure type named by name.
If the structure supports "boa construction", then, optionally, the arguments may be given using the syntax (name arg*) instead of name.
Slot values may also be specified by the slot and init-form arguments.
Note: the evaluation order in new is surprising: namely, init-forms are evaluated before args if both are present.
When the object is constructed, all default initializations take place first. If the object's structure type has a supertype, then the supertype initializations take place. Then the type's initializations take place, followed by the slot init-form overrides from the new macro, and lastly the "boa constructor" overrides.
If any of the initializations abandon the evaluation of new by a nonlocal exit such as an exception throw, the object's finalizers, if any, are invoked.
The macro lnew differs from new in that it specifies the construction of a lazy struct, as if by the make-lazy-struct function. When lnew is used to construct an instance, a lazy struct is returned immediately, without evaluating any of the arg and init-form expressions. The expressions are evaluated when any of the object's instance slots is accessed for the first time. At that time, these expressions are evaluated (in the same order as under new) and initialization proceeds in the same way.
If any of the initializations abandon the delayed initializations steps arranged by lnew by a nonlocal exit such as an exception throw, the object's finalizers, if any, are invoked.
Lazy initialization does not detect cycles. Immediately prior to the lazy initialization of a struct, the struct is marked as no longer requiring initialization. Thus, during initialization, its instance slots may be freely accessed. Slots not yet initialized evaluate as nil.
(new* {expr | (expr arg*)} {slot init-form}*)
(lnew* {expr | (expr arg*)} {slot init-form}*)
The new* and lnew* macros are variants, respectively, of new and lnew.
The difference in behavior in these macros relative to new and lnew is that the name argument is replaced with an expression expr which is evaluated. The value of expr must be a struct type, or a symbol which is the name of a struct type.
With one exception, if expr0 is a compound expression, then (new* expr0 ...) is interpreted as (new* (expr1 args...) ...) where the head of expr0, expr1, is actually the expression which is evaluated to produce the type, and the remaining constituents of expr0, args, become the boa arguments. The same requirement applies to lnew*.
The exception is that if expr1 is the symbol dwim, this interpretation does not apply. Thus (new* [fun args...] ...) evaluates the [fun args...] expression, rather than treating it as (dwim fun args...) where dwim would be evaluated as a variable reference expected to produce a type.
;; struct with boa constructor
(defstruct (ab a : b) () a b)
;; error: find-struct-type is interpreted as a variable
(new* (find-struct-type 'ab) a 1) -> ;; error
;; OK: extra nesting.
(new* ((find-struct-type 'ab)) a 1) -> #S(ab a 1 b nil)
;; OK: dwim brackets without nesting.
(new* [find-struct-type 'ab] a 1) -> #S(ab a 1 b nil)
;; boa construction
(new* ([find-struct-type 'ab] 1 2)) -> #S(ab a 1 b 2)
(new* ((find-struct-type 'ab) 1 2)) -> #S(ab a 1 b 2)
;; mixed construction
(new* ([find-struct-type 'ab] 1) b 2) -> #S(ab a 1 b 2)
(let ((type (find-struct-type 'ab)))
(new* type a 3 b 4))
-> #S(ab a 3 b 4)
(let ((type (find-struct-type 'ab)))
(new* (type 3 4)))
-> #S(ab a 3 b 4)
(with-slots ({slot | (sym slot)}*) struct-expr
body-form*)
The with-slots binds lexical macros to serve as aliases for the slots of a structure.
The struct-expr argument is expected to be an expression which evaluates to a struct object. It is evaluated once, and its value is retained. The aliases are then established to the slots of the resulting struct value.
The aliases are specified as zero or more expressions which consist of either a single symbol slot or a (sym slot) pair. The simple form binds a macro named slot to a slot also named slot. The pair form binds a macro named sym to a slot named slot.
The lexical aliases are syntactic places: assigning to an alias causes the value to be stored into the slot which it denotes.
After evaluating struct-expr the with-slots macro arranges for the evaluation of body-forms in the lexical scope in which the aliases are visible.
The intent of the with-slots macro is to help reduce the verbosity of code which makes multiple references to the same slot. Use of with-slots is less necessary in TXR Lisp than other Lisp dialects thanks to the dot operator for accessing struct slots.
Lexical aliases to struct places can also be arranged with considerable convenience using the placelet operator. However, placelet will not bind multiple aliases to multiple slots of the same object such that the expression which produces the object is evaluated only once.
(defstruct point nil x y)
;; Here, with-slots introduces verbosity because
;; each slot is accessed only once. The function
;; is equivalent to:
;;
;; (defun point-delta (p0 p1)
;; (new point x (- p1.x p0.x) y (- p1.y p0.y)))
;;
;; Also contrast with the use of placelet:
;;
;; (defun point-delta (p0 p1)
;; (placelet ((x0 p0.x) (y0 p0.y)
;; (x1 p1.x) (y1 p1.y))
;; (new point x (- x1 x0) y (- y1 y0)))))
(defun point-delta (p0 p1)
(with-slots ((x0 x) (y0 y)) p0
(with-slots ((x1 x) (y1 y)) p1
(new point x (- x1 x0) y (- y1 y0)))))
(qref object-form
{slot | (slot arg*) | [slot arg*]}+)
The qref macro ("quoted reference") performs structure slot access. Structure slot access is more conveniently expressed using the referencing dot notation, which works by translating to qref qref syntax, according to the following equivalence:
a.b.c.d <--> (qref a b c d) ;; a b c d must not be numbers
(See the Referencing Dot section under Additional Syntax.)
The leftmost argument of qref is an expression which is evaluated. This argument is followed by one or more reference designators. If there are two or more designators, the following equivalence applies:
(qref obj d1 d2 ...) <---> (qref (qref obj d1) d2 ...)
That is to say, qref is applied to the object and a single designator. This must yield an object, which to which the next designator is applied as if by another qref operation, and so forth.
If the null-safe syntax (t ...) is present, the equivalence becomes more complicated:
(qref (t obj) d1 d2 ...) <---> (qref (qref (t obj) d1) d2 ...)
(qref obj (t d1) d2 ...) <---> (qref (t (qref obj d1)) d2 ...)
Thus, qref can be understood in terms of the semantics of the binary form (qref object-form designator)
Designators come in three basic forms: a lone symbol, an ordinary compound expression consisting of a symbol followed by arguments, or a DWIM expression consisting of a symbol followed by arguments.
A lone symbol designator indicates the slot of that name. That is to say, the following equivalence applies:
where slot is the structure slot accessor function. Because slot is an accessor, this form denotes the slot as a syntactic place; slots can be modified via assignment to the qref form and the referencing dot syntax.
The slot name being implicitly quoted is the basis of the term "quoted reference", giving rise to the qref name.
A compound designator indicates that the named slot is a function, which is to be applied to arguments. The following equivalence applies in this case, except that o is evaluated only once:
(qref o (n arg ...)) <--> (call (slot o 'n) o arg ...)
A DWIM designator similarly indicates that the named slot is a function, which is to be applied to arguments. The following equivalence applies:
(qref obj [name arg ...]) <--> [(slot obj 'name) o arg ...]
Therefore, under this equivalence, this syntax provides the usual Lisp-1-style evaluation rule via the dwim operator.
If the object-form has the syntax (t expression) this indicates null-safe access: if expression evaluates to nil then the entire expression (qref (t expression) designator) form yields nil. This syntax is produced by the .? notation.
The null-safe access notation prevents not only slot access, but also method or function calls on nil. When a method or function call is suppressed due to the object being nil, no aspect of the method or function call is evaluated; not only is the slot not accessed, but the argument expressions are not evaluated.
(defstruct foo nil
(array (vec 1 2 3))
(increment (lambda (self index delta)
(inc [self.array index] delta))))
(defvarl s (new foo))
;; access third element of s.array:
[s.array 2] --> 3
;; increment first element of array by 42
s.(increment 0 42) --> 43
;; access array member
s.array --> #(43 2 3)
Note how increment behaves much like a single-argument-dispatch object-oriented method. Firstly, the syntax s.(increment 0 42) effectively selects the increment function which is particular to the s object. Secondly, the object is passed to the selected function as the leftmost argument, so that the function has access to the object.
(uref {slot | (slot arg*) | [slot arg*]}+)
The uref macro ("unbound reference") expands to an expression which evaluates to a function. The function takes exactly one argument: an object. When the function is invoked on an object, it references slots or methods relative to that object.
Note: the uref syntax may be used directly, but it is also produced by the unbound referencing dot syntactic sugar:
.a --> (uref a)
.?a --> (uref t a)
.(f x) --> (uref (f x))
.(f x).b --> (uref (f x) b)
.a.(f x).b --> (uref a (f x) b)
The macro may be understood in terms of the following translation scheme:
(uref a b ...) --> (lambda (o) (qref o a b ...))
(uref t a b ...) --> (lambda (o) (if o (qref o a b ...)))
where o is understood to be a unique symbol (for instance, as produced by the gensym function).
When only one uref argument is present, these equivalences also hold:
(uref (f a b c ...)) <--> (umeth f a b c ...)
(uref s) <--> (usl s)
The terminology "unbound reference" refers to the property that uref expressions produce a function which isn't bound to a structure object. The function binds a slot or method; the call to that function then binds an object to that function, as an argument.
Suppose that the objects in list have slots a and b. Then, a list of the a slot values may be obtained using:
(mapcar .a list)
because this is equivalent to
(mapcar (lambda (o) o.a) list)
Because uref produces a function, its result can be operated upon by functional combinators. For instance, we can use the juxt combinator to produce a list of two-element lists, which hold the a and b slots from each object in list:
(meth struct slot curried-expr*)
The meth macro allows indirection upon a method-like function stored in a function slot.
The meth macro binds struct as the leftmost argument of the function stored in slot, returning a function which takes the remaining arguments. That is to say, it returns a function f such that [f arg ...] calls [struct.slot struct arg ...] except that struct is evaluated only once.
If one or more curried-expr expressions are present, their values are bound inside f also, and when f is invoked, these are passed to the function stored in the slot. Thus if f is produced by (meth struct slot c1 c2 c3 ...) then [f arg ...] calls [struct.slot struct c1v c2v c3v ... arg ...] except that struct is evaluated only once, and c1v, c2v and c3v are the values of expressions c1, c2 and c3.
The argument struct must be an expression which evaluates to a struct. The slot argument is not evaluated, and must be a symbol denoting a slot. The syntax can be understood as a translation to a call of the method function:
If curried-arg expressions are present, the translation may be be understood as:
(meth a b c1 c2 ...) <--> [(fun method) a 'b c1 c2 ...]
In other words the curried-arg expressions are evaluated under the dwim operator evaluation rules.
;; struct for counting atoms eq to key
(defstruct (counter key) nil
key
(count 0)
(:method increment (self key)
(if (eq self.key key)
(inc self.count))))
;; pass all atoms in tree to func
(defun map-tree (tree func)
(if (atom tree)
[func tree]
(progn (map-tree (car tree) func)
(map-tree (cdr tree) func))))
;; count occurrences of symbol a
;; using increment method of counter,
;; passed as func argument to map-tree.
(let ((c (new (counter 'a)))
(tr '(a (b (a a)) c a d)))
(map-tree tr (meth c increment))
c)
--> #S(counter key a count 4
increment #<function: type 0>)
(umeth slot curried-expr*)
The umeth macro binds the symbol slot to a function and returns that function.
The curried-expr arguments, if present, are evaluated as if they were arguments to the dwim operator.
When that function is called, it expects at least one argument. The leftmost argument must be an object of struct type.
The slot named slot is retrieved from that object, and is expected to be a function. That function is called with the object, followed by the values of the curried-exprs, if any, followed by that function's arguments.
The syntax can be understood as a translation to a call of the umethod function:
(umeth s ...) <--> [umethod 's ...]
The macro merely provides the syntactic sugar of not having to quote the symbol, and automatically treating the curried argument expressions using Lisp-1 semantics of the dwim operator.
;; seal and dog are variables which hold structures of
;; different types. Both have a method called bark.
(let ((bark-fun (umeth bark)))
[bark-fun dog] ;; same effect as dog.(bark)
[bark-fun seal]) ;; same effect as seal.(bark)
The u in umeth stands for "unbound". The function produced by umeth is not bound to any specific object; it binds to an object whenever it is invoked by retrieving the actual method from the object's slot at call time.
(usl slot)
The usl macro binds the symbol slot to a function and returns that function.
When that function is called, it expects exactly one argument. That argument must be an object of struct type. The slot named slot is retrieved from that object and returned.
The name usl stands for "unbound slot". The term "unbound" refers to the returned function not being bound to a particular object. The binding of the slot to an object takes place whenever the function is called.
(make-struct-type name super static-slots slots
static-initfun initfun boactor
boactor postinitfun)
The make-struct-type function creates a new struct type.
The name argument must be a bindable symbol, according to the bindable function. It specifies the name property of the struct type as well as the name under which the struct type is globally registered.
The super argument indicates the supertype for the struct type. It must be either a value of type struct-type, a symbol which names a struct type, or else nil, indicating that the newly created struct type has no supertype.
The static-slots argument is a list of symbol which specify static slots. The symbols must be bindable and the list must not contain duplicates.
The slots argument is a list of symbols which specifies the instance slots. The symbols must be bindable and there must not be any duplicates within the list, or against entries in the static-slots list.
The new struct type's effective list of slots is formed by appending together static-slots and slots, and then appending that to the list of the supertype's slots, and de-duplicating the resulting list as if by the uniq function. Thus, any slots which are already present in the supertype are removed. If the structure has no supertype, then the list of supertype slots is taken to be empty. When a structure is instantiated, it shall have all the slots specified in the effective list of slots. Each instance slot shall be initialized to the value nil, prior to the invocation of initfun and boactor.
The static-initfun argument either specifies an initialization function, or is nil, which is equivalent to specifying a function which does nothing.
Prior to the invocation of static-initfun, each new static slot shall be initialized the value nil. Inherited static slots retain their values from the supertype.
If specified, static-initfun function must accept one argument. When the structure type is created (before the make-struct-type function returns) the static-initfun function is invoked, passed the newly created structure type as its argument.
The initfun argument either specifies an initialization function, or is nil, which is equivalent to specifying a function which does nothing. If specified, this function must accept one argument. When a structure is instantiated, every initfun in its chain of supertype ancestry is invoked, in order of inheritance, so that the root supertype's initfun is called first and the structure's own specific initfun is called last. These calls occur before the slots are initialized from the arg arguments or the slot-init-plist of make-struct. Each function is passed the newly created structure object, and may alter its slots. If multiple inheritance occurs, the initfun functions of multiple supertypes are called in right-to-left order.
The boactor argument either specifies a by-order-of-arguments initialization function ("boa constructor") or is nil, which is equivalent to specifying a constructor which does nothing. If specified, it must be a function which takes at least one argument. When a structure is instantiated, and boa arguments are given, the boactor is invoked, with the structure as the leftmost argument, and the boa arguments as additional arguments. This takes place after the processing of initfun functions, and after the processing of the slot-init-plist specified in the make-struct call. Note that the boactor functions of the supertypes are not called, only the boactor specific to the type being constructed.
The postinitfun argument either specifies an initialization function, or is nil, which is equivalent to specifying a function which does nothing. If specified, this function must accept one argument. The postinitfun function is similar to initfun. The difference is that postinitfun functions are called after all other initialization processing, rather than before. They are are also called in order of inheritance: the postinitfun of a structure's supertype is called before its own, and in right-to-left order among multiple supertypes under multiple inheritance.
(find-struct-type name)
The find-struct-type returns a struct-type object corresponding to the symbol name.
If no struct type is registered under name, then it returns nil.
A struct-type object exists for each structure type and holds information about it. These objects are not themselves structures and are all of the same type, struct-type.
(struct-type-p obj)
The struct-type-p function returns t if obj is a structure type, otherwise it returns nil.
A structure type is an object of type struct-type, returned by find-struct-type.
(struct-type-name type-or-struct)
The struct-type-name function determines a structure type from the type-or-struct argument and returns that structure type's symbolic name.
The type-or-struct argument must be either a struct type object (such as the return value of a successful lookup via find-struct-type), a symbol which names a struct type, or else a struct instance.
(super [type-or-struct])
The super function determines a structure type from the type-or-struct argument and returns the struct type object which is the supertype of that type, or else nil if that type has no supertype.
The type-or-struct argument must be either a struct type object, a symbol which names a struct type, or else a struct instance.
(make-struct type slot-init-plist arg*)
The make-struct function returns a new object which is an instance of the structure type type.
The type argument must either be a struct-type object, or else a symbol which is the name of a structure.
The slot-init-plist argument gives a list of slot initializations in the style of a property list, as defined by the prop function. It may be empty, in which case it has no effect. Otherwise, it specifies slot names and their values. Each slot name which is given must be a slot of the structure type. The corresponding value will be stored into the slot of the newly created object. If a slot is repeated, it is unspecified which value takes effect.
The optional args specify arguments to the structure type's boa constructor. If the arguments are omitted, the boa constructor is not invoked. Otherwise the boa constructor is invoked on the structure object and those arguments. The argument list must match the trailing parameters of the boa constructor (the remaining parameters which follow the leftmost argument which passes the structure to the boa constructor).
When a new structure is instantiated by make-struct, its slot values are first initialized by the structure type's registered functions as described under make-struct-type. Then, the slot-init-plist is processed, if not empty, and finally, the args are processed, if present, and passed to the boa constructor.
If any of the initializations abandon the evaluation of make-struct by a nonlocal exit such as an exception throw, the object's finalizers, if any, are invoked.
(make-lazy-struct type argfun)
The make-lazy-struct function returns a new object which is an instance of the structure type type.
The type argument must either be a struct-type object, or else a symbol which is the name of a structure.
The argfun argument should be a function which can be called with no parameters and returns a cons cell. More requirements are specified below.
The object returned by make-lazy-struct is a lazily-initialized struct instance, or lazy struct.
A lazy struct remains uninitialized until just before the first access to any of its instance slots. Just before an instance slot is accessed, initialization takes place as follows. The argfun function is invoked with no arguments. Its return value must be a cons cell. The car of the cons cell is taken to be a property list, as defined by the prop function. The cdr field is taken to be a list of arguments. These values are treated as if they were, respectively, the slot-init-plist and the boa constructor arguments given in a make-struct invocation. Initialization of the structure proceeds as described in the description of make-struct.
(struct-from-plist type {slot value}*)
(struct-from-args type arg*)
The struct-from-plist and struct-from-args functions are interfaces to the make-struct function.
The struct-from-plist function passes its slot and value arguments as the slot-init-plist argument of make-struct. It passes no boa constructor arguments.
The struct-from-args function calls make-struct with an empty slot-init-plist, passing down the list of args.
The following equivalences hold:
(struct-from-plist a s0 v0 s1 v1 ...)
<--> (make-struct a (list s0 v0 s1 v1 ...))
(struct-from-args a v0 v1 v2 ...)
<--> (make-struct a nil v0 v1 v2 ...)
(allocate-struct type)
The allocate-struct provides a low-level allocator for structure objects.
The type argument must either be a struct-type object, or else a symbol which is the name of a structure.
The allocate-struct creates and returns a new instance of type all of whose instance slots take on the value nil. No initializations are performed. The struct type's registered initialization functions are not invoked.
(copy-struct struct-obj)
The copy-struct function creates and returns a new object which is a duplicate of struct-obj, which must be a structure.
The duplicate object is a structure of the same type as struct-obj and has the same slot values.
The creation of a duplicate does not involve calling any of the struct type's initialization functions.
Only instance slots participate in the duplication. Since the original structure and copy are of the same structure type, they already share static slots.
This is a low-level, "shallow" copying mechanism. If an object design calls for a higher level cloning mechanism with deep copying or other additional semantics, one can be built on top of copy-struct. For instance, a structure can have a copy method similar to the following:
(:method copy (me)
(let ((my-copy (copy-struct me)))
;; inform the copy that it has been created
;; by invoking its copied method.
my-copy.(copied)
my-copy))
which can then be invoked on whatever object needs copying.
Note that a method named copy is a special structure function. When an object provides this method, the copy function uses the method to copy the object, rather than using copy-struct.
Since this logic is generic, it can be placed in a base method. The copied method which it calls is the means by which the new object is notified that it is a copy. This method takes on whatever special responsibilities are required when a copy is produced, such as registering the object in various necessary associations, or performing a deeper copy of some of the objects held in the slots.
The copied handler can be implemented at multiple levels of an inheritance hierarchy. The initial call to copied from copy will call the most derived override of that method.
To call the corresponding method in the base class, a given derived method can use the call-super-fun function, or else the (meth ...) syntax in the first position of a compound form, in place of a function name. Examples of both are given in the documentation for call-super-fun.
Thus derived structs can inherit the copy handling logic from base structs, and extend it with their own.
(slot struct-obj slot-name)
(set (slot struct-obj slot-name) new-value)
The slot function retrieves a structure's slot. The struct-obj argument must be a structure, and slot-name must be a symbol which names a slot in that structure.
Because slot is an accessor, a slot form is a syntactic place which denotes the slot's storage location.
A syntactic place expressed by slot does not support deletion.
(slotset struct-obj slot-name new-value)
The
slotset
function stores a value in a structure's slot.
The
struct-obj
argument must be a structure, and
slot-name
must be a symbol which names a slot in that structure.
The new-value argument specifies the value to be stored in the slot.
If a successful store takes place to an instance slot of struct-obj, then the dirty flag of that object is set, causing the test-dirty function to report true for that object.
The slotset function returns new-value.
(test-dirty struct-obj)
(clear-dirty struct-obj)
(test-clear-dirty struct-obj)
The test-dirty, clear-dirty and test-clear-dirty functions comprise the interface for interacting with structure dirty flags.
Each structure instance has a dirty flag. When this flag is set, the structure instance is said to be dirty, otherwise it is said to be clean. A newly created structure is dirty. A structure remains dirty until its dirty flag is explicitly reset. If a structure is clean, and one of its instance slots is overwritten with a new value, it becomes dirty.
The test-dirty function returns the dirty flag of struct-obj: t if struct-obj is dirty, otherwise nil.
The clear-dirty function clears the dirty flag of struct-obj and returns struct-obj itself.
The test-clear-dirty flag combines these operations: it makes a note of the dirty flag of struct-obj and clears it. Then it returns the noted value, t or nil.
(structp obj)
The structp function returns t if obj is a structure, otherwise it returns nil.
(struct-type struct-obj)
The struct-type function returns the structure type object which represents the type of the structure object instance struct-obj.
(clear-struct struct-obj [value])
The clear-struct replaces all instance slots of struct-obj with value, which defaults to nil if omitted.
Note that finalizers are not executed prior to replacing the slot values.
(reset-struct struct-obj)
The reset-struct function reinitializes the structure object struct-obj as if it were being newly created. First, all the slots are set to nil as if by the clear-struct function. Then the slots are initialized by invoking the initialization functions, in order of the supertype ancestry, just as would be done for a new structure object created by make-struct with an empty slot-init-plist and no boa arguments.
Note that finalizers registered against struct-obj are not invoked prior to the reset operation, and remain registered.
If the structure has state which is cleaned up by finalizers, it is advisable to invoke them using call-finalizers prior to using reset-struct, or to take other measures to deal with the situation.
If the structure specifies :fini handlers, then the reinitialization will cause these to registered, just like when a new object it constructed. Thus if call-finalizers is not used prior to reset-struct, this will result in the existence of duplicate registrations of the finalization functions.
Finalizers registered against struct-obj are invoked if an exception is thrown during the reinitialization, just like when a new structure is being constructed.
(replace-struct target-obj source-obj)
The replace-struct function causes target-obj to take on the attributes of source-obj without changing its identity.
The type of target-obj is changed to that of source-obj.
All instance slots of target-obj are discarded, and it is given new slots, which are copies of the instance slots of source-obj.
Because of the type change, target-obj implicitly loses all of its original static slots, and acquires those of source obj.
Note that finalizers registered against target-obj are not invoked, and remain registered. If target-obj has state which is cleaned up by finalizers, it is advisable to invoke them using call-finalizers prior to using replace-struct, or to take other measures to handle the situation.
If the target-obj and source-obj arguments are the same object, replace-struct has no effect.
The return value is target-obj.
(method struct-obj slot-name curried-arg*)
The method function retrieves a function m from a structure's slot and returns a new function which binds that function's left argument. If curried-arg arguments are present, then they are also stored in the returned function. These are the curried arguments.
The struct-obj argument must be a structure, and slot-name must be a symbol denoting a slot in that structure. The slot must hold a function of at least one argument.
The function f which method function returns, when invoked, calls the function m previously retrieved from the object's slot, passing to that function struct-obj as the leftmost argument, followed by the curried arguments, followed by all of f's own arguments.
Note: the meth macro is an alternative interface which is suitable if the slot name isn't a computed value.
(super-method struct-obj slot-name)
The super-method function retrieves a function from a static slot belonging to one of the direct supertypes of the structure type of struct-obj.
It then returns a function which binds that function's left argument to the structure.
The struct-obj argument must be a structure which has at least one supertype, and slot-name must be a symbol denoting a static slot in one of those supertypes. The slot must hold a function of at least one argument. The supertypes are searched from left to right for a static slot named slot-name; when the first such slot is found, its value is used.
The super-method function returns a function which, when invoked, calls the function previously retrieved from the supertype's static slot, passing to that function struct-obj as the leftmost argument, followed by the function's own arguments.
(umethod slot-name curried-arg*)
The umethod returns a function which represents the set of all methods named by the slot slot-name in all structure types, including ones not yet defined. The slot-name argument must be a symbol.
If one or more curried-arg argument are present, these values represent the curried arguments which are stored in the function object which is returned.
This returned function must be called with at least one argument. Its leftmost argument must be an object of structure type, which has a slot named slot-name. The function will retrieve the value of the slot from that object, expecting it to be a function, and calls it, passing to it the following arguments: the object itself; all of the curried arguments, if any; and all of its remaining arguments.
Note: the umethod name stands for "unbound method". Unlike the method function, umethod doesn't return a method whose leftmost argument is already bound to an object; the binding occurs at call time.
(uslot slot-name)
The uslot returns a function which represents all slots named slot-name in all structure types, including ones not yet defined. The slot-name argument must be a symbol.
The returned function must be called with exactly one argument. The argument must be a structure which has a slot named slot-name. The function will retrieve the value of the slot from that object and return it.
Note: the uslot name stands for "unbound slot". The returned function isn't bound to a particular object. The binding of slot-name to a slot in the structure object occurs when the function is called.
(slots type)
The slots function returns a list of all of the slots of struct type type.
The type argument must be a structure type, or else a symbol which names a structure type.
(slotp type name)
The slotp function returns t if name name is a symbol which names a slot in the structure type type. Otherwise it returns nil.
The type argument must be a structure type, or else a symbol which names a structure type.
(static-slot-p type name)
The static-slot-p function returns t if name name is a symbol which names a slot in the structure type type, and if that slot is a static slot. Otherwise it returns nil.
The type argument must be a structure type, or else a symbol which names a structure type.
(static-slot type name)
The static-slot function retrieves the value of the static slot named by symbol name of the structure type type.
The type argument must be a structure type or a symbol which names a structure type, and name must be a static slot of this type.
(static-slot-set type name new-value)
The static-slot-set function stores new-value into the static slot named by symbol name of the structure type type.
It returns new-value.
The type argument must be a structure type or the name of a structure type, and name must be a static slot of this type.
(static-slot-ensure type name new-value [no-error-p])
The static-slot-ensure ensures, if possible, that the struct type type, as well as possibly one or more struct types derived from it, have a static slot called name, that this slot is not shared with a supertype, and that the value stored in it is new-value.
Note: this function supports the redefinition of methods, as the implementation underlying the defmeth macro; its semantics is designed to harmonize with expected behaviors in that usage.
The function operates as follows.
If type itself already has an instance slot called name then an error is thrown, and the function has no effect, unless a true argument is specified for the no-error-p Boolean parameter. In that case, in the same situation, the function has no effect and simply returns new-value.
If type already has a non-inherited static slot called name then this slot is overwritten with new-value and the function returns new-value. Types derived from type may also have this slot, via inheritance; consequently, its value changes in those types also.
If type already has an inherited static slot called name then its inheritance is severed; the slot is converted to a non-inherited static slot of type and initialized with new-value. Then all struct types derived from type are scanned. In each such type, if the original inherited static slot is found, it is replaced with the same newly converted static slot that was just introduced into type, so that all these types now inherit this new slot from type rather than the original slot from some supertype of type. These types all share a single instance of the slot with type, but not with supertypes of type.
In the remaining case, type has no slot called name. The slot is added as a static slot to type. Then it is added to every struct type derived from type which does not already have a slot by that name, as if by inheritance. That is to say, types to which this slot is introduced share a single instance of that slot. The value of the new slot is new-value, which is also returned from the function. Any subtypes of type which already have a slot called name are ignored, as are their subtypes.
(static-slot-home type name)
The static-slot-home method determines which structure type actually defines the static slot name present in struct type type.
If type isn't a struct type, or the name of a struct type, the function throws an error. Likewise, if name isn't a static slot of type.
If name is a static slot of type then the function returns a struct type name symbol which is either then name of type itself, if the slot is defined specifically for type or else the most distant ancestor of type from which the slot is inherited.
(call-super-method struct-obj name argument*)
The call-super-method function is deprecated. Solutions involving call-super-method should be reworked in terms of call-super-fun.
The call-super-method retrieves the function stored in the static slot name of one of the direct supertypes of struct-obj and invokes it, passing to that function struct-obj as the leftmost argument, followed by the given arguments, if any.
The struct-obj argument must be of structure type. Moreover, that structure type must be derived from one or more supertypes, and name must name a static slot available from at least one of those supertypes. The supertypes are searched left to right in search of this slot.
The object retrieved from that static slot must be callable as a function, and accept the arguments.
Note that it is not correct for a method that is defined against a particular type to use call-super-method to call the same method (or any other method) in the supertype of that particular type. This is because call-super-method refers to the type of the object instance struct-obj, not to the type against which the calling method is defined.
(call-super-fun type name argument*)
The call-super-fun retrieves the function stored in the slot name of one of the supertypes of type and invokes it, passing to that function the given arguments, if any.
The type argument must be a structure type. Moreover, that structure type must be derived from one or more supertypes, and name must name a static slot available from at least one of those supertypes. The supertypes are searched left to right in search of this slot.
The object retrieved from that static slot must be callable as a function, and accept the arguments.
Print a message and call supertype method:
(defstruct base nil)
(defstruct derived base)
(defmeth base fun (obj arg)
(format t "base fun method called with arg ~s\n" arg))
(defmeth derived fun (obj arg)
(format t "derived fun method called with arg ~s\n" arg)
(call-super-fun 'derived 'fun obj arg))
;; Interactive Listener:
1> (new derived).(fun 42)
derived fun method called with arg 42
base fun method called with arg 42
Note that a static method or function in any structure type can be invoked by using the (meth ...) name syntax in the first position of a compound form, as a function name. Thus, the above derived fun can also be written:
(defmeth derived fun (obj arg)
(format t "derived fun method called with arg ~s\n" arg)
((meth base fun) obj arg))
(struct-get-initfun type)
(struct-get-postinitfun type)
The struct-get-initfun and struct-get-postinitfun functions retrieve, respectively, a structure type's initfun and postinitfun functions. These are the functions which are initially configured in the call to make-struct-type via the initfun and postinitfun arguments.
Either one may be nil, indicating that the type has no initfun or postinitfun.
(struct-set-initfun type function)
(struct-set-postinitfun type function)
The struct-set-initfun and struct-set-postinitfun functions overwrite, respectively, a structure type's initfun and postinitfun functions. These are the functions which are initially configured in the call to make-struct-type via the initfun and postinitfun arguments.
The function argument must either be nil or else a function which accepts one argument.
Note that initfun has the responsibility for all instance slot initializations. The defstruct syntax compiles the initializing expressions in the slot specifier syntax into statements which are placed into a function, which becomes the initfun of the struct type.
(with-objects ({(sym init-form)}*) body-form*)
The with-objects macro provides a binding construct similar to let*.
Each sym must be a symbol suitable for use as a variable name.
Each init-form is evaluated in sequence, and a binding is established for its corresponding sym which is initialized with the value of that form. The binding is visible to subsequent init-forms.
Additionally, the values of the init-forms are noted as they are produced. When the with-objects form terminates, by any means, the call-finalizers function is invoked on each value which was returned by an init-form and had been noted. These calls are performed in the reverse order relative to the original evaluation of the forms.
After the variables are established and initialized, the body-forms are evaluated in the scope of the variables. The value of the last form is returned, or else nil if there are no forms. The invocations of call-finalizers take place just before the value of the last form is returned.
(define-struct-clause keyword params [body-form]*)
The define-struct-clause macro makes available a new, application-defined defstruct clause. The clause is named by keyword, which must be a keyword symbol, and is implemented as a macro transformation by the params and body-forms of the definition. The definition established by define-struct-clause is called a struct clause macro.
A struct clause macro is invoked when defstruct syntax is processed which contains one or more clauses which are headed by the matching keyword symbol.
The params comprise a macro-style parameter list which must match the invoking clause, otherwise an error exception is thrown. When params successfully matches the clause parameters, the parameters are destructured into the parameters and the body-forms are evaluated in the scope of those parameters.
The body-forms must return a possibly list of defstruct clauses, not a single clause.
Each of the returned clauses is examined for the possibility that it may be a struct clause macro; if so, it is expanded.
The built-in clause keywords :static, :instance, :function, :method, :init, :postinit, :fini and :postfini. may not be used as the names of a struct clause macro; if any of these symbols is used as the keyword parameter of define-struct-clause, an error exception is thrown.
The return value of a define-struct-clause macro invocation is the keyword argument.
;; Trivial struct clause macro which consumes any number of
;; arguments and produces no slots:
(define-struct-clause :nothing (. ignored-args))
;; Consequently, the following defines a struct with one slot, x:
;; The (:nothing ...) clause disappears by producing no clauses.
(defstruct foo ()
(:nothing 1 2 3 beeblebrox)
x)
;; struct clause macro called :multi which takes an initial value
;; and zero or more slot names. It produces instance slot definitions
;; which all use that same initial value.
(define-struct-clause :multi (init-val . names)
(mapcar (lop list init-val) names))
;; define a struct with three slots initialized to zero:
(defstruct bar ()
(:multi 0 a b c)) ;; expands to (a 0) (b 0) (c 0)
;; struct clause macro to define a slot along with a
;; get and set method.
(define-struct-clause :getset (slot getter setter : init-val)
^((,slot ,init-val)
(:method ,getter (obj) obj.,slot)
(:method ,setter (obj new) (set obj.,slot new))))
;; Example use:
(defstruct point ()
(:getset x get-x set-x 0)
(:getset y get-y set-y 0))
;; This has exactly the same effect as the following defstruct:
(defstruct point ()
(x 0)
(y 0)
(:method get-x (obj) obj.x)
(:method set-x (ob new) (set obj.x new))
(:method get-y (obj) obj.y)
(:method set-y (ob new) (set obj.y new)))
(:delegate name (param+) delegate-expr [target-name])
The :delegate struct clause macro provides a way to define a method which is implemented entirely by delegation to a different object. The name of the method is name and its parameter list is specified in the same way as in the :method clause. Instead of a method body, the :delegate clause has an expression delegate-expr and an optional target-name which defaults to name. The delegate-expr must be an expression which the delegate method can evaluate to produce a delegate object. The delegate method then passes its arguments to the target method, given by the target-name argument, invoked on the delegate object.
If the delegate method specifies an optional parameter without a default initializing expression, and that optional parameter is not given an argument value, it receives the colon symbol : as its argument. That value is passed on to the corresponding parameter of the delegate target method. Thus, if the target method has an optional parameter in that same parameter position, that colon symbol argument then has the effect of requesting the default value. If the target method has an ordinary parameter in that position, then the colon symbol is received as an ordinary argument value.
If the delegate method specifies an optional parameter with a default initializing expression, and that optional parameter is not given an argument value, then the expression is evaluated to produce a value for that parameter, in the usual manner, and that value is passed as an argument to the corresponding parameter of the delegate target. Thus, delegates are able to specify different optional argument defaulting from their targets.
A delegate may have an optional parameter in a position where the target has a required parameter and vice versa.
The three-element optional parameter expression, specifying a Boolean variable which indicates whether the optional parameter has been given an argument, is not supported by the :delegate clause, and is diagnosed.
If the delegate method has variadic parameters, they are passed on to the target after the fixed parameters.
Structure definitions:
(defstruct worker ()
name
(:method work (me)
`worker @{me.name} works`)
(:method relax (me : (min 15))
`worker @{me.name} relaxes for @min min`))
;; "contractor" class has a sub ("subcontractor") slot
;; which is another contractor of the same type.
;; The subcontractor's own sub slot, however is going
;; to be a worker.
(defstruct contractor ()
sub
(:delegate work (me) me.sub.sub)
(:delegate break (me : min) me.sub.sub relax))
The contractor structure's work and break methods delegate to the sub-subcontractor, which is going to be instantiated as a worker object. Note that the break method delegates to a differently named method relax.
;; The objects are set up as described above.
;; general contractor co has a co.sub subcontractor,
;; and co.sub.sub is a worker:
(defvar co (new contractor
sub (new contractor
sub (new worker name "foo"))))
;; Call work method on general contractor:
;; this invokes co.sub.sub.(work) on the worker.
co.(work) -> "worker foo works"
;; Call break method on general contractor with
;; no argument. This causes co.sub.sub.(relax :)
;; to be invoked, triggering argument defaulting:
co.(break) -> "worker foo relaxes for 15 min"
;; Call break method with argument. This
;; invokes co.sub.sub.(relax 5), specifying a
;; value for the default argument:
co.(break 5) -> "worker foo relaxes for 5 min"
(:mass-delegate self-var delegate-expr
from-type [*] [method]*)
The :mass-delegate struct macro provides a way to define multiple methods which are implemented as delegates to corresponding methods on another object. The implementation of :mass-delegate depends on the :delegate macro.
The self-var argument must be a bindable symbol. In each generated delegate method, this symbol will be the first argument. The purpose of this symbol is to enable the delegate-expr to refer to the delegating object.
The delegate-expr is an expression which is inserted into every method. Its evaluation is expected to produce the delegate object. This expression may reference self-var in order to retrieve or otherwise obtain the delegate from the delegating object.
The from-type argument is a symbol naming an existing structure type. If no such structure type has been defined, an error exception is thrown.
After the from-type argument, either zero or more slot names appear, optionally preceded by the * (asterisk) symbol.
If the * symbol is present, and isn't followed by any other symbols, it indicates that all methods from from-type are to be delegated. If symbols appear after the * then those specify exceptions: methods not to be delegated. No validation is performed on the exception list; it may specify nonexistent method names which have no effect.
If the * symbol is absent, then every method symbol specifies a method to be delegated. It is consequently expected to name a method of the from-type: a static slot which contains a function. If any method isn't a static slot of from-type, or not a static slot which contains a function, an error exception is thrown.
The :mass-delegate struct macro iterates over all of the methods of from-type that selected for delegation, and for each one it generates a :delegate macro clause based on the existing method's parameter list. For instance, the delegate for a method which has two required arguments and one optional will itself have two required arguments and one optional. Delegates are not simply wrapper functions which take any number of arguments and try to pass them to the target.
The generated :delegate clauses are then processed by that struct clause macro.
Note: composition with delegation is a useful alternative when multiple inheritance is not applicable or desired for various reasons. One such reason is that structures that would be used as multiple inheritance bases use the same symbols for certain slots, and the semantics of those slots conflict. Under inheritance, same-named slots coming from different bases become one slot,
Note: a particular from-type being nominated in the :mass-delegate clause doesn't mean that the specific methods of that type shall be called by the generated delegates. The methods that shall be called are those of the calculated delegate object selected by the delegate-expr. The from-type is used as a source of the argument info, and method existence validation. It is up to the application to ensure that the delegation using from-type makes sense with respect to the delegate object that is selected by the delegate-expr: for instance, by ensuring that this object is an instance of from-type or a subtype thereof.
(defstruct foo-api ()
name
(:method begin (me) ^(foo ,me.name begin))
(:method increment (me delta) ^(foo ,me.name increment ,delta))
(:method end (me) ^(foo ,me.name end)))
(defstruct bar-api ()
name
(:method open (me) ^(bar ,me.name open))
(:method read (me buf) ^(bar ,me.name read ,buf))
(:method write (me buf) ^(bar ,me.name write ,buf))
(:method close (me) ^(bar ,me.name close)))
;; facade holds the two API objects by composition:
(defstruct facade ()
(foo (new foo-api name "foo"))
(bar (new bar-api name "bar"))
;; delegate foo-api style calls via me.foo
(:mass-delegate me me.foo foo-api *)
;; delegate bar-api style calls via me.bar
;; exclude the write method.
(:mass-delegate me me.bar bar-api * write))
;; instantiate facade as variable fa
(defvar fa (new facade)) -> fa
;; begin call on facade delegates through foo-api object.
fa.(begin) -> (foo "foo" begin)
fa.(increment) -> ;; error: too few arguments
fa.(increment 3) -> (foo "foo" increment 3)
fa.(open) -> (bar "bar" open)
fa.(write 4) -> ;; error: fa has no such method
(macroexpand-struct-clause clause [form])
If clause is a compound expression whose operator symbol was defined by define-struct-clause then macroexpand-struct-clause expands the clause and returns the expansion, which is a list of zero or more clauses. Otherwise, the function returns a one-element list containing the clause argument, as if by the (list clause) expression.
The form parameter, if present, is used for reporting errors. Note: clauses are usually expanded during the processing of a defstruct macro; in that situation, the entire unexpanded defstruct form serves the role,
;; try to expand :delegate, using incorrect syntax.
(macroexpand-struct-clause '(:delegate x (a b)))
--> error "** source location n/a: nil: too few elements ..."
;; same, but with error reporting form.
(macroexpand-struct-clause '(:delegate x (a b)) '(abc xyz))
--> error: "** expr-1:1: abc: too few elements ..."
;; correct :delegate syntax
(macroexpand-struct-clause '(:delegate x (a b) a.foo))
--> ((:method x (a b) (qref a.foo (x b))))
;; not a defstruct macro clause
(macroexpand-struct-clause '(1 2 3))
-> ((1 2 3))
The *struct-clause-expander* special variable holds the hash table of associations between keyword symbols and struct clause expander functions, defined by define-struct-clause.
If the expression [*struct-clause-expander* :sym] yields a function, then symbol :sym has a binding as a struct clause macro. If that expression yields nil, then there is no such binding.
The macro expanders in *struct-clause-expander* are two-parameter functions. The first parameter accepts the clause to be expanded. The second parameter accepts the defstruct form in which that clause is found; this is useful for error reporting.
An expander function returns a list of clauses, which may be any, possibly empty mixture of primary clauses accepted by defstruct and clause macros.
(define-struct-prelude name struct-name-or-list clause*)
The define-struct-prelude macro defines a prelude. A prelude is a named entity which implicitly provides clauses to defstruct macro invocations. Preludes are processed during the macroexpansion of defstruct; prelude definitions have no effect on previously compiled defstruct forms loaded from a file.
A prelude has a name which must be a bindable symbol. The purpose of this name is that if multiple define-struct-prelude forms are evaluated which specify the same name, they replace each others' definition. Only the most recent prelude of a given name is retained; the previous definitions are overwritten.
The struct-name-or-list argument is either a symbol or a list of symbols, which are valid for use as structure names. The prelude being defined shall be applicable to each of the structures whose names are given by this argument.
The zero or more clause arguments give the clauses which comprise the prelude. In the future, when a defstruct form is macroexpanded which targets any of the structures given by the struct-name-or-list argument, the specified clauses will be inserted into that definition, as if they appeared in the defstruct form literally.
Multiple preludes may be defined with different names, which each target the same structure. When the structure is defined, or redefined, it will receive all those preludes, in the order in which they were defined.
;; define init-fini-log prelude which targets fox and bear structs
(define-struct-prelude init-fini-log (fox bear)
(:init (me) (put-line `@me created`))
(:fini (me) (put-line `@me finalized`)))
;; The behavior is as if the following defstruct forms included
;; the above :init and :fini clauses
(defstruct fox ())
(defstruct bear ())
(with-object ((f (new fox))
(b (new bear)))
(put-line "inside with-object"))
Output:
#S(fox) created
#S(bear) created
inside with-object
#S(bear) finalized
#S(fox) finalized
Special structure functions are user-defined methods or structure functions which are specially recognized by certain functions in TXR Lisp. They endow structure objects with the ability to participate in certain usage scenarios, or to participate in a customized way.
Special functions are required to bound to static slots, which is the case if the defmeth macro is used, or when methods or functions are defined using syntax inside a defstruct form. If a special function or method is defined as an instance slot, then the behavior of library functions which depend on this method is unspecified.
Special functions introduced below by the word "Method" receive an object instance as an argument. Their syntax is indicated using the same notation which may be used to invoke them, such as:
object.(function-name arg ...)
However, those introduced as "Function" do not operate on an instance. Their syntax is likewise indicated using the notation that may be used to invoke them:
'['
object.function-name arg ...']'
If such a invocation is actually used, the object instance only serves for identifying the struct type whose static slot function-name provides the function; object doesn't participate in the call. An object is not strictly required since the function can be called using
[(static-slot
type 'function-name) arg ...]
which looks up the function in the struct type directly.
object.(copy)
The special method copy is expected to produce a copy of the object. The copy function will use this method if it is available, otherwise fall back on copy-struct. The method is responsible for all semantics of the copy operation; whatever object the method returns is taken to be a copy of object.
It is a recommended practice that the returned object be of the same type as object. It is also a recommended practice that the returned object be newly created, distinct from any object which existed prior to the method being called. The objects held in that object's slots need not be new.
object.(equal)
Normally, two struct values are not considered the same under the equal function unless they are the same object.
However, if the equal method is defined for a structure type, then instances of that structure type support equality substitution.
The equal method must not require any arguments other than object. Moreover, the method must never return nil.
When a struct which supports equality substitution is compared using equal, less or greater, its equal method is invoked, and the return value is used in place of that structure for the purposes of the comparison.
The same applies when a struct is hashed using the hash-equal function, or implicitly by an :equal-hash hash table.
Note: if an equal method is defined or redefined with different semantics for a struct type whose instances have already been inserted as keys in an :equal-based hash table, the behavior of subsequent insertion and lookup operations on that hash table becomes unspecified.
object.(print stream pretty-p)
If a method named by the symbol print is defined for a structure type, then it is used for printing instances of that type.
The stream argument specifies the output stream to which the printed representation is to be written.
The pretty-p argument is a Boolean flag indicating whether pretty-printing is requested. Its value may simply be passed to recursive calls to print, or used to select between ~s or ~a formatting if format is used.
The value returned by the print method is significant. If the special keyword symbol : (colon) is returned, then the system will print the object in the default way, as if no print method existed: it is understood that the method declined the responsibility for printing the object.
If any other value is returned, then it is understood that the method print method accepted the responsibility for printing the object, and the system consequently will generate into stream any output output pertaining to object's representation.
object.(slot slot-name)
object.(slot-set slot-name new-value)
Defining these methods allows a struct type to handle the situation when a nonexistent slot is accessed.
The slot method, if it exists, is invoked if a slot named slot-name is accessed by the slot function, or equivalent syntax, and that slot does not exist. The value returned by the method is taken to be the nonexistent slot's value.
When a value is stored in a slot named slot-name by the slotset function, or equivalent syntax, and the slot does not exist, then the slotset method is invoked, if it exists. It is recommended that the slotset function return new-value, since the value returned propagates out of the slotset function, which in all other cases returns new-value, which is important to the implementation of syntactic places that designate slots.
object.[static-slot type slot-name)
object.[static-slot-set type slot-name new-value)
The static-slot and static-slot-set functions are analogous to the slot and slotset, methods. These functions, if they exist, are only invoked when a static slot lookup fails. Static slot lookups occur through the static-slot and static-slot-set functions, which can be used directly and are used in certain situations. For instance when (meth ...) syntax is looked up with symbol-function, static slot lookup is used. It is recommended that for simulating the existence of structure functions and methods, these methods be used.
The type, argument is an object of type struct-type giving the structure type on which the static slot lookup is taking place.
It is recommended that the static-slot-set function return new-value.
object.(lambda arg*)
If a structure type provides a method called lambda then it can be used as a function.
This method can be called by name, using the syntax given in the above syntactic description.
However, the intended use is that it allows the structure instance itself to be used as a function. When a structure is applied to arguments as if it were a function, this is erroneous, unless that object has a lambda method. In that case, the arguments are passed to the lambda method. The leftmost argument of the method is the structure instance itself.
That is to say, the following equivalences apply, except that s is evaluated only once:
(call s args ...) <--> s.(lambda args ...)
[s args ...] <--> [s.lambda s args ...]
(mapcar s list) <--> (mapcar (meth s lambda) list)
Note: a form such as [s args ...] where s is a structure can be treated as a place if the method lambda-set is also implemented.
object.(lambda-set arg* new-value)
The lambda-set method, in conjunction with a lambda method, allows structures to be used as place accessors. If structure s supports a lambda-set with four arguments, then the following use of the dwim operator is possible:
(set [s a b c d] v)
(set (dwim s a b c d) v) ;; precisely equivalently
This has an effect which can be described by the following code:
(progn
s s.(lambda-set a b c d v)
v)
except that s and v are evaluated only once, and a through d are evaluated using the Lisp-1 semantics due the dwim operator.
If a place-mutating operator is used on this form which requires the prior value, such as the inc macro, then the structure must support the lambda function also.
If lambda takes n arguments, then lambda-set should take n+1 arguments. The first n arguments of these two methods are congruent; the extra rightmost argument of lambda-set is the new value to be stored into the place denoted by the prior arguments.
The return value of lambda-set is ignored.
Note: the lambda-set method is also used by the rplaca function, if no rplaca method exists.
The following defines a structure with a single instance slot hash which holds a hash table, as well as lambda and lambda-set methods:
(defstruct hash-wrapper nil
(hash (hash))
(:method lambda (self key)
[self.hash key])
(:method lambda-set (self key new-val)
(set [self.hash key] new-val) self))
An instance of this structure can now be used as follows:
(let ((s (new hash-wrapper)))
(set [s "apple"] 3
[s "orange] 4)
[s "apple"]) -> 3
object.(length)
If a structure has length method, then it can be used as an argument to the length function.
Structures which implement the methods lambda, lambda-set and length can be treated as abstract vector-like sequences, because such structures support the ref, refset and length functions.
For instance, the nreverse function will operate on such objects.
Note: a structure which supports the car method also supports the length function, in a different way. Such a structure is treated by length as a list-like sequence, and its length is measured by walking the sequence with cdr operations. If a structure supports both length and car, preference is given to length, which is likely to be much more efficient.
object.(length-< len)
If a structure has length-< method, then it can be used as the left argument to the length-< function. The len argument receives the right argument.
If an object doesn't implement the length-< method, but does implement the length it can also be used as an argument to the length-< function. In that situation, the length-< function will call the length method instead, and then compare the returned value against the len parameter.
object.(car)
object.(cdr)
object.(nullify)
Structures may be treated as sequences if they define methods named by the symbols car, cdr, and nullify.
If a structure supports these methods, then these methods are used by the functions car, cdr, nullify, empty and various other sequence manipulating functions derived from them, when those functions are applied to that object.
An object which implements these three methods can be considered to represent a list-like abstract sequence.
The object's car method should return the first value in that abstract sequence, or else nil if that sequence is empty.
The object's cdr method should return an object denoting the remainder of the sequence, or else nil if the sequence is empty or contains only one value. This returned object can be of any type: it may be of the same structure type as that object, a different structure type, a list, or whatever else. If a non-sequence object is returned.
The nullify method should return nil if the object is considered to denote an empty sequence. Otherwise it should either return that object itself, or else return the sequence which that object represents.
object.(rplaca new-car-value)
object.(rplacd new-cdr-value)
If a structure type defines the methods rplaca and rplacd then, respectively, the rplaca and rplacd functions will use these methods if they are applied to instances of that type.
That is to say, when the function call (rplaca o v) is evaluated, and o is a structure type, the function inquires whether o supports a rplaca method. If so, then, effectively, o. (rplaca v) is invoked. The return value of this method call is ignored; rplaca returns o. The analogous requirements apply to rplacd in relation to the rplacd method.
Note: if the rplaca method doesn't exist, the rplaca function falls back on trying to store new-car-value by means of the structure type's lambda-set method, using an index of zero. That is to say, if the type has no rplaca method, but does have a lambda-set method, then o. (lambda-set 0 v) is invoked.
'['object.from-list list']'
If a from-list structure function is defined for a structure type, it is called in certain situations with an argument which is a list object. The function's purpose is to construct a new instance of the structure type, derived from that list.
The purpose of this function is to allow sequence processing operations such as mapcar and remove to operate on a structure object as if it were a sequence, and return a transformed sequence of the same type. This is analogous to the way such functions can operate on a vector or string, and return a vector or string.
If a structure object behaves as a sequence thanks to providing car, cdr and nullify methods, but does not have a from-list function, then those sequence-processing operations which return a sequence will always return a plain list of items.
'['object.derived supertype subtype']'
If a structure type supports a function called derived, this function is called whenever a new type is defined which names that type as its supertype.
The function is called with two arguments which are both struct types. The supertype argument gives the type that is being inherited from. The subtype gives the new type that is inheriting from supertype.
When a new structure type is defined, its list of immediate supertypes is considered. For each of those supertypes which defines the derived function, the function is invoked.
The function is not retroactively invoked. If it is defined for a structure type from which subtypes have already been derived, it is not invoked for those existing subtypes.
If derived directly inherits supertype more than once, it is not specified whether this function is called once, or multiple times.
Note: the supertype parameter exists because the derived function is itself inherited. If the same version of this function is shared by multiple structure types due to inheritance, this argument informs the function which of those types it is being invoked for.
object.(iter-begin)
object.(iter-reset iter)
If an object supports the iter-begin method, it is considered iterable; the iterable function will return t if invoked on this object.
The responsibility of the iter-begin method is to return an iterator object: an object which supports certain special methods related to iteration, according to one of two protocols, described below.
The iter-reset method is optional. It is similar to iter-begin but takes an additional iter argument, an iterator object that was previously returned by the iter-begin method of the same object.
If iter-reset determines that iter can be reused for a new iteration, then it can suitably mutate the state of iter and return it. Otherwise, it behaves like iter-begin and returns a new iterator.
There are two protocols for iteration: the fast protocol, and the canonical protocol. Both protocols require the iterator object returned by the iter-begin method to provide the methods iter-item and iter-step. If the iterator also provides the iter-more method, then the protocol which applies is the canonical protocol. If that method is absent, then the fast protocol is followed.
Under the fast protocol, the iter-more method does not exist and is not involved. The iterable object's iter-begin method must return nil if the abstract sequence is empty. If an iterator is returned, it is assumed that an object can be retrieved from the iterator by invoking its iter-item method. The iterator's iter-next method should return nil if there are no more objects in the abstract sequence, or else it should return an iterator that obeys the fast protocol (possibly itself).
Under the canonical protocol, the iterator implements the iter-more function. The iterable object's iter-begin always returns an iterator object. The iterator object's iter-more method is always invoked to determine whether another item is available from the sequence. The iterator object's iter-step method is expected to return an iterator object which conforms to the canonical protocol.
object.(iter-item)
The iter-item method is invoked on an iterator object to retrieve the next item in the sequence.
Under the fast protocol, it is assumed that if object was returned by an iterable object's iter-begin method, or by an iterator's iter-step method, that an item is available. This method will be unconditionally invoked.
Under the canonical protocol for iteration, the iter-more method will be invoked on object first. If that method yields true, then iter-item is expected to yield the next available item in the sequence.
Note: calls to the iter-item function, with object as its argument, invoke the iter-item method. It is possible for an application to call iter-item through this function or directly as a method call without first calling iter-more. No iteration mechanism in the TXR Lisp standard library behaves this way. If the iterator object has no more items available and iter-more is invoked anyway, no requirements apply to its behavior or return value.
object.(iter-step)
The iter-step method is invoked on an iterator object to produce an iterator object for the remainder of the sequence, excluding the current item.
Under the fast iteration protocol, this method returns nil if there are no more items in the sequence.
Under the canonical iteration protocol, this method always returns an iterator object. If no items remain in the sequence, then that iterator object's iter-more method returns nil. Furthermore, under this protocol, iter-step is not called if iter-more returns nil.
Note: calls to the iter-step function, with object as its argument, invoke the iter-step method. It is possible for an application to call iter-step through this function or directly as a method call without first calling iter-more. No iteration mechanism in the TXR Lisp standard library behaves this way. If the iterator object has no more items available and iter-step is invoked anyway, no requirements apply to its behavior or return value.
object.(iter-more)
If an iterator object returned by iter-begin supports the iter-more method, then the canonical iteration protocol applies to that iteration session. All subsequent iterators that are involved in the iteration are assumed to conform to the protocol and should implement the iter-more method also. The behavior is unspecified otherwise.
The iter-more method is used to interrogate an iterator whether more unvisited items remain in the sequence. This method does not advance the iteration, and does not change the state of the iterator. It is idempotent: if it is called multiple times without any intervening call to any other method, it yields the same value.
If an iterator does not implement the iter-more method, then if the iter-more function is applied to that iterator, it unconditionally returns t.
Functions in this category uniformly manipulate abstract sequences. Lists, strings and vectors are sequences.
Structure objects can behave like sequences, either list-like or vector-like sequences, if they have certain methods: see the previous section Special Structure Functions.
Moreover, hash tables behave like sequences of key-value entries represented by cons pairs. Not all sequence-processing functions accept hash-table sequences.
Additionally, some sequence-processing functions work not only with sequences but with all iterable objects: objects that can be used as arguments to the iter-begin function. Such arguments are called iterable rather than sequence, possibly abbreviated to iter with or without a numeric suffix. Hash tables are always supported if they appear as iterable arguments.
(seqp object)
The function seqp returns t if object is a sequence, otherwise nil.
Lists, vectors and strings are sequences. The object nil denotes the empty list and so is a sequence.
Objects of type buf and carray are sequences, as are hash tables.
Structures which implement the length or car methods are considered sequences.
No other objects are sequences. However, future revisions of the language may specify additional objects that are sequences.
(iterable object)
The iterable function returns t if object is iterable, otherwise nil.
If object is a sequence according to the seqp function, then it is iterable.
If object is a structure which supports the iter-begin method, then it is iterable.
Additional objects that are not sequences are also iterable: numeric or character ranges, and numbers. Future revisions of the language may specify additional iterable objects.
(make-like seq object)
(seq-like object arg*)
The make-like function's seq argument must be a sequence. If object is a sequence type, then list is converted to the same type of sequence, if possible, and returned. Otherwise the original seq is returned.
Conversion is supported to string and vector type, plus additional types as follows.
Conversion to a structure type is possible for structures. If object is an object of a structure type which has a static function from-list, then make-like calls that function, passing to it, and the resulting value is returned. seq and returns whatever value that function returns.
If object is a carray, then list is passed to the carray-list function, and the resulting value is returned. The second argument in the carray-list call is the element type taken from object. The third argument is nil, indicating that the resulting carray is not to be null terminated.
The object may be an iterator returned by iter-begin. In this situation, if that object makes the original sequence available, then make-like takes that sequence in place of object,
The seq-like function creates, if possible, a sequence of the same kind as object populated by the remaining arg values. If some of the arg values are not suitable elements for a sequence of that type, then a list of those values is returned.
The result of seq-like is consistent with what the make-like function would return if given a sequence of the arg values as the seq argument. That is to say, the following equivalence holds:
(make-like (list a0 a1 ...) o) <-> (seq-like o a0 a1 ...)
Note: the make-like function is a helper which supports the development of unoptimized versions of a generic function that accepts any type of sequence as input, and produces a sequence of the same type as output. The implementation of such a function can internally accumulate a list, and then convert the resulting list to the same type as an input value by using make-like.
(list-seq iterable)
(vec-seq iterable)
(str-seq iterable)
The list-seq, vec-seq and str-seq functions convert an iterable object of any type into a list, vector or string, respectively.
The list returned by list-seq is lazy.
The list-seq and vec-seq iterate the items of iterable and accumulate these items into a new list or vector.
The str-seq similarly iterates the items of iterable, requiring them to be a mixture of characters and strings.
(length iterable)
(len iterable)
The length function returns the number of items contained in iterable.
The len function is a synonym of length.
An attempt to calculate the length of infinite lazy lists will not terminate. Iterable objects representing infinite ranges, such as integers and characters are invalid arguments.
(length-< iterable len)
The length-< function efficiently determines whether (length iterable) is less than the integer value len. In cases when iterable would have to be fully traversed in order to measure its length, the length-< function avoids this traversal, by making use of the functions length-str-< or length-list-< as appropriate.
Note: this function is useful when a decision must be made between two algorithms, depending on whether the length is less than a certain small constant. It is also safe on lazy, infinite sequences and circular lists, for which length will fail to terminate.
(empty iterable)
If iterable is a suitable argument for the length function, then the empty Returns t if (length iterable) is zero, otherwise nil.
The empty function also supports certain objects not suitable as arguments for length.
An infinite lazy list is not empty, and so empty returns nil for such an object.
The function also returns nil for iterable objects representing nonempty spaces, even if those spaces are infinite. For instance (empty 0) yields nil because the set of integers beginning with 0 isn't empty.
(nullify iterable)
The nullify function returns nil if iterable denotes an empty sequence. Otherwise, if iterable is not an empty sequence, or isn't a sequence, then iterable itself is returned.
If iterable is a structure object which supports the nullify method, then that method is called. If it returns nil then nil is returned. If the nullify method returns a substitute object other than the iterable object itself, then nullify is invoked on that returned substitute object.
Note: the nullify function is a helper to support unoptimized generic traversal of sequences. Thanks to the generalized behavior of cdr, non-list sequences can be traversed using cdr, similarly to proper lists, by checking for cdr returning the terminating value nil. However, empty non-list sequences are handled incorrectly because since they are not the nil object, they look nonempty under this paradigm of traversal. The nullify function provides a correction: if the input sequence is filtered through nullify then the subsequent list-like iteration works correctly.
Examples:
;; Incorrect for empty strings:
(defun print-chars (string)
(while string
(prinl (pop string))))
;; Corrected with nullify:
(defun print-chars (string)
(let ((s (nullify string)))
(while s
(prinl (pop s)))))
Note: optimized generic iteration is available in the form of iteration based on iter-begin rather than car/cdr and nullify.
Examples:
;; Efficient with iterators,
;; at the cost of verbosity:
(defun print-chars (string)
(let ((i (iter-begin string)))
(while (iter-more i)
(prinl (iter-item s))
(set s (iter-step s)))))
;; Using mapping function built on iterators:
(defun print-chars (string)
[mapdo prinl string])
(sub sequence [from [to]])
(set (sub sequence [from [to]]) new-val)
The sub function extracts a slice from input sequence sequence. The slice is a sequence of the same type as sequence.
If the from argument is omitted, it defaults to 0. If the to parameter is omitted, it defaults to t. Thus (sub a) means (sub a 0 t).
The following semantic equivalence exists between a call to the sub function and the DWIM-bracket syntax, except that sub is an ordinary function call form, which doesn't apply the Lisp-1 evaluation semantics to its arguments:
;; from is not a list
(sub seq from to) <--> [seq from..to]
The description of the dwim operator—in particular, the section on Range Indexing—explains the semantics of the range specification.
The output sequence may share structure with the input sequence.
If sequence is a carray object, then the function behaves like carray-sub.
If sequence is a buf object, then the function behaves like sub-buf.
If sequence is a tree object, then the function behaves like sub-tree. Note: because sub-tree is not an accessor, assigning to the sub syntax in this case will produce an error.
The sequence argument may also be any other object type that is suitable as input to the iter-begin function. In this situation, assigning to sub syntax produces an error. Furthermore, in cases where the from and to arguments imply that a suffix of sequence is required, an lazy list of the suffix of the iterated sequence will be returned. In other cases, a regular list of the elements selected by sub is returned.
If sequence is a structure, it must support the lambda method. The sub operation is transformed into a call to the lambda method according to the following equivalence:
(sub o from to) <--> o.(lambda (rcons from to))
(sub o : to) <--> o.(lambda (rcons : to))
(sub o from) <--> o.(lambda (rcons from :))
(sub o) <--> o.(lambda (rcons : :))
That is to say, the from and to arguments are converted to range object. If either argument is missing, the : (colon) keyword symbol is used for the corresponding element of the range.
When a sub form is used as a syntactic place, that place denotes a slice of seq. The seq argument must be itself be syntactic place, because receives a new value, which may be different from its original value in cases when seq is a list.
Overwriting that slice is equivalent to using the replace function. The following equivalences give the semantics, except that x, a, b and v are evaluated only once, in left-to-right order:
(set (sub x a b) v) <--> (progn (set x (replace x v a b))
v)
(del (sub x a b)) <--> (prog1 (sub x a b)
(set x (replace x nil a b)))
Note that the value of x is overwritten with the value returned by replace. If x is a vector or string, then the return value of replace is just x: the identity of the object doesn't change under mutation. However, if x is a list, its identity changes when items are added to or removed from the front of the list, and in those cases replace will return a value different from its first argument. Similarly, if x is an object with a lambda-set method, that method's return value becomes the return value of replace and must be taken into account.
(replace sequence replacement-sequence [from [to]])
(replace sequence replacement-sequence index-seq)
The replace function modifies sequence in the ways described below.
The operation is destructive: it may work "in place" by modifying the original sequence. The caller should retain the return value and stop relying on the original input sequence.
The return value of replace is the modified version of sequence. This may be the same object as sequence or it may be a newly allocated object.
Note that the form:
(set seq (replace seq new fr to))
has the same effect on the variable seq as the form:
(set [seq fr..to] new)
except that the former set form returns the entire modified sequence, whereas the latter returns the value of the new argument.
The replace function has two invocation styles, distinguished by the type of the third argument. If the third argument is a sequence, then it is deemed to be the index-seq parameter of the second form. Otherwise, if the third argument is missing, or is not a sequence, then it is deemed to be the from argument of the first form.
The first form of the replace function replaces a contiguous subsequence of the sequence with replacement-sequence. The replaced subsequence may be empty, in which case an insertion is performed. If replacement-sequence is empty (for example, the empty list nil), then a deletion is performed.
If the from and to arguments are omitted, their values default to 0 and t respectively.
The description of the dwim operator—in particular, the section on Range Indexing—explains the semantics of the range specification.
The second form of the replace function replaces a subsequence of elements from sequence given by index-seq, with their counterparts from replacement-sequence. If replacement-sequence has at least as many elements as are indicated in index-seq, then the indicated elements of sequence are overwritten with successive elements from replacement-sequence. If replacement-sequence contains fewer elements than index-seq, then the excess elements indicated in index-seq which have no counterparts in the replacement-sequence are deleted. Whenever a negative value occurs in index-seq the original length of sequence (before any deletions) is added to that value. Furthermore, similar restrictions apply on index-seq as under the select function. Namely, the replacement stops when an index value in index-seq is encountered which is out of range for sequence. furthermore, if sequence is a list, or if any deletions take place, then index-seq must be monotonically increasing, after consideration of the displacement of negative values, or else the behavior is unspecified.
If replacement-sequence shares storage with the target range of sequence, or, in the case when that range is resized by the replace operation, shares storage with any portion of sequence above that range, then the effect of replace on either object is unspecified.
If sequence is a carray object, then replace behaves like carray-replace.
If sequence is a buf object, then replace behaves like buf-replace.
If sequence is a structure, then the structure must support the lambda-set method. The replace operation is translated into a call of the lambda-set method according to the following equivalences:
(replace o items from to)
<--> o.(lambda-set (rcons from to) items)
(replace o items index-seq)
<--> o.(lambda-set index-seq items)
Thus, the from and to arguments are converted to single range object, whereas an index-seq is passed as-is. It is an error if the from argument is a sequence, indicating an index-seq, and a to argument is also given; the situation is diagnosed. If either from or to are omitted, the range object contains the : (colon) keyword symbol in the corresponding place:
(replace o items from)
<--> o.(lambda-set (rcons from :) items)
(replace o items : to)
<--> o.(lambda-set (rcons : to) items)
(replace o items)
<--> o.(lambda-set (rcons : :) items)
It is the responsibility of the object's lambda-set method to implement semantics consistent with the description of replace.
(take count sequence)
The take function returns sequence with all except the first count items removed.
If sequence is a list, then take returns a lazy list which produces the first count items of sequence.
For other kinds of sequences, including lazy strings, take works eagerly.
If count exceeds the length of sequence then a sequence is returned which has all the items. This object may be sequence itself, or a copy.
If count is negative, it is treated as zero.
(take-while predfun sequence [keyfun])
(take-until predfun sequence [keyfun])
The take-while and take-until functions return a prefix of sequence whose items satisfy certain conditions.
The take-while function returns the longest prefix of sequence whose elements, accessed through keyfun satisfy the function predfun.
The keyfun argument defaults to the identity function: the elements of sequence are examined themselves.
The take-until function returns the longest prefix of sequence which consists of elements, accessed through keyfun, that do not satisfy predfun followed by an element which does satisfy predfun. If sequence has no such prefix, then an empty sequence is returned of the same kind as sequence.
If sequence is a list, then these functions return a lazy list.
(drop count sequence)
The drop function returns sequence with the first count items removed.
If count is negative, it is treated as zero.
If count is zero, then sequence is returned.
If count exceeds the length of sequence then an empty sequence is returned of the same kind as sequence.
(drop-while predfun sequence [keyfun])
(drop-until predfun sequence [keyfun])
The drop-while and drop-until functions return sequence with a prefix of that sequence removed, according to conditions involving predfun and keyfun.
The drop-while function removes the longest prefix of sequence whose elements, accessed through keyfun satisfy the function predfun, and returns the remaining sequence.
The keyfun argument defaults to the identity function: the elements of sequence are examined themselves.
The drop-until function removes the longest prefix of sequence which consists of elements, accessed through keyfun, that do not satisfy predfun followed by an element which does satisfy predfun. A sequence of the remaining elements is returned.
If sequence has no such prefix, then a sequence same as sequence is returned, which may be sequence itself or a copy.
(last sequence [num])
(set (last sequence [num]) new-value)
The last function returns a subsequence of sequence consisting of the last num of its elements, where num defaults to 1.
If num is zero or negative, then an empty sequence is returned. If num is positive, and greater than or equal to the length of sequence, then sequence sequence is returned.
If a last form is used as a place, then sequence must be a place. The following equivalence gives the semantics of assignment to a last:
(set (last x n) v) <--> (set (sub x (- (max n 0)) t) v)
A last place is deletable. The semantics of deletion may be understood in terms of the following equivalence:
(del (last x n)) <--> (del (sub x (- (max n 0)) t))
(butlast sequence [num])
(set (butlast sequence [num]) new-value)
The butlast function returns the prefix of sequence consisting of a copy of it, with the last num items removed.
The parameter num defaults to 1 if an argument is omitted.
If sequence is empty, an empty sequence is returned.
If num is zero or negative, then sequence is returned.
If num is positive, and meets or exceeds the length of sequence, then an empty sequence is returned.
If a butlast form is used as a place, then sequence must itself be a place. The following equivalence gives the semantics of assignment to a last:
(set (butlast x n) v) <--> (set (sub x 0 (- (max n 0))) v)
A butlast place is deletable. The semantics of deletion may be understood in terms of the following equivalence:
(del (last x n)) <--> (del (sub x 0 (- (max n 0))))
Note: the TXR Lisp take function also computes the prefix of a list; however, it counts items from the beginning, and provides lazy semantics which allow it to work with infinite lists.
See also: the butlastn accessor, which operates on lists. That function has useful semantics for improper lists and treats an atom as the terminator of a zero-length improper list.
Dialect Note: a destructive function similar to Common Lisp's nbutlast isn't provided. Assignment to a butlast form is destructive; Common Lisp doesn't support butlast as a place.
(ldiff sequence tail-sequence)
The ldiff function is a somewhat generalized version of the same-named classic Lisp function found in traditional Lisp dialects.
The ldiff function supports the original ldiff semantics when both inputs are lists. It determines whether the tail-sequence list is a structural suffix of sequence, which is to say: is tail-sequence one of the cons cells which comprise sequence? If so, then a list is returned consisting of all the items of sequence before tail-sequence: a copy of sequence with the tail-sequence part removed, and replaced by the nil terminator. If tail-sequence is nil or the lists are unrelated, then sequence is returned.
The TXR Lisp ldiff function supports the following additional semantics.
;;; unspecified: the compiler could make
;;; '(2 3) a suffix of '(1 2 3),
;;; or they could be separate objects.
(ldiff '(1 2 3) '(2 3)) -> either (1) or (1 2 3)
;; b is the (1 2) suffix of a, so the ldiff is (1)
(let* ((a '(1 2 3)) (b (cdr a)))
(ldiff a b))
-> (1)
;; Rule 5: strings and vector
(ldiff "abc" "bc") -> "a"
(ldiff "abc" nil) -> "abc"
(ldiff #(1 2 3) #(3)) -> #(1 2)
;; Rule 5: mixed vector kinds
(ldiff "abc" #(#\b #\c)) -> "abc"
;; Rule 6:
(ldiff #(1 2 3) '(3)) -> #(1 2 3)
;; Rule 4:
(ldiff '(1 2 3) #(3)) -> '(1 2 3)
(ldiff '(1 2 3 . #(3)) #(3)) -> '(1 2 3)
(ldiff '(1 2 3 . 4) #(3)) -> '(1 2 3 . 4)
;; Rule 6
(ldiff 1 2) -> 1
(ldiff 1 1) -> nil
(search haystack needle [testfun [keyfun]])
The search function determines whether the sequence needle occurs as substring within haystack, under the given comparison function testfun and key function keyfun. If this is the case, then the zero-based position of the leftmost occurrence of key within haystack is returned. Otherwise nil is returned to indicate that key does not occur within haystack. If key is empty, then zero is always returned.
The arguments haystack and needle are sequences. They may not be hash tables.
If needle is not empty, then it occurs at some position N within haystack if the first element of needle matches the element at position N of haystack, the second element of needle matches the element at position N+1 of haystack and so forth, for all elements of needle. A match between elements is determined by passing each element through keyfun, and then comparing the resulting values using testfun.
If testfun is supplied, it must be a function which can be called with two arguments. If it is not supplied, it defaults to eql.
If keyfun is supplied, it must be a function which can be called with one argument. If it is not supplied, it defaults to identity.
;; fails because 3.0 doesn't match 3
;; under the default eql function
[search #(1.0 3.0 4.0 7.0) '(3 4)] -> nil
;; occurrence found at position 1:
;; (3.0 4.0) matches (3 4) under =
[search #(1.0 3.0 4.0 7.0) '(3 4) =] -> 1
;; "even odd odd odd even" pattern
;; matches at position 2
[search #(1 1 2 3 5 7 8) '(2 1 1 1 2) : evenp] -> 2
;; Case insensitive string search
[search "abcd" "CD" : chr-toupper] -> 2
;; Case insensitive string search
;; using vector of characters as key
[search "abcd" #(#\C #\D) : chr-toupper] -> 2
(contains needle haystack [testfun [keyfun]])
The syntax of the contains function differs from that of search: that the needle and haystack arguments are reversed. The semantics is identical.
(rsearch haystack needle [testfun [keyfun])
The rsearch function is like search except for two differences.
Firstly, if needle matches haystack in multiple places, rsearch returns the rightmost matching position rather than the leftmost.
Secondly, if needle is an empty sequence, then rsearch returns the length of haystack, thereby effectively declaring that the rightmost match for an empty needle key occurs at the imaginary position past the element of haystack.
(search-all haystack needle [testfun [keyfun])
The search-all function is closely related to the search and rsearch functions. Whereas those two functions return the leftmost or rightmost position, respectively, of needle within haystack, the search-all function returns a list of all the positions where needle occurs. The positions of overlapping matches are included in the list.
If needle is not found in haystack, search-all returns the empty list nil.
If needle is empty, then search-all returns a list of all positions in haystack including the one position past the last element. In this situation, if haystack is empty, the list (0) is returned. If haystack contains one item, then the list (0 1) is returned and so forth.
In all situations in which search-all returns a non-empty list, the first element of that list is what search would return for the same arguments, and the last element is what rsearch would return.
(ref sequence index)
(set (ref sequence index) new-value)
The ref accessor performs array-like indexing into sequences, as well as hash tables and objects of type buf, carray, tree as well as structure objects which define a lambda method.
If the sequence parameter is a hash, then these functions perform has retrieval and storage; in that case index isn't restricted to an integer value.
If sequence is a structure, it supports ref directly if it has a lambda method. The index argument is passed to that method, and the resulting value is returned. If a structure lacks a lambda method, but has a car method, then ref treats it as a list, traversing the structure using car/cdr operations. In the absence of support for these operations, the function fails with an error exception.
If sequence is a sequence then index argument must be an integer. The first element of the sequence is indexed by zero. Negative values are permitted, denoting backward indexing from the end of the sequence, such that the last element is indexed by -1, the second last by -2 and so on. See also the Range Indexing section under the description of the dwim operator.
If sequence is a list, then out-of-range indices, whether positive or negative, are treated leniently by ref: such accesses produce the value nil, rather than an error. For other sequence types, such accesses are erroneous. For hashes, accesses to nonexistent elements are treated leniently, and produce nil.
If sequence is a search tree, then ref behaves like tree-lookup.
If sequence is a range object, then ref behaves like rangeref.
A ref expression may be used as a place. Storing a value into a ref place is performed using the refset function.
When the del operator is used to delete an index value from a ref place, the sequence itself must be a place. The deletion calculates a new sequence with the item at index deleted; that new sequence is stored back into the sequence place. Deletion does not use refset but rather the replace function.
(refset sequence index new-value)
The refset function performs indexing into sequence in a manner identical to ref with the purpose of overwriting the indexed element with new-value. It is a companion function to ref which is used in the implementation of the ref place.
The return value of ref-set is new-value.
If sequence is a structure, it supports refset directly if it has a lambda-set method. This gets called with index and new-value as arguments. Then new-value is returned. If a structure lacks a lambda-set method, then refset treats it as a list, traversing the structure using car/cdr operations, and storing new-value using rplaca. In the absence of support for these operations, the function fails with an error exception.
The refset function is not supported by search trees.
The refset function is strict for out-of-range indices over all sequences, including lists. In the case of hashes, a refset of a nonexistent key creates the key.
(mref sequence index*)
(set (mref sequence index+) new-value)
The mref accessor provides a mechanism for invoking a curried function. Its name reflects its usefulness for multi-dimensional indexing into nested sequences.
The associated mref place which makes the operator an accessor provides in-place replacement of values in multi-dimensional sequences. There are some restrictions on the index arguments when mref is used as a place.
The sequence argument is not necessarily a sequence, but may be object that can be called as a function with one argument. Except that call isn't a place, the expression (mref x i) is equivalent to (call x i): invoke the function/object x with argument i.
When multiple index arguments are present, the return value of each previous application is expected to be another callable object, to which the next index argument is applied. Thus (mref x i j k) is equivalent to (call (call (call x i) j) k). This is also equivalent to [[[x i] j] k], provided that under the Lisp-1-style name resolution semantics of the DWIM brackets, the symbols x, i, j and k all resolve to bindings in the variable namespace.
The expression (mref x) is not equivalent to (call x); rather, it is equivalent to x: there are no index arguments and so the x object is taken as-is, not being applied to any index.
In more detail, the mref function begins by taking sequence as its an accumulator object. Then if there are index arguments, it iterates over them. At each iteration step, it replaces the accumulator by treating the accumulator as a callable object and applying it to index value and taking the resulting value as the new accumulator. After the iteration, the accumulator becomes the return value of the function.
When mref is used as a place, only the rightmost index argument may be a range. If any other argument is a range object, the behavior is unspecified.
When mref is used as a place, and there is only one index which is a range object, then the sequence expression is also required to be a place, if it denotes a list or range object. If there are no index augments then sequence is unconditionally required to be a place.
Note: the functions nested-vec and nested-vec-of may be used to create nested vectors which simulate multi-dimensional arrays.
;; Indexing:
(let ((ar '((1 2 3)
(4 5 6)
(7 8 9))))
(mref ar 1 1))
--> 5
;; Updating value in nested sequence:
(let ((ar (vec (vec (vec 0 1 2 3)
(vec 4 5 6 7))
(vec (vec 8 9 10 11)
(vec 12 13 14 15)))))
(set (mref ar 0 0 1..3) "AB")
ar)
--> #(#(#( 0 #\A #\B 3)
#( 4 5 6 7))
#(#( 8 9 10 11)
#(12 13 14 15)))
;; Invoking curried function:
(let ((cf (lambda (x)
(lambda (y)
(lambda (z)
(+ x y z))))))
[mref cf 1 2 3])
--> 6
(update sequence function)
The update function replaces each elements in sequence in a hash table, with the result of function being applied to that element value.
The sequence is returned.
The sequence may be a hash table. In that case, function is invoked with each hash value, which is replaced with the function's return value.
(remq object sequence [key-function])
(remql object sequence [key-function])
(remqual object sequence [key-function])
The remq, remql and remqual functions produce a new sequence based on sequence, removing the elements whose associated keys are eq, eql or equal to object.
The input sequence is unmodified, but the returned sequence may share substructure with it. If no items are removed, it is possible that the return value is sequence itself.
If key-function is omitted, then the element keys compared to object are the elements themselves. Otherwise, key-function is applied to each element and the resulting value is that element's key which is compared to object.
(remq* object sequence)
(remql* object sequence)
(remqual* object sequence)
The remq*, remql* and remqual* functions are lazy analogs of remq, remql and remqual. Rather than computing the entire new sequence prior to returning, these functions return a lazy list.
Caution: these functions can still get into infinite looping behavior. For instance, in (remql* 0 (repeat '(0))), remql will keep consuming the 0 values coming out of the infinite list, looking for the first item that does not have to be deleted, in order to instantiate the first lazy value.
;; Return a list of all the natural numbers, excluding 13,
;; then take the first 100 of these.
;; If remql is used, it will loop until memory is exhausted,
;; because (range 1) is an infinite list.
[(remql* 13 (range 1)) 0..100]
(keepq object sequence [key-function])
(keepql object sequence [key-function])
(keepqual object sequence [key-function])
The keepq, keepql and keepqual functions produce a new sequence based on sequence, removing the items whose keys are not eq, eql or equal to object.
The input sequence is unmodified, but the returned sequence may share substructure with it. If no items are removed, it is possible that the return value is sequence itself.
The optional key-function is applied to each element from the sequence to convert it to a key which is compared to object. If key-function is omitted, then each element itself of sequence is compared to object.
(remove-if predfun sequence [keyfun [mapfun]])
(keep-if predfun sequence [keyfun [mapfun]])
(separate predfun sequence [keyfun [mapfun]])
(remove-if* predfun sequence [keyfun [mapfun]])
(keep-if* predfun sequence [keyfun [mapfun]])
The remove-if function produces a sequence whose contents are those of sequence but with those elements removed which satisfy predfun. Those elements which are not removed appear in the same order. The result sequence may share substructure with the input sequence, and may even be the same sequence object if no items are removed.
The optional keyfun specifies how each element from the sequence is transformed to an argument to predfun. If this argument is omitted then the predicate function is applied to the elements directly, a behavior which is identical to keyfun being (fun identity).
The optional mapfun argument specifies a function which is applied to the elements of sequence that are identified for retention, mapping them to the actual values that are accumulated into the output. In the absence of this argument, the behavior is to accumulate the elements themselves.
If keyfun and mapfun are the same object, it is unspecified whether mapfun is called, or whether the result of keyfun is used.
The keep-if function is exactly like remove-if, except the sense of the predicate is inverted. The function keep-if retains those items which remove-if will delete, and removes those that remove-if will preserve.
The separate function combines keep-if and remove-if into one, returning a list of two elements whose car and cadr are the result of calling keep-if and remove-if, respectively, on sequence (with the predfun and keyfun arguments passed through). One of the two elements may share substructure with the input sequence, and may even be the same sequence object if all items are either kept or removed (in which case the other element will be nil).
Note: the separate function may be understood in terms of the following reference implementation:
(defun separate (pred seq : (keyfun :))
[(juxt (op keep-if pred @1 keyfun)
(op remove-if pred @1 keyfun))
seq])
The remove-if* and keep-if* functions are like remove-if and keep-if, but produce lazy lists.
;; remove any element numerically equal to 3.
(remove-if (op = 3) '(1 2 3 4 3.0 5)) -> (1 2 4 5)
;; remove those pairs whose first element begins with "abc"
[remove-if (op equal [@1 0..3] "abc")
'(("abcd" 4) ("defg" 5))
car]
-> (("defg" 5))
;; equivalent, without test function
(remove-if (op equal [(car @1) 0..3] "abc")
'(("abcd" 4) ("defg" 5)))
-> (("defg" 5))
(keep-keys-if predfun sequence [keyfun [mapfun]])
(separate-keys predfun sequence [[keyfun])
The functions keep-keys-if and separate-keys are derived, respectively, from the functions keep-if and separate, and have the same syntax and argument semantics. They differ in that rather than accumulating the elements of the input sequence, they accumulate the transformed values of those elements, as projected through the keyfun.
If all arguments of keep-keys-if are specified, then it behaves exactly like keep-if for those same arguments. The same is true if both the keyfun and mapfun arguments are omitted, or if keyfun is specified as identity.
The difference between keep-keys-if and keep-if is the defaulting of the mapfun argument. If mapfun is omitted, then it defaults to being the same function as the keyfun argument.
In the case of separate-keys-if, when keyfun is omitted, thus defaulting to identity, or else explicitly specified as identity or equivalent function, the behavior is same as that of separate.
;; square the values 1 to 20, keeping the even squares
[keep-keys-if evenp (range 1 20) square]
-> (4 16 36 64 100 144 196 256 324 400)
;; square the values 1 to 20 separating into even and odd:
[separate-keys evenp (range 1 20) square]
-> ((4 16 36 64 100 144 196 256 324 400)
(1 9 25 49 81 121 169 225 289 361))
;; contrast with keep-if: values are of input sequence
[keep-if evenp (range 1 20) square]
-> (2 4 6 8 10 12 14 16 18 20)
(countq object iterable)
(countql object iterable)
(countqual object iterable)
The countq, countql and countqual functions count the number of objects in iterable which are eq, eql or equal to object, and return the count.
(count key sequence [testfun [keyfun]])
(count-if predfun iterable [keyfun])
The count and count-if functions search through sequence for items which match key, or satisfy the predicate function predfun, respectively. They return the number of matching or predicate-satisfying items.
The keyfun argument specifies a function which is applied to the elements of sequence to produce the comparison key. If this argument is omitted, then the untransformed elements of sequence are examined.
The count function's testfun argument specifies the test function which is used to compare the comparison keys from sequence to key. If this argument is omitted, then the equal function is used. The count function returns the number of elements of sequence whose comparison key (as retrieved by keyfun) matches the key object, as compared by testfun.
The count-if function's predfun argument specifies a predicate function which is applied to the successive comparison keys taken from sequence. The function returns the count of the number keys for which predfun returns true.
(cons-count obj tree [test-function])
The cons-count function returns the number of times the object obj occurs in the cons cell structure tree, under the equality imposed by the test-function.
If the optional test-function argument is omitted, it defaults to equal.
First, obj and tree are compared using test-function. If they are equal, that counts as one occurrence.
Then, if tree is a cons cell, the function recurses over the car and cdr fields.
The sum of all these counts is returned.
(posq object sequence)
(posql object sequence)
(posqual object sequence)
The posq, posql and posqual functions return the zero-based position of the first item in sequence which is, respectively, eq, eql or equal to object.
(pos key sequence [testfun [keyfun]])
(pos-if predfun sequence [keyfun])
The pos and pos-if functions search through sequence for an item which matches key, or satisfies the predicate function predfun, respectively. They return the zero-based position of the matching item.
The keyfun argument specifies a function which is applied to the elements of sequence to produce the comparison key. If this argument is omitted, then the untransformed elements of sequence are examined.
The pos function's testfun argument specifies the test function which is used to compare the comparison keys from sequence to key. If this argument is omitted, then the equal function is used. The pos function returns the position of the first element of sequence whose comparison key (as retrieved by keyfun) matches key, as compared by the testfun function. If no such element is found, nil is returned.
The pos-if function's predfun argument specifies a predicate function which is applied to the successive comparison keys taken from sequence by applying keyfun to successive elements. The position of the first element for which predfun yields true is returned. If no such element is found, nil is returned.
(rposq object sequence)
(rposql object sequence)
(rposqual object sequence)
(rpos key sequence [testfun [keyfun]])
(rpos-if predfun sequence [keyfun])
These functions are counterparts of rposq, rposql, rposqual, rpos and rpos-if which report position of the rightmost matching item, rather than the leftmost.
(pos-max sequence [testfun [keyfun]])
(pos-min sequence [testfun [keyfun]])
The pos-min and pos-max functions implement exactly the same algorithm; they differ only in their defaulting behavior with regard to the testfun argument. If testfun is not given, then the pos-max function defaults testfun to the greater function, whereas pos-min defaults it to the less function.
If sequence is empty, both functions return nil.
Without a testfun argument, the pos-max function finds the zero-based position index of the numerically maximum value occurring in sequence, whereas pos-min without a testfun argument finds the index of the minimum value.
If a testfun argument is given, the two functions are equivalent. The testfun function must be callable with two arguments. If testfun behaves like a greater-than comparison, then pos-max and pos-min return the index of the maximum element. If testfun behaves like a less-than comparison, then the functions return the index of the minimum element.
The keyfun argument defaults to the identity function. Each element from sequence is passed through this one-argument function, and the resulting value is used in its place.
If a sequence contains multiple equivalent maxima, whether the position of the leftmost or rightmost such maximum is reported depends on whether testfun compares for strict inequality, or whether it reports true for equal arguments also. Under the default testfun, which is less, the pos-max function will return the position leftmost of a duplicate set of maximum elements. To find the rightmost of the maxima, the lequal function can be substituted. Analogous reasoning applies to other test functions.
(subst old new seq [testfun [keyfun]])
The subst function returns a sequence of the same type as seq in which elements of seq which match the old object have been replaced with the new object.
To form the comparison keys, the elements of seq are projected through the testfun function, which defaults to identity, so the items themselves are used as keys by default.
Keys are compared to the old value using testfun, which defaults to equal.
(subst "brown" "black" #("how" "now" "brown" "cow"))
-> #("how" "now" "black" "cow"))
;; elements are converted to lower case to form keys
[subst "brown" "black"
#("how" "now" "BROWN" "cow") : downcase-str]
-> #("how" "now" "black" "cow")
;; using < instead of equality, replace elements
;; greater than 5 with 0
[subst 5 0 '(1 2 3 4 5 6 7 8 9 10) <] (1 2 3 4 5 0 0 0 0 0))
(subq old new sequence)
(subql old new sequence)
(subqual old new sequence)
The subq, subql and subqual functions return a sequence of the same kind as sequence in which elements matching the old object are replaced by new object.
The matching elements are identified by comparing with old using, respectively, the functions eq, eql, and equal.
(subq #\b #\z "abc") -> "azc"
(subql 1 3 #(0 1 2)) -> #(0 3 2)
(subqual "are" "do" '#"how are you")
-> ("how" "do" "you")
(mismatch left-seq right-seq [testfun [keyfun]])
The mismatch function compares corresponding elements from the sequences left-seq and right-seq, returning the position at which the first mismatch occurs.
If the sequences are of the same length, and their corresponding elements are the same, then nil is returned.
If one sequence is shorter than the other, and matches a prefix of the other, then the mismatching position returned is one position after the last element of the shorter sequence, the same value as its length. An empty sequence is a prefix of every sequence.
The keyfun argument defaults to the identity function. Each element from sequence is passed to keyfun and the resulting value is used in its place.
After being converted through keyfun, items are then compared using testfun, which must accept two arguments, and defaults to equal.
(where function iterable)
If iterable is a sequence, the where function returns a lazy list of the numeric indices of those of its elements which satisfy function. The numeric indices appear in increasing order.
If iterable is a hash, the following special behavior applies: where returns a lazy list of of keys which have values which satisfy function. These keys are not subject to an order.
function must be a function that can be called with one argument. For each element of iterable, function is called with that element as an argument. If a non-nil value is returned, then the zero-based index of that element is added to a list. Finally, the list is returned.
(wheref function)
The wheref function is a combinator related to the where function.
The wheref function returns a function that takes one argument. When a sequence is passed to that function, it returns the index positions where the sequence elements satisfy the given function, which must be capable of taking one argument.
Certain uses of where can be expressed more briefly using wheref, according to the following equivalence:
(where f s) <--> [(wheref f) s]
;; partition list of integers by odd, using where:
[partition 0..10 (op where oddp)]
--> ((0) (1 2) (3 4) (5 6) (7 8) (9))
;; using wheref
[partition 0..10 [wheref oddp]]
--> ((0) (1 2) (3 4) (5 6) (7 8) (9))
(whereq object)
(whereql object)
(wherequal object)
The functions whereq, whereql and wherequal are combinators related to the where function.
The whereq function returns a function that takes one argument. When a sequence is passed to that function, it returns the index positions where the elements of the sequence are eq to object.
The whereql function differs only in that the test is eql rather than eq, and the wherequal function uses equal equality.
;; indices where the string has a 'c', using where:
(where (op eq #\c) "abcabc") -> (2 5)
;; same, using whereq:
[(whereq #\c) "abcabc"] -> (2 5)
(rmismatch left-seq right-seq [testfun [keyfun]])
Similarly to mismatch, the rmismatch function compares corresponding elements from the sequences left-seq and right-seq, returning the position at which the first mismatch occurs. All of the arguments have the same semantics as that of mismatch.
Unlike mismatch, rmismatch compares the sequences right-to-left, finding the suffix which they have in common, rather than prefix.
If the sequences match, then nil is returned. Otherwise, a negative index is returned giving the mismatching position, regarded from the end. If the sequences match only in the rightmost element, then -1 is returned. If they match in two elements then -2 and so forth.
(starts-with short-seq long-seq [testfun [keyfun]])
(ends-with short-seq long-seq [testfun [keyfun]])
The starts-with and ends-with functions compare corresponding elements from sequences short-seq and long-seq.
The starts-with function returns t if short-seq is prefix of long-seq; otherwise, it returns nil.
The ends-with function returns t if short-seq is suffix of long-seq; otherwise, it returns nil.
Element from both sequences are mapped to comparison keys using keyfun, which defaults to identity.
Comparison keys are compared using testfun which defaults to equal.
(select sequence {index-seq | function})
The select function returns a sequence, of the same kind as sequence, which consists of those elements of sequence which are identified by the indices in index-seq, which is required to be a sequence.
If a function argument is given instead of index-seq, then function is invoked with sequence as its argument. The return value is then taken as if it were the index-seq argument .
If sequence is a sequence, then index-seq consists of numeric indices. The length of the sequence, as reported by the length function, is added to every index-seq value which is negative. The select function stops collecting values upon encountering an index value which is greater than or equal to the length of the sequence. (Rationale: without this strict behavior, select would not be able to terminate if index-seq is infinite.)
If sequence is, more specifically, a list-like sequence, then index-seq must contain monotonically increasing numeric values, even if no value is out of range, since the select function makes a single pass through the list based on the assumption that indices are ordered. (Rationale: optimization.) This requirement for monotonicity applies to the values which result after negative indices are displaced by the sequence length Also, in this list-like sequence case, values taken from index-seq which are still negative after being displaced by the sequence length are ignored.
If sequence is a hash, then index-seq is a list of keys. A new hash is returned which contains those elements of sequence whose keys appear in index-seq. All of index-seq is processed, even if it contains keys which are not in sequence. The nonexistent keys are ignored.
The select function also supports objects of type carray, in a manner similar to vectors. The indicated elements are extracted from the input sequence, and a new carray is returned whose storage is initialized by converting the extracted values back to the foreign representation.
(reject sequence {index-seq | function})
The reject function returns a sequence, of the same kind as sequence, which consists of all those elements of sequence which are not identified by the indices in index-seq, which may be a list or a vector.
If function is given instead of index-seq, then function is invoked with sequence as its argument. The return value is then taken as if it were the index-seq argument .
If sequence is a hash, then index-seq represents a list of keys. The reject function returns a duplicate of the hash, in which the keys specified in index-seq do not appear.
Otherwise if sequence is a vector-like sequence, then the behavior of reject may be understood by the following equivalence:
(reject seq idx) --> (make-like
[apply append (split* seq idx)]
seq)
where it is to be understood that seq is evaluated only once.
If sequence is a list, then, similarly, the following equivalence applies:
(reject seq idx) --> (make-like
[apply append* (split* seq idx)]
seq)
The input sequence is split into pieces at the indicated indices, such that the elements at the indices are removed and do not appear in the pieces. The pieces are then appended together in order, and the resulting list is coerced into the same type of sequence as the input sequence.
(relate domain-seq range-seq [default-val])
The relate function returns a one-argument function which implements the relation formed by mapping the elements of domain-seq to the positionally corresponding elements of range-seq. That is to say, the function searches through the sequence domain-seq to determine the position where its argument occurs, using equal as the comparison function. Then it returns the element from that position in the range-seq sequence. This returned function is called the relation function.
If the relation function's argument is not found in domain-seq, then the behavior depends on the optional parameter default-val. If an argument is given for default-val, then the relation function returns that value. Otherwise, the relation function returns its argument.
Note: the relate function may be understood in terms of the following equivalences:
(relate d r) <--> (lambda (arg)
(iflet ((p (posqual arg d)))
[r p]
arg))
(relate d r v) <--> (lambda (arg)
(iflet ((p (posqual arg d)))
[r p]
v))
Note: relate may return a hash table instead of a function, if such an object can satisfy the semantics required by the arguments.
(mapcar (relate "_" "-") "foo_bar") -> "foo-bar"
(mapcar (relate "0123456789" "ABCDEFGHIJ" "X") "139D-345")
-> "BJDXXDEF"
(mapcar (relate '(nil) '(0)) '(nil 1 2 nil 4)) -> (0 1 2 0 4)
(in sequence key [testfun [keyfun]])
(in hash key)
The in function tests whether key is found inside sequence or hash.
If the testfun argument is specified, it specifies the function which is used to comparison keys from the sequence to key. Otherwise the equal function is used.
If the keyfun argument is specified, it specifies a function which is applied to the elements of sequence to produce the comparison keys. Without this argument, the elements themselves are taken as the comparison keys.
If the object being searched is a hash, then if neither of the arguments keyfun nor testfun is specified, in performs a hash lookup for key, returning t if the key is found, nil otherwise. If either of keyfun or testfun is specified, then in performs an exhaustive search of the hash table, as if it were a sequence of cons cells whose car fields are keys, and whose cdr keys are values. Thus to search by key, the car function must be specified as keyfun.
The in function returns t if it finds key in sequence or hash, otherwise nil.
(partition sequence {index-seq | index | function})
If sequence is empty, then partition returns an empty list, and the second argument is ignored; if it is function, it is not called.
Otherwise, partition returns a lazy list of partitions of sequence. Partitions are consecutive, non-overlapping, nonempty substrings of sequence, of the same kind as sequence, such that if these substrings are catenated together in their order of appearance, a sequence equal to the original is produced.
If the second argument is of the form index-seq, or if an index-seq was produced from the index or function arguments, each value in that sequence must be an integer. Each integer value which is nonnegative specifies the index position given by its value. Each integer value which is negative specifies an index position given by adding the length of sequence to its value. The sequence index positions thus denoted by index-seq shall be strictly nondecreasing. Each successive element is expected to designate an index position at least as high as all previous elements, otherwise the behavior is unspecified. Index values which are still negative after the addition of the sequence length are ignored, as are index values greater than the sequence length. Nondecreasing means that repeated values are permitted; they have the same effect as a single value.
If index-seq is empty then a one-element list containing the entire sequence is returned.
If index-seq is an infinite lazy list, the function shall terminate if that list eventually produces an index position which is greater than or equal to the length of sequence.
If the second argument is a function, then this function is applied to sequence, and the return value of this call is then used in place of the second argument, which must either be a single index value, which is then taken as if it were the index argument, or else a sequence of indices, which are taken as the index-seq argument.
If the second argument is neither a sequence, nor a function, then it is assumed to be an integer index, and is turned into an index-seq sequence containing one element.
After the index-seq is obtained as an argument, or determined from the index or function arguments, the partition function then divides sequence according to the indices. The first partition begins with the first element of sequence. The second partition begins at the first position in index-seq, and so on. Indices beyond the length of the sequence are ignored, as are indices less than or equal to zero.
(partition '(1 2 3) 1) -> ((1) (2 3))
;; split the string where there is a "b"
(partition "abcbcbd" (op where (op eql #\b))) -> ("a" "bc"
"bc" "bd")
(split sequence {index-seq | index | function})
(split* sequence {index-seq | index | function})
If sequence is empty, then both split and split* return an empty list, and the second argument is ignored; if it is function, it is not called.
Otherwise, split returns a lazy list of pieces of sequence: consecutive, non-overlapping, possibly empty substrings of sequence, of the same kind as sequence. A catenation of these pieces in the order they appear would produce a sequence that is equal to the original sequence.
The split* function differs from split in that the elements indicated by the split indices are removed.
The index, index-seq, and function arguments are subject to the same restrictions and treatment as the corresponding arguments of the partition function, with the following difference: the index positions indicated by index-seq are required to be strictly increasing, rather than nondecreasing. As with partition, this consideration applies to the transformed indices, after the displacement of negative values by the length of the sequence. If any element of index-seq is not higher than the previous element, the behavior is unspecified.
If the second argument is of the form index-seq, or if an index-seq was produced from the index or function arguments, then the split function divides sequence according to the indices indicated in index-seq. The first piece always begins with the first element of sequence. Each subsequent piece begins with the position indicated by an element of index-seq. Negative indices are ignored. If index-seq includes index zero, then an empty first piece is generated. If index-seq includes an index greater than or equal to the length of sequence (equivalently, an index beyond the last element of the sequence) then an additional empty last piece is generated. The length of sequence is added to any negative indices. An index which is still negative after being thus displaced is discarded.
Note: the principal difference between split and partition is that partition does not produce empty pieces.
(split '(1 2 3) 1) -> ((1) (2 3))
(split "abc" 0) -> ("" "abc")
(split "abc" 3) -> ("abc" "")
(split "abc" 1) -> ("a" "bc")
(split "abc" '(0 1 2 3)) -> ("" "a" "b" "c" "")
(split "abc" '(1 2)) -> ("a" "b" "c")
(split "abc" '(-1 1 2 15)) -> ("a" "b" "c")
;; triple split at makes two additional empty pieces
(split "abc" '(1 1 1)) -> ("a" "" "" "bc")
(split* "abc" 0) -> ("" "bc") ;; "a" is removed
;; all characters removed
(split* "abc" '(0 1 2)) -> ("" "" "" "")
(partition* sequence {index-seq | index | function})
If sequence is empty, then partition* returns an empty list, and the second argument is ignored; if it is function, it is not called.
The index, index-seq, and function arguments are subject to the same restrictions and treatment as the corresponding arguments of the partition function, with the following difference: the index positions indicated by index-seq are required to be strictly increasing, rather than nondecreasing.
If the second argument is of the form index-seq, then partition* produces a lazy list of pieces taken from sequence. The pieces are formed by deleting from sequence the elements at the positions given in index-seq, such that the pieces are the remaining nonempty substrings from between the deleted elements, maintaining their order.
If index-seq is empty then a one-element list containing the entire sequence is returned.
(partition* '(1 2 3 4 5) '(0 2 4)) -> ((2) (4))
(partition* "abcd" '(0 3)) -> "bc"
(partition* "abcd" '(0 1 2 3)) -> nil
(find key sequence [testfun [keyfun]])
(find-if predfun {sequence | hash} [keyfun])
(find-true predfun {sequence | hash} [keyfun])
The find and find-if functions search through a sequence for an item which matches a key, or satisfies a predicate function, respectively. The find-true function is a variant of find-if which returns the value of the predicate function instead of the item.
The keyfun argument specifies a function which is applied to the elements of sequence to produce the comparison key. If this argument is omitted, then the untransformed elements of the sequence are searched.
The find function's testfun argument specifies the test function which is used to compare the comparison keys from sequence to the search key. If this argument is omitted, then the equal function is used. The first element from the list whose comparison key (as retrieved by keyfun) matches the search (under testfun) is returned. If no such element is found, nil is returned.
The find-if function's predfun argument specifies a predicate function which is applied to the successive comparison keys pulled from the list by applying keyfun to successive elements. The first element for which predfun yields true is returned. If no such element is found, nil is returned.
In the case of find-if, a hash table may be specified instead of a sequence. The hash is treated as if it were a sequence of hash key and hash value pairs represented as cons cells, the car slots of which are the hash keys, and the cdr of which are the hash values. If the caller doesn't specify a keyfun then these cells are taken as their keys.
The find-true function's argument conventions and search semantics are identical to those of find-if, but the return value is different. Instead of returning the found item, find-true returns the value which predfun returned for the found item's key.
(rfind key sequence [testfun [keyfun]])
(rfind-if predfun {sequence | hash} [keyfun])
The rfind and rfind-if functions are almost exactly like find and find-if except that if there are multiple matches for key in sequence, they return the rightmost element rather than the leftmost.
In the case of rfind-if when a hash is specified instead of a sequence, the function searches through the hash entries in the same order as find-if, but finds the last match rather than the first. Note: hashes are inherently not ordered; the relative order of items in a hash table can change when other items are inserted or deleted.
(find-max iterable [testfun [keyfun]])
(find-min iterable [testfun [keyfun]])
The find-min and find-max function implement exactly the same algorithm; they differ only in their defaulting behavior with regard to the testfun argument. If testfun is not given, then the find-max function defaults it to the greater function, whereas find-min defaults it to the less function.
Without a testfun argument, the find-max function finds the numerically maximum value occurring in iterable, whereas pos-min without a testfun argument finds the minimum value.
If a testfun argument is given, the two functions are equivalent. The testfun function must be callable with two arguments. If testfun behaves like a greater-than comparison, then find-max and find-min both return the maximum element. If testfun behaves like a less-than comparison, then the functions return the minimum element.
The keyfun argument defaults to the identity function. Each element from sequence is passed through this one-argument function, and the resulting value is used in its place for the purposes of the comparison. However, the original element is returned.
If there are multiple equivalent maxima, then under the default testfun, that being less, the first one encountered while traversing iterable is the one that is reported. See the notes under pos-max regarding duplicate maxima.
(find-maxes iterable [testfun [keyfun]])
(find-mins iterable [testfun [keyfun]])
The find-maxes and find-mins functions have the same argument conventions as, respectively, find-max and find-min. These functions differ in that they return a sequence of all the elements of iterable which maximize the value of keyfun. The returned sequence is of the same kind as iterable.
(find-max-key iterable [ testfun [keyfun]])
(find-min-key iterable [ testfun [keyfun]])
The find-max-key and find-min-key functions have the same argument conventions as, respectively, find-max and find-min and agree with those functions in regard to which element of the input sequence is identified: all these functions identify the element which maximizes or minimizes the value of keyfun.
Whereas find-max and find-min return the maximizing or minimizing element itself, the find-max-key and find-min-key functions return the value of keyfun applied to the element.
Under the default keyfun value, that being the identity function, these functions behave the same as find-max and find-min.
(uni iter1 iter1 [testfun [keyfun]])
(isec iter1 iter1 [testfun [keyfun]])
(isecp iter1 iter1 [testfun [keyfun]])
(diff iter1 iter1 [testfun [keyfun]])
(symdiff iter1 iter2 [testfun [keyfun]])
The functions uni, isec, diff and symdiff treat the sequences iter1 and iter2 as if they were sets.
They, respectively, compute the set union, set intersection, set difference and symmetric difference of iter1 and iter2, returning a new sequence.
The isecp is Boolean: it returns t for those arguments for which isec returns a non-empty list, otherwise nil.
The arguments iter1 and iter2 need not be of the same kind. They may be hash tables.
The returned sequence is of the same kind as iter1. If iter1 is a hash table, the returned sequence is a list.
For the purposes of these functions, an input which is a hash table is considered as if it were a sequence of hash key and hash value pairs represented as cons cells, the car slots of which are the hash keys, and the cdr of which are the hash values. This means that if no keyfun is specified, these pairs are taken as keys.
Since the input sequences are defined as representing sets, they are assumed not to contain duplicate elements. These functions are not required, but may, de-duplicate the sequences.
The union sequence produced by uni contains all of the elements which occur in both iter1 and iter2. If a given element occurs exactly once only in iter1 or exactly once only in iter2, or exactly once in both sequences, then it occurs exactly once in the union sequence. If a given element occurs at least once in either iter1, iter2 or both, then it occurs at least once in the union sequence.
The intersection sequence produced by isec contains all of the elements which occur in both iter1 and iter2. If a given element occurs exactly once in iter1 and exactly once in iter2, then in occurs exactly once in the intersection sequence. If a given element occurs at least once in iter1 and at least once in iter2, then in occurs at least once in the intersection sequence.
The difference sequence produced by diff contains all of the elements which occur in iter1 but do not occur in iter2. If an element occurs exactly once in iter1 and does not occur in iter2, then it occurs exactly once in the difference sequence. If an element occurs at least once in iter1 and does not occur in iter2, then it occurs at least once in the difference sequence. If an element occurs at least once in iter2, then it does not occur in the difference sequence.
The symmetric difference sequence produced by symdiff contains all of the elements of iter1 which do not occur in iter2 and vice versa: it also contains all of the elements of iter2 which do not occur in iter1.
Element equivalence is determined by a combination of testfun and keyfun. Elements are compared pairwise, and each element of a pair is passed through keyfun function to produce a comparison value. The comparison values are compared using testfun. If keyfun is omitted, then the untransformed elements themselves are compared, and if testfun is omitted, then the equal function is used.
Note: a function similar to diff named set-diff exists. This became deprecated starting in TXR 184. For the set-diff function, the requirement was specified to preserve the original order of items from iter1 that survive into the output sequence. This requirement is not documented for the diff function, but is de facto honored by the implementation for at as long as the set-diff synonym continues to be available. The set-diff function doesn't support hash tables and is inefficient for vectors and strings.
Note: these functions are not efficient for the processing of hash tables, even when both inputs are hashes, the keyfun argument is car, and testfun matches the equality used by both hash-table inputs. If applicable, the operations hash-uni, hash-isec and hash-diff should be used instead.
(mapcar function iterable*)
(mappend function iterable*)
(mapcar* function iterable*)
(mappend* function iterable*)
When given only one argument, the mapcar function returns nil. function is never called.
When given two arguments, the mapcar function applies function to each elements of iterable and returns a sequence of the resulting values in the same order as the original values. The returned sequence is the same kind as iterable, if possible. If the accumulated values cannot be elements of that type of sequence, then a list is returned.
When additional sequences are given as arguments, this filtering behavior is generalized in the following way: mapcar traverses the sequences in parallel, taking a value from each sequence as an argument to the function. If there are two lists, function is called with two arguments and so forth. The traversal is limited by the length of the shortest sequence. The return values of the function are collected into a new sequence which is returned. The returned sequence is of the same kind as the leftmost input sequence, unless the accumulated values cannot be elements of that type of sequence, in which case a list is returned.
The mappend function works like mapcar, with the following difference. Rather than accumulating the values returned by the function into a sequence, mappend expects the items returned by the function to be sequences which are catenated with append, and the resulting sequence is returned. The returned sequence is of the same kind as the leftmost input sequence, unless the values cannot be elements of that type of sequence, in which case a list is returned.
The mapcar* and mappend* functions work like mapcar and mappend, respectively. However, they return lazy lists rather than generating the entire output list prior to returning.
Like mappend, mappend* must "consume" empty lists. For instance, if the function being mapped puts out a sequence of nils, then the result must be the empty list nil, because (append nil nil nil nil ...) is nil.
But suppose that mappend* is used on inputs which are infinite lazy lists, such that the function returns nil values indefinitely. For instance:
;; Danger: infinite loop!!!
(mappend* (fun identity) (repeat '(nil)))
The mappend* function is caught in a loop trying to consume and squash an infinite stream of nils, and so doesn't return.
;; multiply every element by two
(mapcar (lambda (item) (* 2 item)) '(1 2 3)) -> (4 6 8)
;; "zipper" two lists together
(mapcar (lambda (le ri) (list le ri)) '(1 2 3) '(a b c))
-> '((1 a) (2 b) (3 c)))
;; like append, mappend allows a lone atom or a trailing atom:
(mappend (fun identity) 3) -> (3)
(mappend (fun identity) '((1) 2)) -> (1 . 2)
;; take just the even numbers
(mappend (lambda (item) (if (evenp x) (list x))) '(1 2 3 4 5))
-> (2 4)
(maprod function iterable*)
(maprend function iterable*)
(maprodo function iterable*)
The maprod, maprend and maprodo functions resemble mapcar, mappend and mapdo, respectively. When given no iterable arguments or exactly one iterable argument, they behave exactly like those three functions.
When two or more iterable arguments are present, maprod differs from mapcar in the following way, as do the remaining functions from their aforementioned counterparts. Whereas mapcar iterates over the iterable values in parallel, taking successive tuples of element values and passing them to function, the maprod function iterates over all combinations of elements from the sequences: the Cartesian product. The prod suffix stands for "product".
If one or more iterable arguments specify an empty sequence, then the Cartesian product is empty. In this situation, function is not called. The result of the function is then nil converted to the same kind of sequence as the leftmost iterable.
The maprod function collects the values into a list just as mapcar does. Just like mapcar, it converts the resulting list into the same kind of sequence as the leftmost iterable argument, if possible. For instance, if the resulting list is a list or vector of characters, and the leftmost iterable is a character string, then the list or vector of characters is converted to a character string and returned.
The maprend function ("map product through function and append") iterates the iterable element combinations exactly like maprod, passing them as arguments to function. The values returned by function are then treated exactly as by the mappend function. The return values are expected to be sequences which are appended together as if by append, and the final result is converted to the same kind of sequence as the leftmost iterable if possible.
The maprodo function, like mapdo, ignores the result of function and returns nil.
The combination iteration gives priority to the rightmost iterable, which means that the rightmost element of each generated tuple varies fastest: the tuples are traversed in "rightmost major" order. This is made clear in the examples.
[maprod list '(0 1 2) '(a b) '(i ii iii)]
->
((0 a i) (0 a ii) (0 a iii) (0 b i) (0 b ii) (0 b iii)
(1 a i) (1 a ii) (1 a iii) (1 b i) (1 b ii) (1 b iii)
(2 a i) (2 a ii) (2 a iii) (2 b i) (2 b ii) (2 b iii))
;; Vectors #(#\a #\x) #(#\a #\y) ... are appended
;; together resulting in #(#\a #\x #\a #\y ...)
;; which is converted to a string:
[maprend vec "ab" "xy"] -> "axaybxby"
;; One of the sequences is empty, so the product is an
;; empty sequence of the same kind as the leftmost
;; sequence argument, thus an empty string:
[maprend vec "ab" ""] -> ""
(mapdo function iterable*)
The mapdo function is similar to mapcar, but always returns nil. It is useful when function performs some kind of side effect, hence the "do" in the name, which is a mnemonic for the execution of imperative actions.
When only the function argument is given, function is never called, and nil is returned.
If a single iterable argument is given, then mapdo iterates over iterable, invoking function on each element.
If two or more iterable arguments are given, then mapdo iterates over the sequences in parallel, extracting parallel tuples of items. These tuples are passed as arguments to function, which must accept as many arguments as there are sequences.
(transpose iterable)
(zip iterable*)
The transpose function performs a transposition on iterable. This means that the elements of iterable must be iterable. These iterables are understood to be columns; transpose exchanges rows and columns, returning a sequence of the rows which make up the columns. The returned sequence is of the same kind as iterable. The rows are also the same kind of sequence as the first element of the original sequence, if possible, otherwise they are lists. The number of rows returned is limited by the shortest column among the sequences.
The zip function takes variable arguments, and is equivalent to calling transpose on a list of the arguments. The following equivalences hold:
(zip . x) <--> (transpose x)
[apply zip x] <--> (transpose x)
A special requirement applies when the first argument of zip or the first element of the iterable argument of transpose is a string. In this situation, the tuples which emerge are strings, if possible. The special requirement is that column elements which are strings are treated as individual items and appended to the row strings. For example, (zip "ab" #("rst" "xyz")) produces ("arst" "bxyz"), rather than (("a" "rst") ("b" "xyz")).
;; transpose list of lists
(transpose '((a b c) (c d e))) -> ((a c) (b d) (c e))
;; transpose vector of strings:
;; - string columns become string rows
;; - vector input becomes vector output
(transpose #("abc" "def" "ghij")) -> #("adg" "beh" "cfi")
;; error: transpose wants to make a list of strings
;; but 1 is not a character
(transpose #("abc" "def" '(1 2 3))) ;; error!
;; String elements are catenated:
(transpose #("abc" "def" ("UV" "XY" "WZ")))
-> #("adUV" "beXY" "cfWZ")
;; Transpose list of ranges
(transpose (list 1..4 4..8 8..12))
-> ((1 4 8) (2 5 9) (3 6 10))
(zip '(a b c) '(c d e)) -> ((a c) (b d) (c e))
(window-map range boundary function sequence)
(window-mappend range boundary function sequence)
(window-mapdo range boundary function sequence)
The window-map and window-mappend functions process the elements of sequence by passing arguments derived from each successive element to function. Both functions return, if possible, a sequence of the same kind as sequence, otherwise a list.
Under window-map, values returned by function are accumulated into a sequence of the same type as sequence and that sequence is returned. Under window-mappend, the values returned by the calls to function are expected to be sequence which are appended together to form the output sequence.
These functions are analogous to mapcar and mappend. Unlike these, they operate only on a single sequence, and over this sequence they perform a sliding window mapping, whose description follows.
The function window-mapdo avoids accumulating a sequence, and instead returns nil; it is analogous to mapdo.
The argument to the range parameter must be a positive integer, not exceeding 512. This parameter specifies the amount of ahead/behind context on either side of each element which is processed. It indirectly determines the window size for the mapping. The window size is twice range, plus one. For instance if range is 2, then the window size is 5: the element being processed lies at the center of the window, flanked by two elements on either side, making five.
The function argument must specify a function which accepts a number of arguments corresponding to the window size. For instance if range is 2, making the window size 5, then function must accept 5 arguments. These arguments constitute the sliding window being processed. Each time function is called, the middle argument is the element being processed, and the arguments surrounding it are its window.
When an element is processed from somewhere in the interior of a sequence, where it is flanked on either side by at least range elements, then the window is populated by those flanking elements taken from sequence.
The boundary parameter specifies the window contents which are used for the processing of elements which are closer than range to either end of the sequence. Except if it is of list type, boundary must be a sequence containing at least twice range number of elements (one less than the window size): if it has additional elements, they are not used. If it is a list, it may be shorter than twice range; in this case, the value nil is substituted for the missing elements. The argument may also be one of the two keyword symbols :wrap or :reflect, described below.
If boundary is a sequence, it may be regarded as divided into two pieces of range length. These two pieces then flank sequence on either end. The left half of boundary is effectively prepended to the sequence, and the right half effectively appended. When the sliding window extends beyond the boundary of sequence near its start or end, the window is populated from these flanking elements obtained from boundary.
If boundary argument is specified as the keyword :wrap, then the sequence is imagined to be flanked at either end by an infinite repetition of copies of itself. These flanks are trimmed to the window size to generate the boundary.
For instance if the sequence is (1 2 3) and the window size is 9 due to the value of range being 4, then the behavior of :wrap is as if boundary value of (3 1 2 3 1 2 3 1) were specified. The left flank is (3 1 2 3), being the last four elements of an infinite repetition of 1 2 3; and the right flank is similarly (1 2 3 1), being the first four elements of an infinite repetition of 1 2 3.
If boundary is given as the keyword :reflect, then the sequence is imagined to be flanked at either end by an infinite repetition of reversed copies of itself. These flanks are trimmed to the window size to generate the boundary. For instance if the sequence is (1 2 3) and the window size is 9 due to the value of range being 4, then the behavior of :reflect is as if boundary value of (1 3 2 1 3 2 1 3) were specified. The left flank is (1 3 2 1), being the last four elements of an infinite repetition of 3 2 1; and the right flank is similarly (3 2 1 3), being the first four elements of an infinite repetition of 3 2 1.
;; change characters between angle brackets to upper case.
[window-map 1 nil (lambda (x y z)
(if (and (eq x #\<)
(eq z #\>))
(chr-toupper y)
y))
"ab<c>de<f>g"]
--> "ab<C>de<F>g"
;; collect all numbers which are the centre element of
;; a monotonically increasing triplet
[window-mappend 1 :reflect (lambda (x y z)
(if (< x y z)
(list y)))
'(1 2 1 3 4 2 1 9 7 5 7 8 5)]
--> (3 7)
;; calculate a moving average with a five-element
;; window, flanked by zeros at the boundaries:
[window-map 2 #(0 0 0 0)
(lambda (. args) (/ (sum args) 5))
#(4 7 9 13 5 1 6 11 10 3 8)]
--> #(4.0 6.6 7.6 7.0 6.8 7.2 6.6 6.2 7.6 6.4 4.2))
(interpose sep sequence)
The interpose function returns a sequence of the same type as sequence, in which the elements from sequence appear with the sep value inserted between them.
If sequence is an empty sequence or a sequence of length 1, then a sequence identical to sequence is returned. It may be a copy of sequence or it may be sequence itself.
If sequence is a character string, then the value sep must be a character.
It is permissible for sequence, or for a suffix of sequence to be a lazy list, in which case interpose returns a lazy list, or a list with a lazy suffix.
(interpose #\- "xyz") -> "x-y-z"
(interpose t nil) -> nil
(interpose t #()) -> #()
(interpose #\a "") -> ""
(interpose t (range 0 0)) -> (0)
(interpose t (range 0 1)) -> (0 t 1)
(interpose t (range 0 2)) -> (0 t 1 t 2)
(reduce-left binary-function list
[init-value [key-function]])
(reduce-right binary-function list
[init-value [key-function]])
The reduce-left and reduce-right functions reduce lists of operands specified by list and init-value to a single value by the repeated application of binary-function.
In the case of reduce-left, the list argument is required to be an object which is iterable according to the iter-begin function. The reduce-right function treats the list argument using list operations.
An effective list of operands is formed by combining list and init-value. If key-function is specified, then the items of list are mapped to new values through key-function, as if by mapcar. If init-value is supplied, then in the case of reduce-left, the effective list of operands is formed by prepending init-value to list. In the case of reduce-right, the effective operand list is produced by appending init-value to list. The init-value isn't mapped through key-function.
The production of the effective list can be expressed like this, though this is not to be understood as the actual implementation:
(append (if init-value-present (list init-value))
[mapcar (or key-function identity) list]))))
In the reduce-right case, the arguments to append are reversed.
If the effective list of operands is empty, then binary-function is called with no arguments at all, and its value is returned. This is the only case in which binary-function is called with no arguments; in all remaining cases, it is called with two arguments.
If the effective list contains one item, then that item is returned.
Otherwise, the effective list contains two or more items, and is decimated as follows.
Note that an init-value specified as nil is not the same as a missing init-value; this means that the initial value is the object nil. Omitting init-value is the same as specifying a value of : (the colon keyword symbol). It is possible to specify key-function while omitting an init-value argument. This is achieved by explicitly specifying : as the init-value argument.
Under reduce-left, the leftmost pair of operands is removed from the list and passed as arguments to binary-function, in the same order that they appear in the list, and the resulting value initializes an accumulator. Then, for each remaining item in the list, binary-function is invoked on two arguments: the current accumulator value, and the next element from the list. After each call, the accumulator is updated with the return value of binary-function. The final value of the accumulator is returned.
Under reduce-right, the list is processed right to left. The rightmost pair of elements in the effective list is removed, and passed as arguments to binary-function, in the same order that they appear in the list. The resulting value initializes an accumulator. Then, for each remaining item in the list, binary-function is invoked on two arguments: the next element from the list, in right to left order, and the current accumulator value. After each call, the accumulator is updated with the return value of binary-function. The final value of the accumulator is returned.
;;; effective list is (1) so 1 is returned
(reduce-left (fun +) () 1 nil) -> 1
;;; computes (- (- (- 0 1) 2) 3)
(reduce-left (fun -) '(1 2 3) 0 nil) -> -6
;;; computes (- 1 (- 2 (- 3 0)))
(reduce-right (fun -) '(1 2 3) 0 nil) -> 2
;;; computes (* 1 2 3)
(reduce-left (fun *) '((1) (2) (3)) nil (fun first)) -> 6
;;; computes 1 because the effective list is empty
;;; and so * is called with no arguments, which yields 1.
(reduce-left (fun *) nil)
(some sequence [predicate-fun [key-fun]])
(all sequence [predicate-fun [key-fun]])
(none sequence [predicate-fun [key-fun]])
The some, all and none functions apply a predicate-test function predicate-fun over a list of elements. If the argument key-fun is specified, then elements of sequence are passed into key-fun, and predicate-fun is applied to the resulting values. If key-fun is omitted, the behavior is as if key-fun were the identity function. If predicate-fun is omitted, the behavior is as if predicate-fun were the identity function.
These functions have short-circuiting semantics and return conventions similar to the and and or operators.
The some function applies predicate-fun to successive values produced by retrieving elements of list and processing them through key-fun. If the list is empty, it returns nil. Otherwise it returns the first non-nil return value returned by a call to predicate-fun and stops evaluating the elements. If predicate-fun returns nil for all elements, some returns nil.
The all function applies predicate-fun to successive values produced by retrieving elements of list and processing them through key-fun. If the list is empty, it returns t. Otherwise, if predicate-fun yields nil for any value, the all function immediately returns without invoking predicate-fun on any more elements. If all the elements are processed, then the all function returns the value which predicate-fun yielded for the last element.
The none function applies predicate-fun to successive values produced by retrieving elements of list and processing them through key-fun. If the list is empty, it returns t. Otherwise, if predicate-fun yields non-nil for any value, the none function immediately returns nil. If predicate-fun yields nil for all values, the none function returns t.
;; some of the integers are odd
[some '(2 4 6 9) oddp] -> t
;; none of the integers are even
[none '(1 3 4 7) evenp] -> t
(multi function sequence*)
The multi function distributes an arbitrary list processing function multi over multiple sequences given by the list arguments.
The sequence arguments are first transposed into a single list of tuples. Each successive element of this transposed list consists of a tuple of the successive items from the lists. The length of the transposed list is that of the shortest list argument.
The transposed list is then passed to function as an argument.
The function is expected to produce a list of tuples, which are transposed again to produce a list of lists which is then returned.
Conceptually, the input sequences are columns and function is invoked on a list of the rows formed from these columns. The output of function is a transformed list of rows which is reconstituted into a list of columns.
;; Take three lists in parallel, and remove from all of them the
;; element at all positions where the third list has an element of 20.
(multi (op remove-if (op eql 20) @1 third)
'(1 2 3)
'(a b c)
'(10 20 30))
-> ((1 3) (a c) (10 30))
;; The (2 b 20) "row" is gone from the three "columns".
;; Note that the (op remove if (op eql 20) @1 third)
;; expression can be simplified using the ap operator:
;;
;; (op remove-if (ap eql @3 20))
(sort sequence [lessfun [keyfun]])
(nsort sequence [lessfun [keyfun]])
(ssort sequence [lessfun [keyfun]])
(snsort sequence [lessfun [keyfun]])
The nsort function destructively sorts sequence, producing a sequence which is sorted according to the lessfun and keyfun arguments.
The keyfun argument specifies a function which is applied to elements of the sequence to obtain the key values which are then compared using the lessfun. If keyfun is omitted, the identity function is used by default: the sequence elements themselves are their own sort keys.
The lessfun argument specifies the comparison function which determines the sorting order. It must be a binary function which can be invoked on pairs of keys as produced by the key function. It must return a non-nil value if the left argument is considered to be lesser than the right argument. For instance, if the numeric function < is used on numeric keys, it produces an ascending sorted order. If the function > is used, then a descending sort is produced. If lessfun is omitted, then it defaults to the generic less function.
The sort function has the same argument requirements as nsort but is non-destructive: it returns a new object, leaving the input sequence unmodified, as if a copy of the input object were made using the function copy and then that copy were sorted in-place using nsort.
The sort and nsort functions are stable for sequences which are lists. This means that the original order of items which are considered identical is preserved. For strings and vectors, sort and nsort are not stable.
The ssort and nsort functions have the same argument syntax and semantics as, respectively, sort and nsort. These functions provide a stable sort for all sequences, not only lists, at the cost of temporarily allocating memory.
All of these functions can be applied to hashes. They produce meaningful behavior for a hash table which contains N keys which are the integers from 0 to N - 1. Such as hash is treated as if it were a vector. The values are sorted and reassigned to sorted order to the integer keys. The behavior is not specified for hashes whose contents do not conform to this convention.
Note: nsort was introduced in TXR 238. Prior to that version, sort behaved like nsort.
(csort sequence [lessfun [keyfun]])
(cnsort sequence [lessfun [keyfun]])
(cssort sequence [lessfun [keyfun]])
(csnsort sequence [lessfun [keyfun]])
The functions csort, cnsort, cssort and csnsort are caching counterparts of, respectively, sort, nsort, ssort and snsort. They have exactly the same argument syntax and semantics.
Caching refers to eliminating repeated calls to keyfun for the same element of sequence, in order to reduce the execution time, at the cost of using more storage.
(grade sequence [lessfun [keyfun]])
The grade function returns a list of integer indices which indicate the position of the elements of sequence in sorted order.
The lessfun and keyfun arguments behave like those of the sort function.
The sequence object is not modified.
The internal sort performed by grade is not stable. The indices of any elements considered equivalent under lessfun may appear in any order in the returned index sequence.
Note: the grade function is inspired by the "grade up" and "grade down" operators in the APL language.
;; Order of the 2 3 positions of the "l"
;; characters is not specified:
[grade "Hello"] -> (0 1 2 3 4)
[grade "Hello" >] -> (4 2 3 1 0)
(shuffle sequence [random-state])
(nshuffle sequence [random-state])
(cshuffle sequence [random-state])
(cnshuffle sequence [random-state])
The nshuffle function pseudorandomly rearranges the elements of sequence. This is performed in place: sequence object is modified.
The return value is sequence itself.
The rearrangement depends on pseudorandom numbers obtained from the rand function. The random-state argument, if present, is passed to that function.
The cnshuffle function also pseudorandomly rearranges the elements of sequence. It differs from nshuffle in that it produces a cyclic permutation: a permutation consisting of a single cycle.
Whereas nshuffle may possibly map some, or even all elements to their original locations, cnshuffle maps every element to a new location (if sequence has at least two elements.
An example of a cyclic permutation is the mapping of (1 2 3 4) to (3 1 4 2). The cycle consists of 1 mapping to 3, 3 mapping to 4, 4 mapping to 2, and 2 mapping back to 1. An example of a permutation which is not cyclic is (1 2 3 4) to (1 3 4 2) which contains two cycles: (1) maps to (1) and (2 3 4) maps to (3 4 2). The cnshuffle function will not produce this permutation; the nshuffle function may.
The nshuffle and cnshuffle functions support hash tables in a manner analogous to the way nsort supports hash tables; the same remarks apply as in the description of that function.
The shuffle and cshuffle functions have the same argument requirements and semantics as nshuffle and cnshuffle respectively, but differ in that it they in-place modification of sequence: a new, shuffled sequence is returned, as if a copy of sequence were made using copy and then that copy were shuffled in-place and returned.
Note: nshuffle was introduced in TXR 238. Prior to that version, shuffle behaved like nshuffle.
Note: The pseudo-random number generator in TXR has only 512 bits of state, which is sufficient for generating the all the permutations of sequences of at most 98 elements, and the cyclic permutations of sequences of at most 99 elements. These figures should not be interpreted as guarantees, but as theoretical maxima.
(rot sequence [displacement])
(nrot sequence [displacement])
The nrot and rot functions rotate the elements of sequence, returning a rotated sequence.
The nrot function does this destructively; it modifies sequence in-place, whereas rot returns a new sequence without modifying the original.
The rot function always returns a new sequence. In cases when no rotation is performed, it copies sequence as if using the copy function. In cases when no rotation is performed, the nrot function returns the original sequence, which is unmodified.
The displacement parameter, an integer, has a default value of 1.
To rotate elements means to displace their position within the sequence by some amount, that being given by the displacement parameter, while partially preserving their circular order. Circular order means that for the purposes of rotation, the sequence is regarded to be cyclic: the first element of the sequence is considered to be the successor of the last element and vice versa. Thus, when an element is displaced past the first or last position, it wraps to the end or beginning of the sequence.
If the sequence is empty, or contains only one element, then rot and nrot terminate, performing no rotation. The following remarks apply to situations when sequence has two or more elements.
The displacement parameter, which may be negative, is first reduced to the smallest positive residue modulo the length of the sequence, resulting in a value ranging from zero to one less than the sequence length. If the resulting value is zero, then no rotation is performed.
The displacement has a negative orientation: each element's position is decreased by this amount. Those elements whose position would become negative move to the end of the sequence.
The default displacement of 1 causes the first element to become last, the second element to become first, and so forth. The opposite rotation can be obtained using -1 as the displacement.
Note: even though nrot operates destructively, the returned object may not be the same object as sequence. Only the returned object is required to be the rotated sequence. If this is different from the original sequence input, the contents of that original object are unspecified.
Note: the symbol rotate is the name of a place-mutating macro, which is much older than these functions. If S is a three-element sequence, then:
(set S (nrot S)) ;; alternatively: (upd S nrot)
has the same effect as:
(rotate [S 0] [S 1] [S 2])
(rot "abc") -> "bca"
(rot #(1 2 3) -1) -> (3 1 2)
;; lower-case rot-13
(mapcar (relate (range #\a #\z)
(rot (range #\a #\z) 13))
"hello, world!")
-> "uryyb, jbeyq!"
(sort-group sequence [keyfun [lessfun]])
The sort-group function sorts sequence according to the keyfun and lessfun arguments, and then breaks the resulting sequence into groups, based on the equivalence of the elements under keyfun.
The csort-group differs from sort-group in that it is based on the caching csort rather than sort.
The following equivalence holds:
(sort-group sq lf kf)
<-->
(partition-by kf (sort sq kf lf))
Note the reversed order of keyfun and lessfun arguments between sort and sort-group.
(uniq sequence)
The uniq function returns a sequence of the same kind as sequence, but with duplicates removed. Elements of sequence are considered equal under the equal function. The first occurrence of each element is retained, and the subsequent duplicates of that element, of any, are suppressed, such that the order of the elements is otherwise preserved.
The uniq function is an alias for the one-argument case of unique. That is to say, this equivalence holds:
(unique sequence [keyfun {hash-arg}*])
The unique function is a generalization of uniq. It returns a sequence of the same kind as sequence, but with duplicates removed.
If neither keyfun nor hash-args are specified, then elements of sequence are considered equal under the equal function. The first occurrence of each element is retained, and the subsequent duplicates of that element, of any, are suppressed, such that the order of the elements is otherwise preserved.
If keyfun is specified, then that function is applied to each element, and the resulting values are compared for equality. In other words, the behavior is as if keyfun were the identity function.
If one or more hash-args are present, these specify the arguments for the construction of the internal hash table used by unique. The arguments are like those of the hash function.
(tuples length sequence [fill-value])
The tuples function produces a lazy list which represents a reorganization of the elements of sequence into tuples of length, where length must be a positive integer.
The length of the sequence might not be evenly divisible by the tuple length. In this case, if a fill-value argument is specified, then the last tuple is padded with enough repetitions of fill-value to make it have length elements. If fill-value is not specified, then the last tuple is left shorter than length.
The output of the function is a list, but the tuples themselves are sequences of the same kind as sequence. If sequence is any kind of list, they are lists, and not lazy lists.
(tuples 3 #(1 2 3 4 5 6 7 8) 0) -> (#(1 2 3) #(4 5 6)
#(7 8 0))
(tuples 3 "abc") -> ("abc")
(tuples 3 "abcd") -> ("abc" "d")
(tuples 3 "abcd" #\z) -> ("abc" "dzz")
(tuples 3 (list 1 2) #\z) -> ((1 2 #\z))
(tuples* length sequence [fill-value])
The tuples* function produces a lazy list of overlapping tuples taken from sequence. The length of the tuples is given by the length argument.
The length argument must be a positive integer.
Tuples are subsequences of consecutive items from the input sequence, beginning with consecutive elements. The first tuple in the returned list begins with the first item of sequence; the second tuple begins with the second item, and so forth.
The output of the function is a list, but the tuples themselves are sequences of the same kind as sequence. If sequence is any kind of list, they are lists, and not lazy lists.
If sequence is shorter than length then it contains no tuples of that length. In this case, if no fill-value argument is specified, then the empty list is returned. In this same situation, if fill-value is specified, then a one-element list is returned, consisting of a tuple of the required length, consisting of the elements from sequence followed by repetitions of fill-value, which must be of a type suitable as an element of the sequence. The fill-value is otherwise ignored.
(tuples* 1 "abc") -> ("a" "b" "c")
(tuples* 2 "abc") -> ("ab" "bc")
(tuples* 3 "abc") -> ("abc")
(tuples* 4 "abc") -> nil
(tuples* 4 "abc" #\z) -> ("abcz")
(tuples* 6 "abc" #\z) -> ("abczzz")
(tuples* 6 "abc" 4) -> error
(tuples* 2 '(a b c)) -> ((a b) (b c))
(take 3 (tuples* 3 0)) -> ((0 1 2) (1 2 3) (2 3 4))
(partition-by function sequence)
If sequence is empty, then partition-by returns an empty list, and function is never called.
Otherwise, partition-by returns a lazy list of partitions of the sequence sequence. Partitions are consecutive, nonempty substrings of sequence, of the same kind as sequence.
The partitioning begins with the first element of sequence being placed into a partition.
The subsequent partitioning is done according to function, which is applied to each element of sequence. Whenever, for the next element, the function returns the same value as it returned for the previous element, the element is placed into the same partition. Otherwise, the next element is placed into, and begins, a new partition.
The return values of the calls to function are compared using the equal function.
[partition-by identity '(1 2 3 3 4 4 4 5)] -> ((1) (2) (3 3)
(4 4 4) (5))
(partition-by (op = 3) #(1 2 3 4 5 6 7)) -> (#(1 2) #(3)
#(4 5 6 7))
(partition-if function iterable [count])
The partition-if function separates the iterable sequence into partitions which are identified by the two-argument function. The principal idea is that successive overlapping pairs from iterable are passed as arguments to function, and whenever function yields true, those elements are identified as belonging to separate partitions: a partitioning division shall take place between them. The detailed semantics is given below, as a procedure.
Firstly, if sequence is empty, then partition-if returns an empty list, and function is never called.
Otherwise, partition-if returns a lazy list of partitions of iterable. Partitions are consecutive, nonempty substrings of iterable, of the same kind as iterable.
The partitioning begins with the first element of iterable being placed into the first partition.
The subsequent partitioning is done according to a Boolean function, which must accept two arguments. Whenever the function yields true, it indicates that a partition is to be terminated and a new partition to begin.
The count argument, if present, must be a nonnegative integer. It indicates a limit on how many partitions will be delimited; after this limit is reached, the remainder of the iterable sequence is placed into a single partition.
After the first element is placed into a partition, the following partition-building process is repeated until the partition is terminated.
If, after a partition is thus produced, a next element is available, it is placed into a new partition, and the above partition-building process takes place from step 1. Otherwise, the lazy list terminates.
;; Start new partition for unequal characters.
[partition-if neql "aaaabbcdee"] -> ("aaaa" "bb" "c" "d" "ee")
;; As above, but partition only twice
[partition-if neql "aaaabbcdee" 2] -> ("aaaa" "bb" "cdee")
;; Start new partition when non-digit follows digit:
[partition-if (do and
(chr-isdigit @1)
(not (chr-isdigit @2)))
"a13cd9foo42z"]
-> ("a13" "cd9" "foo42" "z")
;; Place ascending runs of consecutive integers
;; into partitions. I.e. start a partition whenever the
;; difference from the previous element isn't 1:
(partition-if (op /= (- @2 @1) 1)
'(1 3 4 5 7 8 9 10 9 8 6 5 3 2))
-> ((1) (3 4 5) (7 8 9 10) (9) (8) (6) (5) (3) (2))
;; Place runs of adjacent integers into partitions.
;; I.e. start a new partition if the the absolute value of
;; the difference from the previous exceeds 1:
(partition-if (op > (abs (- @2 @1)) 1)
'(1 3 4 5 7 8 9 10 9 8 6 5 3 2))
-> ((1) (3 4 5) (7 8 9 10 9 8) (6 5) (3 2))
Functions in this category perform efficient traversal of sequences.
There are two flavors of these functions: functions in the iter-begin group, and functions in the seq-begin group. The latter are obsolescent.
User-defined iteration is possible via defining special methods on structures. An object supports iteration by defining the special method iter-begin which is different from the iter-begin function. This special function returns an iterator object which supports special methods iter-item, iter-more and iter-step. Two protocols are supported, one of which is more efficient by eliminating the iter-more method. Details are specified in the section Special Structure Functions.
(iter-begin seq)
The iter-begin function returns an iterator object suitable for traversing the elements of the sequence denoted by the seq object.
If seq is a list-like sequence, then iter-begin may return seq itself as the iterator. Likewise if seq is a number.
If seq is a structure which supports the iter-begin method, then that method is called and its return value is returned. A structure which does not support this method is possibly considered to be a sequence according to the usual criteria, based on whether it supports the nullify, length or car methods. A struct object supporting none of these methods is deemed not iterable.
Otherwise, if seq is an iterator object of seq-iter type, such as one produced by iter-begin, then an iterator similar to that iterator is returned, as if produced by applying the copy-iter function to seq.
In all other cases, if seq is iterable, an object of type seq-iter is returned.
Range objects are iterable if they are numeric. A range consisting of two strings may also be iterable, as described below.
A range is considered to be a numeric or character range if the from element is a number or character. The to is then required to be either a value which is comparable with that number or character using the < function, or else it must be one of the two objects t or :, either of which indicate that the range is unbounded. In this unbounded range case, the expressions (iter-begin X..:) and (iter-begin X..t) are equivalent to (iter-begin X). Numeric ranges are half-open: the to value of ascending ranges is excluded, as is the from value of descending ranges, so that 0..10 steps through the values 0 through 9, and 10..0 steps through the same values in reverse order.
A string range consists of two strings of equal length. If the strings are of unequal length, an error exception is thrown. The sequence denoted by a string range is a sequence of strings formed from the the Cartesian product of the character ranges formed by positionally-corresponding characters from the two strings. The order of the sequence is such that the rightmost character varies most frequently. In more detail, the string range iterates over successive strings by incrementing or decrementing the characters of the from string until they are equal to those of the to string. The rightmost character has priority. For instance, the range "AA".."CC" iterates over the strings AA, AB, AC, BA, BB, BC, CA, CB and CC. The descending range "CC".."AA" iterates over the same strings, in reverse order. Whenever the incrementing character attains the value of the corresponding character in the to string, that character is reset to its starting value, and its left neighbor, if it exists, is incremented instead. If no left neighbor exists, the iteration terminates. For every character position in the string pair, it is independently determined whether the iteration for that position is ascending or descending, such that the range "AC".."CA" iterates over the strings AC, AB, AA, BC, BB, BA, CC, CB and CA.
Search trees are iterable. Iteration entails an in-order visits of the elements of a tree. A tree iterator created by tree-begin is also iterable. It is unspecified whether iteration over a tree-iter object modifies that object to perform the traversal, or whether it uses a copy of the iterator.
If seq is not an iterable object, an error exception is thrown.
(iter-more iter)
The iter-more function returns t if there remain more elements to be traversed. Otherwise it returns nil.
The iter argument must be a valid iterator returned by a call to iter-begin, iter-step or iter-reset.
The iter-more function doesn't change the state of iter.
If iter is the object nil then nil is returned. Note: the iter-begin may return nil if its argument is nil or any empty sequence, or an empty range (a range whose to and from fields are the same number or character).
If iter is a cons cell, then iter-more returns t.
If iter is a number, then iter-more returns t. This is the case even if calculating the successor of that number isn't possible due to floating-point overflow or insufficient system resources.
If iter is a character, then iter-more returns t if iter isn't the highest possible character code, otherwise nil.
If iter was formed from a descending range, meaning that iter-begin was invoked on a range with a from fielding exceeding its to value, then iter-begin returns true while the current iterator value is greater than the the limiting value given by the to field. For an ascending range, it returns true if the current iterator value is lower than the limiting value. However, note the peculiar semantics of iter-item with regard to descending range iteration.
If iter is a structure, then if it supports an iter-more method, then that method is called with no arguments, and its return value is returned. If the structure does not have an iter-more method, then t is returned.
(iter-item iter)
If the iter-more function indicates that more items remain to be visited, then the next item can be retrieved using iter-item.
The iter argument must be a valid iterator returned by a call to iter-begin, iter-step or iter-reset.
The iter-more function doesn't change the state of iter.
If iter-more is invoked on an iterator which indicates that no more items remain to be visited, the return value is nil.
If iter is a cons cell, then iter-item returns the car field of that cell.
If iter is a character or number, then iter-item returns that character or number itself.
If iter is based on an ascending numeric or character range, then iter-item returns the current iteration value, which is initialized by iter-begin as a copy of the range's from field. Thus, the range 0..3 traverses the values 0, 1 and 2, excluding the 3.
If iter is based on a descending numeric or character range, then iter-item returns the predecessor of the current iteration value, which is initialized iter-begin as a copy of the range's from field. Thus, the range 3..0 traverses the values 2, 1 and 0, excluding the 3: exactly the same values are visited as for the range 0..3 only in reverse order.
If iter is a structure which supports the iter-item method, then that method is called and its return value is returned.
(iter-step iter)
If the iter-more function indicates that more items remain to be visited, then the iter-step function may be used to consume the next item.
The function returns an iterator denoting the traversal of the remaining items in the sequence.
The iter argument must be a valid iterator returned by a call to iter-begin, iter-step or iter-reset.
The iter-step function may return a new object, in which case it avoids changing the state of iter, or else it may change the state of iter and return it.
If the application discontinues the use of iter, and continues the traversal using the returned iterator, it will work correctly in either situation.
If iter-step is invoked on an iterator which indicates that no more items remain to be visited, the return value is unspecified.
If iter is a cons cell, then iter-step returns the cdr field of that cell. That value must itself be a cons or else nil, otherwise an error is thrown. This is to prevent iteration from wrongly iterating into the non-null terminators of improper lists. Without this rule, iteration of a list like (1 2 . 3) would reach the cons cell (2 . 3) at which point a subsequent iter-step would return the cdr field 3. But that value is a valid iterator which will then continue by stepping through 4, 5 and so on.
If iter is a list-like sequence, then cdr is invoked on it and that value is returned. The value must also be a list-like sequence, or else nil. The reasoning for this is the same as for the similar restriction imposed in the case when iter is a cons.
If iter is a character or number, then iter-step returns its successor, as if using the succ function.
If iter is a structure which supports the iter-step method, then that method is called and its return value is returned.
(iter-reset iter seq)
The iter-reset function returns an iterator object specialized for the task of traversing the sequence seq.
If it is possible for iter to be that object, then the function may adjust the state of iter and return it.
If iter-reset doesn't use iter, then it behaves exactly like iter-begin being invoked on seq.
If seq is a structure which supports the iter-reset method, then that method is called and its return value is returned. Note the reversed arguments. The iter-reset method is of the seq object, not of iter. That is to say, the call (iter-reset iter obj) results in the obj.(iter-reset iter) call. If seq is a structure which doesn't support iter-reset then iter is ignored, iter-begin is invoked on seq and the result is returned.
(copy-iter iter)
The copy-iter produces a duplicate iter such that the duplicate iterator will traverse the same sequence of items as iter starting at the current point in the sequence indicated by iter.
For some kinds of iterators, such as integers and conses, copy-iter just returns iter.
If iter is a structure object, then if it supports the copy method, that method is invoked and its return value is taken as the iterator copy. Otherwise, iter must implement a list-like sequence, in which case the object is just returned. If iter is a structure which neither supports a copy method nor implements a list-like sequence by supporting the car method, an error exception is thrown.
Note: iterators of type seq-iter can be copied with the copy function (which for those objects is defined in terms of copy-iter). However, the copy function has the wrong semantics for other kinds of iterator objects. It refuses to copy certain atoms such as numbers, and in the case of conses it behaves like copy-list, which is unnecessary.
(iter-cat seq*)
The iter-cat function produces a catenated iterator: an object suitable for traversing the abstract sequence formed by the catenation of the seq arguments. This is accomplished without actually catenating the argument sequences.
If no arguments are given to iter-cat then it returns nil. Otherwise, the abstract semantics of the catenated iterator is as follows. The iterator retains all of the seq objects. It converts the first seq object to an iterator as if by iter-begin on it. This is referred to as the individual iterator. When that iterator is exhausted of items, iter-begin is called on the next seq object to produce the next individual iterator.
Note: under this semantics, the catenated iterator's iter-more operation does not simply report the value returned by the iter-more call on the individual iterator. When the individual iterator's iter-more function returns nil, the catenated operator then switches to the individual iterator of the next seq object in the argument sequence. This is repeated as many times as necessary until an iterator is found for which The iter-item function, when applied to a catenated iterator, similarly potentially searches through the argument space. iter-more yields true, or the arguments are exhausted.
;; Create an iterator that produces 0, 1, ... 9, 20, 21, ... 29
(iter-cat 0..10 20..30)
(seq-begin object)
The obsolescent seq-begin function returns an iterator object specialized to the task of traversing the sequence represented by the input object.
If object isn't a sequence, an exception is thrown.
Note that if object is a lazy list, the returned iterator maintains a reference to the head of that list during the traversal; therefore, generic iteration based on iterators from seq-begin is not suitable for indefinite iteration over infinite lists.
(seq-next iter end-value)
The obsolescent seq-next function retrieves the next available item from the sequence iterated by iter, which must be an object returned by seq-begin.
If the sequence has no more items to be traversed, then end-value is returned instead.
Note: to avoid ambiguities, the application should provide an end-value which is guaranteed distinct from any item in the sequence, such as a freshly allocated object.
(seq-reset iter object)
The obsolescent seq-reset reinitializes the existing iterator object iter to begin a new traversal over the given object, which must be a value of a kind that would be a suitable argument for seq-begin.
The seq-reset function returns iter.
TXR Lisp provides an a structure type called list-builder which encapsulates state and methods for constructing lists procedurally. Among the advantages of using list-builder is that lists can be constructed in the left-to-right direction without requiring multiple traversals or reversal. For example, list-builder naturally combines with iteration or recursion: items visited in an iterative or recursive process can be collected easily using list-builder in the order they are visited.
The list-builder type provides methods for adding and removing items at either end of the list, making it suitable where a dequeue structure is required.
The basic workflow begins with the instantiation of a list-builder object. This object may be initialized with a piece of list material which begins the to-be-constructed list, or it may be initialized to begin with an empty list. Methods such as add and pend are invoked on this object to extend the list with new elements. At any point, the list constructed so far is available using the get method, which is also how the final version of the list is eventually retrieved.
The list-builder methods which add material to the list all return the list builder, making chaining possible.
(new list-builder).(add 1).(add 2).(pend '(3 4 5)).(get)
-> (1 2 3 4 5)
The build macro is provided which syntactically streamlines the process. It implicitly creates a list-builder instance and binds it to a hidden lexical variable. It then evaluates forms in a lexical scope in which shorthand macros are available for building the list.
(defstruct list-builder nil
head tail)
The list-builder structure encapsulates the state for a list building process. Programs should use the build-list function for creating an instance of list-builder. The head and tail slots should be regarded as internal variables.
(build-list [initial-list])
The build-list function instantiates and returns an object of struct type list-builder.
If no initial-list argument is supplied, then the object is implicitly initialized with an empty list.
If the argument is supplied, then it is equivalent to calling build-list without an argument to produce an object obj by invoking the method call obj.(ncon initial-list) on this object. The object produced by the expression list is installed (without being copied) into the object as the prefix of the list to be constructed.
The initial-list argument can be a sequence other than a list.
;; build the list (a b) trivially
(let ((lb (build-list '(a b))))
lb.(get)
-> (a b)
list-builder.(add element*)
list-builder.(add* element*)
The add and add* methods extend the list being constructed by a list-builder object by adding individual elements to it. The add method adds elements at the tail of the list, whereas add* adds elements at the front.
These methods return the list-builder object.
The precise semantics is as follows. All of the element arguments are combined into a list as if by the list function, and the resulting list combined with the current contents of the list-builder object as if using the append function. The resulting list becomes the new contents.
;; Build the list (1 2 3 4)
(let ((lb (build-list)))
lb.(add 3 4)
lb.(add* 1 2)
lb.(get))
-> (1 2 3 4)
;; Add "c" to "abc"
;; same semantics as (append "abc" #\c)
(let ((lb (build-list "ab")))
lb.(add #\c)
lb.(get))
-> "abc"
list-builder.(pend list*)
list-builder.(pend* list*)
The pend and pend* methods extend the list being constructed by a list-builder object by adding lists to it. The pend method catenates the list arguments together as if by the append function, then appends the resulting list to the end of the list being constructed. The pend* method is similar, except it prepends the catenated lists to the front of the list being constructed.
The pend and pend* operations do not mutate the input lists, but may cause the resulting list to share structure with the input lists.
These functions may mutate the list already contained in list-builder; however, they avoid mutating those parts of the current list that are shared with inputs that were given in earlier calls to these functions.
These methods return the list-builder object.
;; Build the list (1 2 3 4)
(let ((lb (build-list)))
lb.(pend '(3 4))
lb.(pend* '(1 2))
lb.(get))
-> (1 2 3 4)
list-builder.(ncon list*)
list-builder.(ncon* list*)
The ncon and ncon* methods extend the list being constructed by a list-builder object by adding lists to it. The ncon method destructively catenates the list arguments as if by the nconc function. The resulting list is appended to the list being constructed. The ncon* method is similar, except it prepends the catenated lists to the front of the list being constructed.
These methods may destructively manipulate the list already contained in the list-builder object, and likewise may destructively manipulate the input lists. They may cause the list being constructed to share substructure with the input lists.
Additionally, these methods may destructively manipulate the list already contained in the list-builder object without regard for shared structure between that list and inputs given earlier any of the pend, pend*, ncon or ncon* functions.
The ncon* function can be called with a single argument which is an atom. This atom will simply be installed as the terminating atom of the list being constructed, if the current list is an ordinary list.
These methods return the list-builder object.
;; Build the list (1 2 3 4 . 5)
(let ((lb (build-list)))
lb.(ncon* (list 1 2))
lb.(ncon (list 3 4))
lb.(ncon 5)
lb.(get))
-> (1 2 3 4 . 5)
list-builder.(oust list*)
The oust method discards the list constructed so far, optionally replacing it with new material.
The oust method catenates the list arguments together as if by the append function. The resulting list, which is empty if there are no list arguments, then replaces the object's list constructed so far.
The oust method returns the list-builder object.
;; Build the list (3 4) by first building (1 2),
;; then discarding that and adding 3 and 4:
(let ((lb (build-list)))
lb.(add 1 2)
lb.(oust)
lb.(add 3 4)
lb.(get))
-> (3 4)
;; Build the list (3 4 5 6) by first building (1 2),
;; then replacing with catenation of (3 4) and (5 6):
(let ((lb (build-list)))
lb.(pend '(1 2))
lb.(oust '(3 4) '(5 6))
lb.(get))
-> (3 4 5 6)
list-builder.(get)
The get method retrieves the list constructed so far by a list-builder object. It doesn't change the state of the object. The retrieved list may be passed as an argument into the construction methods on the same object.
;; Build the circular list (1 1 1 1 ...)
;; by appending (1) to itself destructively:
(let ((lb (build-list '(1))))
lb.(ncon* lb.(get))
lb.(get))
-> (1 1 1 1 ...)
;; build the list (1 2 1 2 1 2 1 2)
;; by doubling (1 2) twice:
(let ((lb (build-list)))
lb.(add 1 2)
lb.(pend lb.(get))
lb.(pend lb.(get))
lb.(get))
-> (1 2 1 2 1 2 1 2)
list-builder.(del)
list-builder.(del*)
The del and del* methods each remove an element from the list and return it. If the list is empty, they return nil.
The del method removes an element from the front of the list, whereas del* removes an element from the end of the list.
Note: this orientation is opposite to add and add*. Thus del pairs with add to produce FIFO queuing behavior.
(build form*)
(buildn form*)
The build and buildn macros provide a shorthand notation for constructing lists using the list-builder structure. They eliminate the explicit call to the build-list function to construct the object, and eliminate the explicit references to the object.
Both of these macros create a lexical environment in which a list-builder object is implicitly constructed and bound to a hidden variable. This lexical environment also provides local functions named add, add*, pend, pend*, ncon, ncon*, oust, get, del and del*, which mimic the list-builder methods, but operate implicitly on this hidden variable, so that the object need not be mentioned as an argument. With the exception of get, del and del*, the local functions return nil, unlike like the same-named list-builder methods, which return the list-builder object.
In this lexical environment, each form is evaluated in order.
When the last form is evaluated, build returns the constructed list, whereas buildn returns the value of the last form.
If no forms are enclosed, both macros return nil.
Note: because the local function del has the same name as a global macro, it is implemented as a macrolet. Inside a build or buildn, if del is invoked with no arguments, then it denotes a call to the list-builder del method. If invoked with an argument, then it resolves to the global del macro for deleting a place.
;; Build the circular list (1 1 1 1 ...)
;; by appending (1) to itself destructively:
(build
(add 1)
(ncon* (get))) -> (1 1 1 1 ...)
;; build the list (1 2 1 2 1 2 1 2)
;; by doubling (1 2) twice:
(build
(add 1 2)
(pend (get))
(pend (get))) -> (1 2 1 2 1 2 1 2)
;; build a list by mapping over the local
;; add function:
(build [mapdo add (range 1 3)]) -> (1 2 3)
;; breadth-first traversal of nested list;
(defun bf-map (tree visit-fn)
(buildn
(add tree)
(whilet ((item (del)))
(if (atom item)
[visit-fn item]
(each ((el item))
(add el))))))
(let (flat)
(bf-map '(1 (2 (3 4 (5))) ((6 7) 8)) (do push @1 flat))
(nreverse flat))
-> (1 2 8 3 4 6 7 5)
(perm seq [len])
(permi seq [len])
The perm function returns a lazy list which consists of all length len permutations of formed by items taken from seq. The permutations do not use any element of seq more than once.
The permi function has identical argument semantics, but returns an iterator instead of a lazy list: an object meeting the same conventions as the return value of iter-begin.
Argument len, if present, must be a positive integer, and seq must be a sequence.
If len is not present, then its value defaults to the length of seq: the list of the full permutations of the entire sequence is returned.
The permutations in the returned list are sequences of the same kind as seq.
If len is zero, then a list containing one permutation is returned, and that permutation is of zero length.
If len exceeds the length of seq, then an empty list is returned, since it is impossible to make a single nonrepeating permutation that requires more items than are available.
The permutations are lexicographically ordered.
(rperm seq len)
(rpermi seq len)
The rperm function returns a lazy list which consists of all the repeating permutations of length len formed by items taken from seq. "Repeating" means that the items from seq can appear more than once in the permutations.
The rpermi function has identical argument semantics, but returns an iterator instead of a lazy list: an object meeting the same conventions as the return value of iter-begin.
The permutations which are returned are sequences of the same kind as seq.
Argument len must be a nonnegative integer, and seq must be a sequence.
If len is zero, then a single permutation is returned, of zero length. This is true regardless of whether seq is itself empty.
If seq is empty and len is greater than zero, then no permutations are returned, since permutations of a positive length require items, and the sequence has no items. Thus there exist no such permutations.
The first permutation consists of le repetitions of the first element of seq. The next repetition, if there is one, differs from the first repetition in that its last element is the second element of seq. That is to say, the permutations are lexicographically ordered.
(rperm "01" 3) -> ("000" "001" "010" "011"
"100" "101" "110" "111")
(rperm #(1) 3) -> (#(1 1 1))
(rperm '(0 1 2) 2) -> ((0 0) (0 1) (0 2) (1 0)
(1 1) (1 2) (2 0) (2 1) (2 2))
(comb seq len)
(combi seq len)
The comb function returns a lazy list which consists of all length len nonrepeating combinations formed by taking items taken from seq. "Nonrepeating combinations" means that the combinations do not use any element of seq more than once. If seq contains no duplicates, then the combinations contain no duplicates.
The combi function has identical argument semantics, but returns an iterator instead of a lazy list: an object meeting the same conventions as the return value of iter-begin.
Argument len must be a nonnegative integer, and seq must be a sequence or a hash table.
The combinations in the returned list are objects of the same kind as seq.
If len is zero, then a list containing one combination is returned, and that combination is of zero length.
If len exceeds the number of elements in seq, then an empty list is returned, since it is impossible to make a single nonrepeating combination that requires more items than are available.
If seq is a sequence, the returned combinations are lexicographically ordered. This requirement is not applicable when seq is a hash table.
;; powerset function, in terms of comb.
;; Yields a lazy list of all subsets of s,
;; expressed as sequences of the same type as s.
(defun powerset (s)
(mappend* (op comb s) (range 0 (length s))))
(rcomb seq len)
(rcombi seq len)
The rcomb function returns a lazy list which consists of all length len repeating combinations formed by taking items taken from seq. "Repeating combinations" means that the combinations can use an element of seq more than once.
The rcombi function has identical argument semantics, but returns an iterator instead of a lazy list: an object meeting the same conventions as the return value of iter-begin.
Argument len must be a nonnegative integer, and seq must be a sequence.
The combinations in the returned list are sequences of the same kind as seq.
If len is zero, then a list containing one combination is returned, and that combination is of zero length. This is true even if seq is empty.
If seq is empty, and len is nonzero, then an empty list is returned.
The combinations are lexicographically ordered.
Because TXR Lisp supports structural macros, TXR processes TXR Lisp expressions in two separate phases: the expansion phase and the evaluation/compilation phase. During the expansion phase, a top-level expression is recursively traversed, and all macro invocations in it are expanded. The result is a transformed expression which contains only function calls and invocations of special operators. This expanded form is then evaluated or compiled, depending on the situation.
Macro invocations are compound forms and whose operator symbol has a macro definition in scope. A macro definition is a kind of function which operates on syntax during macro-expansion, called upon to calculate a transformation of the syntax. The return value of a macro replaces its invocation, and is traversed to look for more opportunities for macro expansion. Macros differ from ordinary functions in three ways: they are called at macro-expansion time, they receive pieces of unevaluated syntax as their arguments, and their parameter lists are macro parameter lists which support destructuring, as well as certain special parameters.
TXR Lisp also supports symbol macros. A symbol macro definition associates a symbol with an expansion. When that symbol appears as a form, the macro-expander replaces it with the expansion.
TXR source files are treated somewhat differently with regard to macro expansion compared to TXR Lisp. When TXR Lisp forms are read from a file by load or compile or read by the interactive listener, each form is expanded and evaluated or compiled before the subsequent form is processed. In contrast, when a TXR file is loaded, expansion of the Lisp forms are its arguments takes place during the parsing of the entire source file, and is complete for the entire file before any of the code is executed.
TXR macros support destructuring, similarly to Common Lisp macros. This means that macro parameter lists are like function argument lists, but support nesting. A macro parameter list can specify a nested parameter list in every place where an argument symbol may appear. For instance, consider this macro parameter list:
((a (b c)) : (c frm) ((d e) frm2 de-p) . g)
The top-level of this nested form has the structure
(I : J K . L)
in which we can identify the major constituent positions as I, J, K and L.
The constituent at position I is the mandatory parameter (a (b c)). Position J holds the optional parameter c (with default init form frm). At K is found the optional parameter (d e) (with default init form frm2 and presence-indicating variable de-p). Finally, the parameter in the dot position L is g, which captures trailing arguments.
Obviously, some of the parameters are compound expressions rather than symbols: (a (b c)) and (d e). These compounds express nested macro parameter lists.
Starting in TXR 285, the symbol t can be used in a macro parameter list in place of a parameter name. This indicates that an object is expected at that position in the corresponding structure, but no variable will be bound. For completeness, the t symbol may also be used for a presence-indicating variable. When the name of an optional parameter is specified as t, and the corresponding structure is missing, the init-val expression, if present, is still evaluated under the same circumstances as it would if a variable were present.
Nested macro parameter lists recursively match the corresponding structure in the argument object. For instance if a simple argument would capture the structure (1 (2 3)) then we can replace the argument with the nested argument list (a (b c)) which destructures the (1 (2 3)) such that the parameters a, b and c will end up bound to 1, 2 and 3, respectively.
Nested macro parameter lists have all the features of the top-level macro parameter lists: they can have optional arguments with default values, use the dotted position, and contain the :env, :whole and :form special parameters, which are described below. In nested parameter lists, the binding strictness is relaxed for optional parameters. If (a (b c)) is optional, and the argument is, say, (1), then a gets 1, and b and c receive nil.
Macro parameter lists also supports three special keywords, namely :env, :whole and :form.
The parameter list (:whole x :env y :form z) will bind parameter x to the entire macro parameter list, bind parameter y to the macro environment and bind parameter z to the entire macro form (the original compound form used to invoke the macro).
The :env, :whole and :form notations can occur anywhere in a macro parameter list, other than to the right of the consing dot. They can be used in nested macro parameter lists also. Note that in a nested macro parameter list, :form and :env do not change meaning: they bind the same object as they would in the top-level of the macro parameter list. However the :whole parameter inside has a restricted scope in a nested parameter list: its parameter will capture just that part of the argument material which matches that parameter list, rather than the entire argument list.
The processing of macro parameter lists omits the feature that when the : (colon) keyword symbol is given as the argument to an optional parameter, that argument is treated as a missing argument. This special logic is implemented only in the function argument passing mechanism, not in the binding of macro parameters to object structure. If the colon symbol appears in the object structure and is matched against an optional parameter, it is an ordinary value. That parameter is considered present, and takes on the colon symbol as its value.
In ANSI Common Lisp, the lambda list keyword &whole binds its corresponding variable to the entire macro form, whereas TXR Lisp's :whole binds its variable only to the arguments of the macro form.
Note, however, that ANSI CL distinguishes between destructuring and macro lambda lists, and the &whole parameter has a different behavior in each. Under destructuring-bind, the &whole parameter receives just the arguments, just like the behavior of TXR Lisp's :whole parameter.
TXR Lisp does not distinguish between destructuring and macro lambda lists; they are the same and behave the same way. Thus :whole is treated the same way in macros as in tree-bind and related binding operators: it binds just the arguments to the parameter. TXR Lisp has the special parameter :form by means of which macros can access their invoking form. This parameter is also supported in tree-bind and binds to the entire tree-bind form.
ANSI CL doesn't support the convention that the t symbol may appear instead of a parameter symbol to suppress the binding of a variable.
The following description omits the treatment of top-level forms by eval and the compiler. This is described, respectively, in the description of eval and the section Top-Level Forms inside the LISP COMPILATION chapter. Certain other details are also omitted, such as the dynamic evolution of the macro-time environment, the expansion of macrolet forms.
Macro expansion is, generally speaking, a recursive process. The expression to be expanded is classified into cases, and as necessary, the constituent expressions are recursively expanded, depending on these cases. Certain aspects of the process may be regarded as iterative. Macro expansion maintains a macro-time lexical environment which is extended and contracted as the expander descends into various nested binding constructs.
The expander may encounter a bindable symbol. If such a symbol has a binding as a symbol macro, then it is replaced by its expansion, and the expander iterates on the resulting form. The form may be another object, including a symbol. If it is the same symbol, than macro expansion terminates; the symbol remains unsubstituted. Symbols are treated differently by the expander if they are in the Lisp-1-style context of the dwim operator, or the equivalent square bracket notation. The expander takes into consideration the semantics of the combined function and variable namespace.
The expander may encounter a compound form headed by a symbol which has a macro binding. In this situation, the macro expander function is called, and the form is replaced by the resulting form. That form is considered again as a potential macro. In any case, the expander makes a note that it has expanded a macro,
If a form isn't a macro, then it's either a function call, special from or an atomic form: a symbol (that has no binding as a symbol macro) or other atom. The interesting cases are special forms and function calls, since the atomic forms are simply returned as-is without expansion. Special forms and function call forms contain other forms, some or all of which require expansion. The expander recognizes the shape of each special form or function call, pulls out the constituent expressions and expands them recursively, combining the results into a new version of the special form or function call form.
Because TXR Lisp allows the same symbol to have a macro and function binding, the expander allows for interplay between the two, which produces useful behaviors. Recall from two paragraphs ago that whenever the expander expands a macro, it makes a note that it has done so. Subsequently, suppose that the rounds of macro expansion happen to terminate in such a way that the result is a function call form. The form's constituents are expanded, If the expansion of those constituents produces any change, then the resulting replacement function call form is again examined for the possibility that it may be a macro. This special requirement, not typically implemented by Lisp macro expanders, greatly simplifies the writing of macros which provide algebraic optimizations of function calls.
An example follows to illustrate the benefit of the rule. Note that the example involves some simple macros which change the number of times that an argument expression is evaluated. A more careful handling of this issue is omitted in order to keep the examples simple.
Suppose a macro is written for the sqrt function like this:
(defmacro sqrt (:match :form f)
(((* @exp @exp)) exp)
(@else f))
The macro uses pattern matching to recognize cases like (sqrt (* a a)) when the argument is a product expression with two identical terms. This pattern implements the arithmetic identity that the positive square root of a real term multiplied by itself is just that term.
Now suppose that a similar macro is written to optimize a certain case of the expt function:
(defmacro expt (:match :form f)
((@exp 2) ^(* ,exp ,exp))
(@else f))
This macro recognizes when the argument is being squared, turning (expt x 2) into (* x x): a strength reduction from exponentiation to multiplication.
What if the following expression is then written:
The special provision in the expander algorithm allows the above combination to reduce to just x, as follows. Firstly, the (sqrt (expt x 2)) expression is treated as a macro call. It doesn't match the main case in the macro, only the fallback case which returns the form unexpanded. The expander notes that it has invoked a macro, and then proceeds to treat the form as a function call. The function call's argument expression (expt x 2) is expanded as a macro. This produces a transformation: our expt macro reduces this quadratic term to (* x x). Here is where the special rule comes into play. The expander sees that the function's arguments have been transformed. It knows that the original function call was the result of expansion. To promote more opportunities for expansion, it tries the transformed function call again as a macro. The (sqrt (* x x)) form is handed to the sqrt macro, which this time has a match for the (* x x) argument pattern, reducing the entire form to x. Effectively, the sqrt macro has the opportunity to work with both the unexpanded argument syntax (expt x 2) as well as its expanded version. It is first offered the one, and when it declines to expand, then the other.
(defmacro name
(param* [: opt-param* ] [. rest-param ])
body-form*)
The defmacro operator is evaluated at expansion time. It defines a macro-expander function under the name name, effectively creating a new operator.
Note that the above syntax synopsis describes only the canonical parameter syntax which remains after parameter list macros are expanded. See the section Parameter List Macros.
Note that the parameter list is a macro parameter list, and not a function parameter list. This means that each param and opt-param can be not only a symbol, but it can itself be a parameter list. The corresponding argument is then treated as a structure which matches that parameter list. This nesting of parameter lists can be carried to an arbitrary depth.
A macro is called like any other operator, and resembles a function. Unlike in a function call, the macro receives the argument expressions themselves, rather than their values. Therefore it operates on syntax rather than on values. Also, unlike a function call, a macro call occurs in the expansion phase, rather than the evaluation phase.
The return value of the macro is the macro expansion. It is substituted in place of the entire macro call form. That form is then expanded again; it may itself be another macro call, or contain more macro calls.
A global macro defined using defmacro may decline to expand a macro form. Declining to expand is achieved by returning the original unexpanded form, which may be captured using the :form parameter. When a global macro declines to expand a form, the form is taken as-is. At evaluation time, it will be treated as a function call. Note: when a local macro defined by macrolet declines, more complicated requirements apply; see the description of macrolet.
A macro in the global namespace introduced by defmacro may coexist with a function of the same name introduced by defun. This is not permitted in ANSI Common Lisp.
ANSI Common Lisp doesn't describe the concept of declining to expand, except in the area of compiler macros. Since TXR Lisp allows global macros and functions of the same name to coexist, ordinary macros can be used to optimize functions in a manner similar to Common Lisp compiler macros. A macro can be written of the same name as a function, and can optimize certain cases of the function call by expanding them to some alternative syntax. Cases which it doesn't optimize are handled by declining to expand, in which case the form remains as the original function call.
;; dolist macro similar to Common Lisp's:
;;
;; The following will print 1, 2 and 3
;; on separate lines:
;; and return 42.
;;
;; (dolist (x '(1 2 3) 42)
;; (format t "~s\n" x))
(defmacro dolist ((var list : result) . body)
(let ((i (gensym)))
^(for ((,i ,list)) (,i ,result) ((set ,i (cdr ,i)))
(let ((,var (car ,i)))
,*body))))
(macrolet ({(name macro-style-params
macro-body-form*)}*)
body-form*)
The macrolet binding operator extends the macro-time lexical environment by making zero or more new local macros visible.
The macrolet symbol is followed by a list of macro definitions. Each definition is a form which begins with a name, followed by macro-style-params which is a macro parameter list, and zero or more macro-body-forms. These macro definitions are similar to those globally defined by the defmacro operator, except that they are in a local environment.
The macro definitions are followed by optional body-forms. The macros specified in the definitions are visible to these forms.
Forms inside the macro definitions such as the macro-body-forms, and initializer forms appearing in the macro-style-params are subject to macro-expansion in a scope in which none of the new macros being defined are yet visible. Once the macro definitions are themselves macro-expanded, they are placed into a new macro environment, which is then used for macro expanding the body-forms.
A macrolet form is fully processed in the expansion phase of a form, and is effectively replaced by progn form which contains expanded versions of body-forms. This expanded structure shows no evidence that any macrolet forms ever existed in it. Therefore, it is impossible for the code evaluated in the bodies and parameter lists of macrolet macros to have any visibility to any surrounding lexical variable bindings, which are only instantiated in the evaluation phase, after expansion is done and macros no longer exist.
A local macro defined using macrolet may decline to expand a macro form. Declining to expand is achieved by returning the original unexpanded form, which may be captured using the :form parameter. When a local macro declines to expand a form, the macro definition is temporarily hidden, as if it didn't exist in the lexical scope. If another macro of the same name is thereby revealed (a global macro or another local macro at a shallower nesting level), then an expansion is tried with that macro. If no such macro is revealed, or if a lexical function binding of that name is revealed, then no expansion takes place; the original form is taken as-is. When another macro is tried, the process repeats, resulting in a search which proceeds as far as possible through outer lexical scopes and finally the global scope.
(macro-form-p obj [env])
The macro-form-p function returns t if obj represents the syntax of a form which is a macro form: either a compound macro or a symbol macro. Otherwise it returns nil.
A macro form is one that will transform under macroexpand-1 or macroexpand; an object which isn't a macro form will not undergo expansion.
The optional env parameter is a macroexpansion environment. A macroexpansion environment is passed down to macros and can be received via their special :env parameter.
env is used by macro-form-p to determine whether obj is a macro in a lexical macro environment.
If env is not specified or is nil, then macro-form-p only recognizes global macros.
;; macro which translates to 'yes if its
;; argument is a macro from, or otherwise
;; transforms to the form 'no.
(defmacro ismacro (:env menv form)
(if (macro-form-p form menv)
''yes ''no))
(macrolet ((local ()))
(ismacro (local))) ;; yields yes
(ismacro (local)) ;; yields no
(ismacro (ismacro foo)) ;; yields yes
During macro expansion, the global macro ismacro is handed the macro-expansion environment via :env menv.
When the macro is invoked within the macrolet, this environment includes the macro-time lexical scope in which the local macro is defined. So when global checks whether the argument form (local) is a macro, the conclusion is yes: the (local) form is a macro call in that environment: macro-form-p yields t.
When (global (local)) is invoked outside of the macrolet, no local macro is visible is there, and so macro-form-p yields nil.
(macroexpand-1 obj [env])
(macroexpand obj [env])
If obj is a macro form (an object for which macro-form-p returns t), these functions expand the macro form and return the expanded form. Otherwise, they return obj.
macroexpand-1 performs a single expansion, expanding just the macro that is referenced by the symbol in the first position of obj, and returns the expansion. That expansion may itself be a macro form.
macroexpand performs an expansion similar to macroexpand-1. If the result is a macro form, then it expands that form, and keeps repeating this process until the expansion yields a non-macro-form. That non-macro-form is then returned.
The optional env parameter is a macroexpansion environment. A macroexpansion environment is passed down to macros and can be received via their special :env parameter. The environment they receive is their lexically apparent macro-time environment in which local macros may be visible. A macro can use this environment to "manually" expand some form in the context of that environment.
;; (rem-num x) expands x, and if x begins with a number,
;; it removes the number and returns the resulting
;; form. Otherwise, it returns the entire form.
(defmacro rem-num (:env menv some-form)
(let ((expanded (macroexpand some-form menv)))
(if (numberp (car expanded))
(cdr expanded)
some-form)))
(macrolet ((foo () '(1 list 42))
(bar () '(list 'a)))
(list (rem-num (foo)) (rem-num (bar))))
--> ((42) (a))
The rem-num macro is able to expand the (foo) and (bar) forms it receives as the some-form argument, even though these forms use local macro that are only visible in their local scope. This is thanks to the macro environment passed to rem-num. It is correctly able to work with the expansions (1 list 42) and (list 'a) to produce (list 42) and (list 'a) which evaluate to 42 and a respectively.
(macroexpand-1-lisp1 obj [env])
(macroexpand-lisp1 obj [env])
The macroexpand-1-lisp1 and macroexpand-lisp1 functions closely resemble, respectively, macroexpand-1 and macroexpand.
The argument and return value syntax and semantics is almost identical, except for one difference. These functions consider argument obj to be syntax in a Lisp-1 evaluation context, such as any argument position of the dwim operator, or the equivalent DWIM Brackets notation.
This makes a difference because in a Lisp-1 evaluation context, an inner function binding is able to shadow an outer symbol macro binding of the same name.
The requirements about this language area are given in more detail in the description of the dwim operator.
Note: the macroexpand-lisp1 function is useful to the implementor of a macro whose semantics requires one or more argument forms to be treated in a Lisp-1 context, in situations when such a macro needs to itself expand the material, rather than merely insert it as-is into the output code template.
(expand form [env])
(expand* form [env])
The functions expand and expand* both perform a complete expansion of form in the macro-environment env, and return that expansion.
If env is omitted, the expansion takes place in the global environment in which only global macros are visible.
The returned object is a structure that is devoid of any macro calls. Also, all macrolet and symacrolet blocks in form form are removed in the returned structure, replaced by their fully expanded bodies.
The difference between expand and expand* is that expand suppresses expansion-time deferred warnings (exceptions of type defr-warning), issued for unbound variables or functions. To suppress a warning means to intercept the warning exception with a handler which throws a continue exception to resume processing. What this requirement means is that if unbound functions or variables occur in the form being expanded by expand, the warning is effectively squelched. Rationale: expand is may be used by macros for expanding fragments which contain references to variables or functions which are not defined in those fragments.
(expand-with-free-refs form [inner-env [outer-env]])
The expand-with-free-refs form performs a full expansion of form, as if by the expand function and returns a list containing that expansion, plus four additional items which provide information about variable and function references which occur in form.
If both inner-env and outer-env are provided, then it is expected that inner-env is lexically nested within outer-env.
Note: it is not required that outer-env be the immediate parent of inner-env.
Note: a common usage situation is that outer-env is the environment of the invocation of a "parent" macro which generates a form that contains local macros. The bodies of those local macros use expand-with-free-refs, specifying their own environment as inner-env and that of their generating "parent" as outer-env.
In detail, the five items of the returned list are (expansion fv-inner ff-inner fv-outer ff-outer) whose descriptions are:
The semantics of the treatment of inner-env and outer-env in the calculation of fv-outer and ff-outer is as follows. A new environment diff-env is calculated from these two environments, and form is expanded in this environment. Variables and functions occurring in form which are not bound in diff-env are listed as fv-outer and ff-outer.
This diff-env is calculated as follows. First diff-env is initialized as a copy of outer-env. Then, all environments below outer-env down to inner-env are examined for bindings which shadow bindings in diff-env. Those shadows are removed from diff-env. Therefore, what remains in diff-env are those bindings from outer-env that are not shadowed by the environments between inner-env and outer-env.
Within each of the lists of variables returned by expand-with-free-refs, the order of the variables is not specified.
Suppose that mac is a macro which somehow has access to the two indicated lexical environments in the following code snippet:
(let (a c) ;; <- outer-env
(let (b)
(let (c) ;; <- inner-env
(mac (list a b c d)))))
Suppose that mac invokes the expand-with-free-refs function, passing in the (list a b c d) argument form as form and two macro-time environment objects corresponding to the indicated environments.
Then the following object shall be a correct return value of expand-with-free-refs:
((list a b c d) (d) nil (d c b) nil)
A complete code example of this is given below.
Other correct return values are possible due to permitted variations in the order of the variables within the four lists. For instance, instead of (d c b) the list (c b d) may appear.
The fv-inner list is (d) because this is the only variable that occurs in (list a b c d) which is free with regard to inner-env. The a, b and c variables are not listed because they appear bound inside inner-env.
The reported fv-outer list is (b c d) because the form is considered against diff-env which is formed by removing the shadowing bindings from outer-env. The difference between (a c) and (b c) is a and so the form is considered in an environment containing the binding a which leaves (b c d) free.
The following is a complete code sample demonstrating the above descriptions:
;; Given this macro:
(defmacro bigmac (:env out-env big-form)
^(macrolet ((mac (:env in-env little-form)
^',(expand-with-free-refs
little-form in-env ,out-env)))
,big-form))
(let (a c) ;; <- outer-env, surrounding bigmac
(bigmac
(let (b)
(let (c) ;; <- inner-env, surrounding mac
(mac (list a b c d))))))
--> ((list a b c d) (d) nil (d c b) nil)
Note: this information is useful because a set difference can be calculated between the two reported sets. The set difference between the fv-outer variables (b c d) and the fv-inner variables (d) is (b c).
That set difference (b c) is significant because it precisely informs about the bound variables which occur in (list a b c d) which appear bound in inner-env, but are not bound due to a binding coming from outer-env. In the above example, these are the variables enclosed in the bigmac macro, but external to the inner mac macro.
The variable d is not listed in (b c) because it is not a bound variable. The variable a is not in (b c) because though it is bound in inner-env, that binding comes from outer-env.
The upshot of this logic is that it allows a macro to inspect a form in order to discover the identities of the variables and functions which are used inside that form, whose definitions come from a specific, bounded scope surrounding that form.
(lexical-var-p env form)
(lexical-fun-p env form)
(lexical-symacro-p env form)
(lexical-macro-p env form)
These four functions are useful to macro writers. They are intended to be called from the bodies of macro expanders, such as the bodies of defmacro or macrolet forms. The env argument is a macro-time environment, which is available to macros via the special :env parameter. Using these functions, a macro can enquire whether a given form is, respectively, a symbol which has a variable binding, a function binding, a symbol macro (defined by symacrolet) or a macro (defined by macrolet) in the environment of the macro's invocation. This information is known during macro expansion. The macro expander recognizes lexical function and variable bindings, because these bindings can shadow macros.
Special variables are not lexical. The function lexical-var-p returns nil if form satisfies special-var-p function, indicating that it is the name of a special variable.
The lexical-var-p function also returns nil for global lexical variables. If form is a symbol for which only a global lexical variable binding is apparent, lexical-var-p returns nil. Testing for the existence for a global variable can be done using boundp; if a symbol is boundp but not special-var-p, then it is a global lexical variable.
Similarly, lexical-fun-p returns nil for global functions, lexical-symacro-p returns nil for global symbol macros and lexical-macro-p returns nil for global macros.
;;
;; this macro replaces itself with :lexical-var if its
;; argument is a lexical variable, :lexical-fun if
;; its argument is a lexical function, or with
;; :not-lex-fun-var if neither is the case.
;;
(defmacro classify (sym :env e)
(cond
((lexical-var-p e sym) :lexical-var)
((lexical-fun-p e sym) :lexical-fun)
(t :not-lex-fun-var)))
;;
;; Use classify macro above to report classification
;; of the x, y and f symbols in the given scope
;;
(let ((x 1) (y 2))
(symacrolet ((y x))
(flet ((f () (+ 2 2)))
(list (classify x) (classify y) (classify f)))))
--> (:lexical-var :not-lex-fun-var :lexical-fun)
;; Locally bound specials are not lexical
(let ((*stdout* *stdnull*))
(classify *stdout*))
--> :not-lex-fun-var
(lexical-binding-kind env symbol)
The lexical-binding-kind function inspects the macro-time environment env to determine what kind of binding, if any, does symbol have in the the variable namespace of that environment.
If the innermost binding for symbol is a variable binding, then either :var is returned if the variable is lexical, otherwise nil is returned if the variable is special.
If the innermost binding for symbol is a symbol macro, then :symacro is returned.
In all other cases, nil is returned. The function does not consider global symbol macros or global lexical variables.
(lexical-fun-binding-kind env symbol)
The lexical-fun-binding-kind function inspects the macro-time environment env to determine what kind of binding, if any, does symbol have in the the function namespace of that environment.
If the innermost binding for symbol is a function binding, then :fun is returned.
If the innermost binding for symbol is a macro, then :macro is returned.
In all other cases, nil is returned. The function does not consider global macros or functions.
(lexical-lisp1-binding env symbol)
The lexical-lisp1-binding function inspects the macro-time environment env to determine what kind of binding, if any, does symbol have in that environment, from a Lisp-1 perspective.
That is to say, it considers function bindings, variable bindings and symbol macro bindings to be in a single name space and finds the innermost binding of one of these types for symbol.
If such a binding is found, then the function returns one of the three keyword symbols :var, :fun, or :symacro.
If no such lexical binding is found, then the function returns nil.
Note that :var is never returned for a special variable, but such a variable can be shadowed by a symbol macro, in which case :symacro is returned.
Note that a nil return doesn't mean that the symbol doesn't have a lexical binding. It could have an operator macro lexical binding (a macro binding in the function namespace established by macrolet). Unlike the lexical-binding-kind function, the lexical-lisp1-binding function never returns :macro because Lisp-1-style evaluation of symbols is blind to the existence of macros, other than symbol macros.
(defsymacro sym form)
A defsymacro form introduces a symbol macro. A symbol macro consists of a binding between a symbol sym and and a form. The binding denotes the form itself, rather than its value.
The form argument is not subject to macro expansion; it is associated with sym in its unexpanded state, as it appears in the defmacro form.
The defsymacro form must be evaluated for its defining to take place; therefore, the definition is not available in the top-level form which contains the defsymacro invocation; it becomes available to a subsequent top-level form.
Subsequent to the evaluation of the defsymacro definition, whenever the macro expander encounters sym sym as a form, it replaces it by form. After this replacement takes place, form itself is then processed for further replacement of macros and symbol macros.
Symbol macros are also recognized in contexts where sym denotes a place which is the target of an assignment operation like set and similar.
Note: if a symbol macro expands to itself directly, expansion stops. However, if a symbol macro expands to itself through a chain of expansions, runaway expansion-time recursion will occur.
If a global variable exists by the name sym, then defsymacro first removes that variable from the global environment, and if that variable is special, the symbol's special marking is removed. defsymacro doesn't alter the dynamic binding of a special variable. Any such a binding remains intact. If defsymacro is evaluated in a scope in which there is any lexical or dynamic binding of sym in the variable namespace, whether as a variable or macro, the global symbol macro is shadowed by that binding.
(symacrolet ({(sym form)}*) body-form*)
The symacrolet operator binds local, lexically scoped macros that are similar to the global symbol macros introduced by defsymacro.
Each sym in the bindings list is bound to its corresponding form, creating a new extension of the expansion-time lexical macro environment.
Each body-form is subsequently macro-expanded in this new environment in which the new symbol macros are visible.
Note: ordinary lexical bindings such as those introduced by let or by function parameters lists shadow symbol macros. If a symbol x is bound by nested instances of macrolet and a let, then the scope enclosed by both constructs will see whichever of the two bindings is more inner, even though the bindings are active in completely separate phases of processing.
From the perspective of the arguments of a dwim form, lexical function bindings also shadow symbol macros. This is consistent with the Lisp-1-style name resolution which applies inside a dwim form. Lexical operator macros do not shadow symbol macros under any circumstances.
(placelet ({(sym place)}*) body-form*)
(placelet* ({(sym place)}*) body-form*)
The placelet macro binds lexically scoped symbol macros in such a way that they behave as aliases for places denoted by place forms.
Each place must be an expression denoting a syntactic place. The corresponding sym is established as an alias for the storage location which that place denotes, over the scope of the body-forms.
This binding takes place in such a way that each place is evaluated exactly once, only in order to determine its storage location. The corresponding sym then serves as an alias for that location, over the scope of the body-forms. This means that whenever sym is evaluated, it stands for the value of the storage location, and whenever a value is apparently stored into sym, it is actually the storage location which receives it.
The placelet* variant implements an alternative scoping rule, which allows a later place form to refer to a sym bound to an earlier place form. In other words, a given sym binding is visible not only to the body-forms but also to place forms which occur later.
Note: certain kinds of places, notably (force promise) expressions, must be accessed before they can be stored, and this restriction continues to hold when those places are accessed through placelet aliases.
Note: placelet differs from symacrolet in that the forms themselves are not aliased, but the storage locations which they denote. (symacrolet ((x y)) z) performs the syntactic substitution of symbol x by form y, wherever x appears inside z as an evaluated form, and is not shadowed by any inner binding. Whereas (placelet ((x y)) z) generates code which arranges for y to be evaluated to a storage location, and syntactically replaces occurrences of x with a form which directly denotes that storage location, wherever x appears inside z as an evaluated form, and is not shadowed by any inner binding. Also, x is not necessarily substituted by a single, fixed form, as in the case of symacrolet. Rather it may be substituted by one kind of form when it is treated as a pure value, and another kind of form when it is treated as a place.
Note: multiple accesses to an alias created by placelet denote multiple accesses to the aliased storage location. That can mean multiple function calls or array indexing operations and such. If the target of the alias is (read-once place) instead of place, then a single access occurs to fetch the prior value of place and stored into a hidden variable. All of the multiple occurrences of the alias then simply retrieve this cached prior value from the hidden variable, rather than accessing the place. The read-once macro is independent of placelet and separately documented.
Implementation of inc using placelet:
(defmacro inc (place : (delta 1))
(with-gensyms (p)
^(placelet ((,p ,place))
(set ,p (+ ,p ,delta)))))
The gensym p is used to avoid accidental capture of references emanating from the delta form.
(expander-let ({(sym init-form)}*) body-form*)
The expander-let operator strongly resembles let* but has different semantics, relevant to expansion. It also has a stricter syntax in that variables may not be symbols without a init-form: only variable binding specifications of the form (sym<< init-form ) are allowed.
Symbols bound using expander-let are expected to be special variables. For every sym, the expression (special-var-p sym) should be true. The behavior is unspecified for any sym which doesn't name a special variable.
The expander-let macro establishes a new dynamic environment which each given sym has the value of the specified init-form which is evaluated in the top-level environment. Then, the body-forms are turned into the arguments of a progn form, and that form is then expanded in the new environment in which the dynamic bindings are visible.
Thus expander-let may be used to bind special variables which are visible to expansion-time computations occurring within body-forms. A macro may generate an expander-let form in order to communicate values to macros contained in that form.
(macro-time form*)
The macro-time macro evaluates its arguments immediately during macro expansion.
The form arguments are processed from left to right. Each form is fully expanded and evaluated in the top-level environment before the next form is considered.
The value of the last form, or else nil if there aren't any arguments, is converted into a literal expression which denotes that value, and the resulting literal is produced as the expansion of macro-time.
Note: macro-time supports techniques that require a calculation to be performed in the environment where the program is being compiled, and inserting the result of that calculation as a literal into the program source. Possibly, the calculation can have some useful effect in that environment, or use as an input information that is available in that environment. The load-time operator also inserts a calculated value as a de facto literal into the program, but it performs that calculation in the environment where the compiled file is being loaded. The two operators may be considered complementary in this sense.
Consider the source file:
(defun host-name-c () (macro-time (uname).nodename))
(defun host-name-l () (load-time (uname).nodename))
If this is compiled via compile-file, the uname call in host-name-c takes place when it is macro-expanded. Thereafter, the compiled version of the function returns the name of the machine where the compilation took place, no matter in what environment it is subsequently loaded and called.
In contrast, the compilation of host-name-l arranges for that function's uname call to take place just one time, whenever the compiled file is loaded. Each time the function is subsequently called, it will return the name of the machine where it was loaded, without making any additional calls to uname.
Note: macro-time can be understood in terms of the following implementation. Note that this implementation always produces a quote expression, which macro-time is not required to do if val is self-evaluating:
(defmacro macro-time (. forms)
(let (val)
(each ((f forms))
(set val (eval f)))
^(quote ,val)))
Because eval treats a top-level progn specially, this implementation is also possible:
(defmacro macro-time (. forms)
^(quote ,(eval ^(progn ,*forms))))
;; The (1 2 3) object is produced at macro-expansion time, becoming
;; a quoted literal which evaluates to (1 2 3).
(macro-time (list 1 2 3)) -> (1 2 3)
;; The above fact is revealed by macroexpand: the list form was
;; evaluated, and then quote was inserted to produce (quote (1 2 3))
;; which is notated '(1 2 3):
(macroexpand '(macro-time (list 1 2 3))) -> '(1 2 3)
;; Quote isn't required on a self-evaluating object; it serves
;; as a literal expression denoting itself:
(macroexpand '(macro-time (join-with "-" "a" "b"))) -> "a-b"
(equot form)
The equot macro ("expand and quote") performs a full expansion of form in the surrounding macro environment. Then it constructs a quote form whose argument is the expansion. This quote form is then returned as the macro replacement for the original equot form.
(symacrolet ((a (+ 2 2)))
(list (quote a) (equot a) a))
--> (a (+ 2 2) 4)
Above, the expansion of a is (+ 2 2). Thus the macro call (equot a) expands to (quote (+ 2 2)). When that is evaluated, it yields (+ 2 2).
If a is quoted, then the result is a: no expansion or evaluation takes place. Whereas if a is presented for evaluation, then not only is it expanded to (+ 2 2), but that expansion is reduced to 4.
The equot operator is a an intermediate point between these two semantics: it permits expansion to proceed, but then suppresses evaluation of the result.
(tree-bind macro-style-params expr form*)
(mac-param-bind context-expr
macro-style-params expr form*)
(mac-env-param-bind context-expr env-expr
macro-style-params expr form*)
The tree-bind operator evaluates expr, and then uses the resulting value as a counterpart to a macro-style parameter list. If the value has a tree structure which matches the parameters, then those parameters are established as bindings, and the forms, if any, are evaluated in the scope of those bindings. The value of the last form is returned. If there are no forms, nil is returned. Under tree-bind, the value of the :form available to macro-style-params is the tree-bind form itself.
The mac-param-bind operator is similar to tree-bind except that it takes an extra argument, context-expr. This argument is an expression which is evaluated. It is expected to evaluate to a compound form. If an error occurs during binding, the error diagnostic message is based on information obtained from this form. By contrast, the tree-bind operator's error diagnostic refers to the tree-bind form, which is cryptic if the binding is used for the implementation of some other construct, hidden from the user of that construct. In addition, context-expr specifies the value for the :form parameter that macro-style-params may refer to.
The mac-env-param-bind is an extension of mac-param-bind which takes one more argument, env-expr, before the macro parameters. This expression is evaluated, and becomes the value of the :env parameter that macro-style-params may refer to.
Under tree-bind and mac-param-bind, the :env parameter takes on the value nil.
Under all three operators, the :whole parameter takes on the value of expr.
These operators throw an exception if there is a structural mismatch between the parameters and the value of expr. One way to avoid this exception is to use tree-case, which is based on the conventions of tree-bind. There exists no tree-case analog for mac-param-bind or mac-env-param-bind.
(tree-case expr {(macro-style-params form*)}*)
The tree-case operator evaluates expr and matches it against a succession of zero or more cases. Each case defines a pattern match, expressed as a macro style parameter list macro-style-params.
If the object produced by expr matches macro-style-params, then the parameters are bound, becoming local variables, and the forms, if any, are evaluated in order in the environment in which those variables are visible. If there are forms, the value of the last form becomes the result value of the case, otherwise the result value of the case is nil.
If the result value of a case is the object : (the colon symbol), then processing continues with the next case. Otherwise the evaluation of tree-case terminates, returning the result value.
If the value of expr does not match the macro-style-params parameter list of a case, processing continues with the next case.
If no cases match, then tree-case terminates, returning nil.
;; reverse function implemented using tree-case
(defun tb-reverse (obj)
(tree-case obj
(() ()) ;; the empty list is just returned
((a) obj) ;; one-element list returned
((a . b) ^(,*(tb-reverse b) ,a)) ;; car/cdr recursion
(a a))) ;; atom is just returned
Note that in this example, the atom case is placed last, because an argument list which consists of a symbol is a "catch all" match that matches any object. We know that it matches an atom, because the previous (a . b) case matches conses. In general, the order of the cases in tree-case is important: even more so than the order of cases in a cond or caseql. The one-element list case is unnecessary; it can be removed.
(tb macro-style-params form*)
The tb macro is similar to the lambda operator but its argument binding is based on a macro-style parameter list. The name is an abbreviation of tree-bind.
A tb form evaluates to a function which takes a variable number of arguments.
When that function is called, those arguments are taken as a list object which is matched against macro-style-params as if by tree-bind. If the match is successful, then the parameters are bound to the corresponding elements from the argument structure and each successive form is evaluated in an environment in which those bindings are visible. The value of the last form is the return value of the function. If there are no forms, the function's return value is nil.
The following equivalence holds, where args should be understood to be a globally unique symbol:
(tb pattern body ...) <--> (lambda (. args)
(tree-bind pattern args body ...))
(tc {(macro-style-params form*)}*)
The tc macro produces an anonymous function whose behavior is closely based on the tree-case operator. Its name is an abbreviation of tree-case.
The anonymous function takes a variable number of arguments. Its argument list is taken to be the value macro is tested against the multiple pattern clauses of an implicit tree-case. The return value of the function is that of the implied tree-case.
The following equivalence holds, where args should be understood to be a globally unique symbol:
(tc clause1 clause2 ...) <--> (lambda (. args)
(tree-case args
clause1 clause2 ...))
(with-gensyms (sym*) body-form*)
The with-gensyms evaluates the body-forms in an environment in which each variable name symbol sym is bound to a new uninterned symbol ("gensym").
The code:
(let ((x (gensym))
(y (gensym))
(z (gensym)))
^(,x ,y ,z))
may be expressed more conveniently using the with-gensyms shorthand:
(with-gensyms (x y z)
^(,x ,y ,z))
Parameter list macros, also more briefly called parameter macros are an original feature of TXR Lisp.
If the first element of a function or macro parameter list is a keyword symbol other than :env, :whole, :form or : (the colon symbol), it denotes a parameter macro. This keyword symbol is expected to have a binding in the parameter macro namespace: a global namespace which associates keyword symbols with parameter list expander functions.
Parameter list macros are recognized in both function parameter lists and macro parameter lists. A macro parameter list can, via nesting, contain multiple nested parameter lists. Each such nested list may contain parameter macro invocations; those are all traversed and processed.
Expansion of a parameter list macro occurs at macro-expansion time, when a function's or macro's parameter list is traversed by the macro expander. It takes place as follows. First, the keyword is removed from the parameter list. The keyword's binding in the parameter macro namespace is retrieved. If it doesn't exist, an exception is thrown. Otherwise, the remaining parameter list is first recursively processed for more occurrences of parameter macros. This expansion produces a transformed parameter list, along with a transformed function body. These two artifacts are then passed to the transformer function retrieved from the keyword symbol's binding. The function returns a further transformed version of the parameter list and body. These are processed for more parameter macros. The process terminates when no more expansion is possible, because a parameter list has been produced which does not begin with a parameter macro. This final parameter list and its accompanying body are then taken in place of the original parameter list and body.
TXR Lisp provides a two built-in parameter list macros. The :key parameter macro endows a function keyword parameters. The :match parameter macro allows a function to be expressed using pattern matching, which requires the body to consist of pattern-matching clauses.
The implementation of both of these macros is written entirely using this parameter list macro mechanism, by means of the public define-param-expander macro.
The variable *param-macro* holds a hash table which associates keyword symbols with parameter list expander functions.
The functions are expected to conform to the following syntax:
(lambda (params body env form) form*)
The params parameter receives the parameter list of the function which is undergoing parameter expansion. All other parameter macros have already been expanded.
The body parameter receives the list of body forms. The function is expected to return a cons cell whose car contains the transformed parameter list, and whose cdr contains the transformed list of body forms. Parameter expansion takes place at macro expansion time.
The env parameter receives the macro-expansion-time environment which surrounds the function being expanded. Note that this environment doesn't take into account the parameters themselves; therefore, it is not the correct environment for expanding macros among the body forms. For that purpose, it must be extended with shadowing entries, the manner of doing which is undocumented. However env may be used directly for expanding init forms for optional parameters occurring in params.
The form parameter receives the overall function-defining form that is being processes, such as a defun or lambda form. This is intended for error reporting.
A parameter transformer returns the transformed parameter list and body as a single object: a list whose first element is the parameter list, and whose remaining elements are the forms of the body. Thus, the following is a correct null transformer:
(lambda (params body env form)
(cons params body))
(define-param-expander name (pvar bvar : evar fvar)
form*)
The define-param-expander macro provides syntax for defining parameter macros. Invocations of this macro expand to code which constructs an anonymous function and installs it into the *param-macro* hash table, under the key given by name.
The name parameter's argument should be a keyword symbol that is valid for use as a parameter macro name.
The pvar, bvar, evar and fvar arguments must be symbols suitable for variable binding. These symbols define the parameters of the expander function which shall, respectively, receive the parameter list, body forms, macro environment and function form. If evar is omitted, a symbol generated by the gensym function is used. Likewise if fvar is omitted.
The form arguments constitute the body of the expander.
The define-param-expander form returns name.
The parameter macro returns the transformed parameter list and body as a single object: a list whose first element is the parameter list, and whose remaining elements are the forms of the body.
The following example shows the implementation of a parameter macro :memo which provides rudimentary memoization. Using the macro is extremely easy. It is a matter of simply inserting the :memo keyword at the front of a function's parameter list. The function is then memoized.
(defvarl %memo% (hash :weak-keys))
(defun ensure-memo (sym)
(or (gethash %memo% sym)
(sethash %memo% sym (hash))))
(define-param-expander :memo (param body)
(let* ((memo-parm [param 0..(posq : param)])
(hash (gensym))
(key (gensym)))
^(,param (let ((,hash (ensure-memo ',hash))
(,key (list ,*memo-parm)))
(or (gethash ,hash ,key)
(sethash ,hash ,key (progn ,*body)))))))
The above :memo macro may be used to define a memoized Fibonacci function as follows:
(defun fib (:memo n)
(if (< n 2)
(clamp 0 1 n)
(+ (fib (pred n)) (fib (ppred n)))))
All that is required is the insertion of the :memo keyword.
(:key non-key-param*
[ -- {sym | (sym [init-form [p-sym]])}* ]
[ . rest-param ])
Parameter list macro :key injects keyword parameter support into functions and macros.
When :key appears as the first item in a function parameter list, a special syntax is recognized in the parameter list. After any required and optional parameters, the symbol -- (two dashes) may appear. Parameters after this symbol are interpreted as keyword parameters. After the keyword parameters, a rest parameter may appear in the usual way as a symbol in the dotted position.
Keyword parameters use the same syntax as optional parameters, except that if used in a macro parameter list, they do not support destructuring whereas optional parameters do. That is to say, regardless whether :key is used in a function or macro, keyword parameters are symbols.
A keyword parameter takes three possible forms:
In a call to a :key-enabled function, keyword arguments begin after those arguments which satisfy all of the required and optional parameters. Keyword arguments consist of interleaved indicators and values, which are separate arguments. Thus passing a keyword argument actually requires the passing of two function arguments: an indicator keyword symbol, followed by the associated value. The indicator keywords are expected to have the same symbol name as the defined keyword parameters. For instance, the indicator-value pair :xyz 42 passes the value 42 to a keyword parameter that may be named xyz in any package: it may be usr:xyz or mypackage:xyz and so forth. Arguments specifying unrecognized keywords are ignored.
If the function has a rest-param, then that parameter receives the keyword arguments as a list. Since that list contains indicators and values, it is a de facto property list. In detail, the :key mechanism generates a regular variadic function which receives the keyword arguments as the trailing argument list. That function parses the recognized keyword arguments out of the trailing list, and binds them to the keyword parameter symbols as local variables. If a rest-param parameter is defined, then the entire keyword argument list is available through that parameter, and the keyword argument parsing logic also refers to the value of that parameter to gain access to the keyword arguments. If there is no rest-param specified, then the :key macro adds a rest-param using a machine-generated symbol. The argument parsing logic then refers to the value of that symbol.
Define a function fun with two required arguments a b, one optional argument c, two keyword arguments foo and bar, and a rest parameter klist:
(defun fun (:key a b : c -- foo bar . klist)
(list a b c foo bar klist))
(fun 1 2 3 :bar 4) -> (1 2 3 nil 4 (:bar 4))
Define a function with only keyword arguments, with default expressions and Boolean indicator params:
(defun keyfun (:key -- (a 10 a-p) (b 20 b-p))
(list a a-p b b-p))
(keyfun :a 3) -> (3 t 20 nil)
(keyfun :b 4) -> (10 nil 4 t)
(keyfun :c 4) -> (10 nil 20 nil)
(keyfun) -> (10 nil 20 nil)
(expand-params proto-form [env])
The expand-param function expands all of the parameter list macros expressed in the prototype form proto-form, returning an expanded version of the form.
The proto-form is a compound form which has a shape very similar to a lambda expression, and may be a lambda expression.
The first element of proto-form is a name, which is an arbitrary object, though the use of a symbol is strongly recommended. This object plays no role in expand-params other than for composing diagnostic messages if errors occur.
The second element of proto-form is the parameter list.
The remaining elements of proto-form are zero or more body forms.
If proto-form contains no parameter macro invocations, then it is returned.
The optional env parameter specifies the macro environment which is passed to the parameter macro expanders, which they can receive via the :env parameter. The default value nil specifies the top-level environment.
;; No expansion: argument is returned
(macroexpand-params '(foo (arg) body)) -> (foo (arg) body)
;; Expand :key macro
(macroexpand-params '(bar (:key a b c -- d (e 1234 f-p)) body))
--> (bar (a b c . #:g0014)
(let (d e f-p)
(let ((#:g0015 (memp :d #:g0014)))
(when #:g0015
(set d (cadr #:g0015))))
(let ((#:g0015 (memp :e #:g0014)))
(cond
(#:g0015 (set e (cadr #:g0015))
(set f-p t))
(t (set e 1234))))
body))
(set {place new-value}*)
The set operator stores the values of expressions in places. It must be given an even number of arguments.
If there are no arguments, then set does nothing and returns nil.
If there are two arguments, place and new-value, then place is evaluated to determine its storage location, then new-value is evaluated to determine the value to be stored there, and then the value is stored in that location. Finally, the value is also returned as the result value.
If there are more than two arguments, then set performs multiple assignments in left-to-right order. Effectively, (set v1 e1 v2 e2 ... vn en) is precisely equivalent to (progn (set v1 e1) (set v2 e2) ... (set vn en)).
(pset {place new-value}*)
The syntax of pset is similar to that of set, and the semantics is similar also in that zero or more places are assigned zero or more values. In fact, if there are no arguments, or if there is exactly one pair of arguments, pset is equivalent to set.
If there are two or more argument pairs, then all of the arguments are evaluated first, in left-to-right order. No store takes place until after every place is determined, and every new-value is calculated. During the calculation, the values to be stored are retained in hidden, temporary locations. Finally, these values are moved into the determined places. The rightmost value is returned as the form's value.
The assignments thus appear to take place in parallel, and pset is capable of exchanging the values of a pair of places, or rotating the values among three or more places. (However, there are more convenient operators for this, namely rotate and swap).
;; exchange x and y
(pset x y y x)
;; exchange elements 0 and 1; and 2 and 3 of vector v:
(let ((v (vec 0 10 20 30))
(i -1))
(pset [vec (inc i)] [vec (inc i)]
[vec (inc i)] [vec (inc i)])
vec)
-> #(10 0 30 20)
(zap place [new-value])
The zap macro assigns new-value to place and returns the previous value of place.
If new-value is missing, then nil is used.
In more detail, first place is evaluated to determine the storage location. Then, the location is accessed to retrieve the previous value. Then, the new-value expression is evaluated, and that value is placed into the storage location. Finally, the previously retrieved value is returned.
(flip place)
The flip macro toggles the Boolean value stored in place.
If place previously held nil, it is set to t, and if it previously held a value other than nil, it is set to nil.
(test-set place)
(test-clear place)
The test-set macro examines the value of place. If it is nil then it stores t into the place, and returns t. Otherwise it leaves place unchanged and returns nil.
The test-clear macro examines the value of place. If it is Boolean true (any value except nil) then it stores nil into the place, and returns t. Otherwise it leaves place unchanged and returns nil.
(compare-swap place cmp-fun cmp-val store-val)
The compare-swap macro examines the value of place and compares it to cmp-val using the comparison function given by the function name cmp-fun.
This comparison takes places as if by evaluating the expression (cmp-fun value cmp-val) where value denotes the current value of place.
If the comparison is false, place is not modified, the store-val expression is not evaluated, and the macro returns nil.
If the comparison is true, then compare-swap evaluates the store-val expression, stores the resulting value into place and returns t.
(ensure place init-expr)
The ensure macro examines the value of place.
If the current value is nil, then init-expr is evaluated. The value is stored in place and becomes the result of the ensure form.
If the value of place is other than nil, then the form yields that value. In this case, init-expr isn't evaluated, and place isn't modified.
The place expression is evaluated only once to determine the place.
(inc place [delta])
(dec place [delta])
The inc macro increments place by adding delta to its value. If delta is missing, the value used in its place the integer 1.
First the place argument is evaluated as a syntactic place to determine the location. Then, the value currently stored in that location is retrieved. Next, the delta expression is evaluated. Its value is added to the previously retrieved value as if by the + function. The resulting value is stored in the place, and returned.
The macro dec works exactly like inc except that addition is replaced by subtraction. The similarly defaulted delta value is subtracted from the previous value of the place.
(pinc place [delta])
(pdec place [delta])
The macros pinc and pdec are similar to inc and dec.
The only difference is that they return the previous value of place rather than the incremented value.
(test-inc place [delta [from-val]])
(test-dec place [delta [to-val]])
The test-inc and test-dec macros provide combined operations which change the value of a place and provide a test whether, respectively, a certain previous value was overwritten, or a certain new value was attained. By default, this tested value is zero.
The test-inc macro notes the prior value of place and then updates it with that value, plus delta, which defaults to 1. If the prior value is eql to from-val then it returns t, otherwise nil. The default value of from-val is zero.
The test-dec macro produces a new value by subtracting delta from the value of place. The argument delta defaults to 1. The new value is stored into place. If the new value is eql to to-val then t is returned, otherwise nil.
(swap left-place right-place)
The swap macro exchanges the values of left-place and right-place and returns the value which is thereby transferred to right-place.
First, left-place and right-place are evaluated, in that order, to determine their locations. Then the prior values are retrieved, exchanged and stored back. The value stored in right-place is also returned.
If left-place and right-place are ranges of the same sequence, the behavior is not specified if the ranges overlap or are of unequal length.
Note: the rotate macro's behavior is somewhat more specified in this regard. Thus, although any correct swap expression can be expressed using rotate, but the reverse isn't true.
(push item place)
The push macro places item at the head of the list stored in place and returns the updated list which is stored back in place.
First, the expression item is evaluated to produce the push value. Then, place is evaluated to determine its storage location. Next, the storage location is accessed to retrieve the list value which is stored there. A new object is produced as if by invoking cons function on the push value and list value. This object is stored into the location, and returned.
(pop place)
The pop macro removes an element from the list stored in place and returns it.
First, place is evaluated to determine the place. The place is accessed to retrieve the original value. Then a new value is calculated, as if by applying the cdr function to the old value. This new value is stored. Finally, a return value is calculated and returned, as if by applying the car function to the original value.
(pushnew item place [testfun [keyfun]])
The pushnew macro inspects the list stored in place. If the list already contains the item, then it returns the list. Otherwise it creates a new list with the item at the front and stores it back into place, and returns it.
First, the expression item is evaluated to produce the push value. Then, place is evaluated to determine its storage location. Next, the storage location is accessed to retrieve the list value which is stored there. The list is inspected to check whether it already contains the push value, as if using the member function. If that is the case, the list is returned and the operation finishes. Otherwise, a new object is produced as if by invoking cons function on the push value and list value. This object is stored into the location and returned.
(shift place+ shift-in-value)
The shift macro treats one or more places as a "multi-place shift register". The values of the places are shifted one place to the left. The first (leftmost) place receives the value of the second place, the second receives that of the third, and so on. The last (rightmost) place receives shift-in-value (which is not treated as a place, even if it is a syntactic place form). The previous value of the first place is returned.
More precisely, all of the argument forms are evaluated left to right, in the process of which the storage locations of the places are determined, shift-in-value is reduced to its value.
The values stored in the places are sampled and saved.
Note that it is not specified whether the places are sampled in a separate pass after the evaluation of the argument forms, or whether the sampling is interleaved into the argument evaluation. This affects the behavior in situations in which the evaluation of any of the place forms, or of shift-in-value, has the side effect of modifying later places.
Next, the places are updated by storing the saved value of the second place into the first place, the third place into the second and so forth, and the value of shift-in-value into the last place.
Finally, the saved original value of the first place is returned.
If any of the places are ranges which index into the same sequence, and the behavior is not otherwise unspecified due to the issue noted in an earlier paragraph, the effect upon the multiply-stored sequence can be inferred from the above-described storage order. Note that even if stores take place which change the length of the sequence and move some elements, not-yet-processed stores whose ranges to refer to these elements are not adjusted.
With regard to the foregoing paragraph, a recommended practice is that if subranges of the same sequence object are shifted, they be given to the macro in ascending order of starting index. Furthermore, the semantics is simpler if the ranges do not overlap.
(rotate place*)
Treats zero or more places as a "multi-place rotate register". If there are no arguments, there is no effect and nil is returned. Otherwise, the last (rightmost) place receives the value of the first (leftmost) place. The leftmost place receives the value of the second place, and so on. If there are two arguments, this equivalent to swap. The prior value of the first place, which is the value rotated into the last place, is returned.
More precisely, the place arguments are evaluated left to right, and the storage locations are thereby determined. The storage locations are sampled, and then the sampled values are stored back into the locations, but rotated by one place as described above. The saved original value of the leftmost place is returned.
It is not specified whether the sampling of the original values is a separate pass which takes place after the arguments are evaluated, or whether this sampling it is interleaved into argument evaluation. This affects the behavior in situations in which the evaluation of any of the place forms has the side effect of modifying the value stored in a later place form.
If any of the places are ranges which index into the same sequence, and the behavior is not otherwise unspecified due to the issue noted in the preceding paragraph, the effect upon the multiply-stored sequence can be inferred from the above-described storage order. Note that even if stores take place which change the length of the sequence and move some elements, not-yet-processed stores whose ranges to refer to these elements are not adjusted.
With regard to the foregoing paragraph, a recommended practice is that if subranges of the same sequence object are shifted, they be given to the macro in ascending order of starting index. Furthermore, the semantics is simpler if the ranges do not overlap.
(del place)
The del macro requests the deletion of place. If place doesn't support deletion, an exception is thrown.
First place is evaluated, thereby determining its location. Then the place is accessed to retrieve its value. The place is then subject to deletion. Finally, the previously retrieved value is returned.
Precisely what deletion means depends on the kind of place. The built-in places in TXR Lisp have deletion semantics which are intended to be unsurprising to the programmer familiar with the data structure which holds the place.
Generally, if a place denotes the element of a sequence, then deletion of the place implies deletion of the element, and deletion of the element implies that the gap produced by the element is closed. The deleted element is effectively replaced by its successor, that successor by its successor and so on. If a place denotes a value stored in a dynamic data set such as a hash table, then deletion of that place implies deletion of the entry which holds that value. If the entry is identified by a key, that key is also removed.
If place is a DWIM bracket expression indexing into a structure, the structure is expected to implement the lambda and lambda-set methods. Moreover, the place form must have only two arguments: the object and an index argument. In other words, the del form must have this syntax:
(del [obj index])
The lambda method will be invoked with the unmodified obj and index arguments to determine the prior value to be returned. Then the lambda-set method will be invoked with three arguments: obj, a possibly modified index value and the argument nil representing an empty replacement sequence.
If index is a sequence or range, it is passed to the lambda-set method unmodified. Otherwise it is expected to be an integer, and converted into a one-element range spanning the indicated element. For instance, if the index value is 3, it is converted to the range #R(3 4). In effect, the lambda-set method is thereby asked to replace the one-element subsequence starting at index 3 with the empty sequence nil.
(lset {place}+ sequence-expr)
The lset operator's parameter list consists of one or more places followed by an expression sequence-expr.
The macro evaluates sequence-expr, which is expected to produce a sequence.
Successive elements of the resulting list are then assigned to each successive place.
If there are fewer elements in the sequence than places, the unmatched places receive the value nil.
Excess elements in the sequence are ignored.
An error exception occurs if the sequence is an improper list with fewer elements than places.
A lset form produces the value of sequence-expr as its result value.
(upd place opip-arg*)
The upd macro evaluates place and passes the value as an argument to the operational pipeline function formed, as if by the opip macro, from the opip-arg arguments. The result of this function is then stored back into place.
The following equivalence holds, except that place p is evaluated only once:
(upd p x y z ...) <--> (set p (call (opip x y z ...) p))
TXR Lisp provides a number of place-modifying operators such as set, push, and inc. It also provides a variety of kinds of syntactic places which may be used with these operators.
Both of these categories are open-ended: TXR Lisp programs may extend the set of place-modifying operators, as well as the vocabulary of forms which are recognized as syntactic places.
Regarding place operators, it might seem obvious that new place operators can be developed, since they are macros, and macros can expand to uses of existing place operators. As an example, it may seem that inc operator could be written as a macro which uses set:
(defmacro new-inc (place : (delta 1))
^(set ,place (+ ,place ,delta)))
However, the above new-inc macro has a problem: the place argument form is inserted into two places in the expansion, which leads to two evaluations. This is visibly incorrect if the place form contains any side effects. It is also potentially inefficient.
TXR Lisp provides a framework for writing place update macros which evaluate their argument forms once, even if they have to access and update the same places.
The framework also supports the development of new kinds of place forms as capsules of code which introduce the right kind of material into the lexical environment of the body of an update macro, to enable this special evaluation.
The expanders operate independently, and it is expected that place-modifying operators choose one of the three, and use only that expander. For example, accessing a place with an update expander and then overwriting its value with a clobber expander may result in incorrect code which contains multiple evaluations of the place form.
The programmer who implements a new place does not write expanders directly, but rather defines them via the defplace, define-accessor or defset macro.
The programmer who implements a new place update macro likewise does not call the expanders directly. Usually, they are invoked via the macros with-update-expander, with-clobber-expander and with-delete-expander. These are sufficient for most kind of macros. In certain complicated cases, expanders may be invoked using the wrapper functions call-update-expander, call-clobber-expander and call-delete-expander. These convenience macros and functions perform certain common chores, like macro-expanding the place in the correct environment, and choosing the appropriate function.
The expanders are described in the following sections.
(lambda (getter-sym setter-sym place-form
body-form) ...)
The update expander is a code-writer. It takes a body-form argument, representing code, and returns a larger form which surrounds this code with additional code.
This larger form returned by the update expander can be regarded as having two abstract actions, when it is substituted and evaluated in the context where place-form occurs. The first abstract action is to evaluate place-form exactly one time, in order to determine the actual run-time location to which that form refers. The second abstract action is to evaluate the caller's body-forms, in a lexical environment in which bindings exist for some lexical functions or (more usually) lexical macros. These lexical macros are explicitly referenced by the body-form; the update expander just provides their definition, under the names it is given via the getter-sym and setter-sym arguments.
The update expander writes local functions or macros under these names: a getter function and a setter function. Usually, update expanders write macros rather than functions, possibly in combination with some lexical anonymous variables which hold temporary objects. Therefore the getter and setter are henceforth referred to as macros.
The code being generated is with regard to some concrete instance of place-form. This argument is the actual form which occurs in a program. For instance, the update expander for the car place might be called with an arbitrary variant of the place-form which might look like (car (inc (third some-list))).
In the abstract semantics, upfront code wrapped around the body-form by the update expander provides the logic to evaluate this place to a location, which is retained in some hidden local context.
The getter local macro named by getter-sym must provide the logic for retrieving the value of this place. The getter macro takes no arguments. The body-form makes free use of the getter function; they may call it multiple times, which must not trigger multiple evaluations of the original place form.
The setter local macro named by setter-sym must generate the logic for storing a new value into the once-evaluated version of place-form. The setter function takes exactly one argument, whose value specifies the value to be stored into the place. It is the caller's responsibility to ensure that the argument form which produces the value to be stored via the setter is evaluated only once, and in the correct order. The setter does not concern itself with this form. Multiple calls to the setter can be expected to result in multiple evaluations of its argument. Thus, if necessary, the caller must supply the code to evaluate the new value form to a temporary variable, and then pass the temporary variable to the setter. This code can be embedded in the body-form or can be added to the code returned by a call to the update expander.
The setter local macro or function must return the new value which is stored. That is to say, when body-form invokes this local macro or function, it may rely on it yielding the new value which was stored, as part of achieving its own semantics.
The update expander does not macro-expand place-form. It is assumed that the expander is invoked in such a way that the place has been expanded in the correct environment. In other words, the form matches the type of place which the expander handles. If the expander had to macro-expand the place form, it would sometimes have to come to the conclusion that the place form must be handled by a different expander. No such consideration is the case: when an expander is called on a form, that is final; it is certain that it is the correct expander, which matches the symbol in the car position of the form, which is not a macro in the context where it occurs.
An update expander is free to assume that any place which is stored (the setter local macro is invoked on it) is accessed at least once by an invocation of the getter. A place update macro which relies on an update expander, but uses only the store macro, might not work properly. An example of an update expander which relies on this assumption is the expander for the (force promise) place type. If promise has not yet been forced, and only the setter is used, then promise might remain unforced as its internal value location is updated. A subsequent access to the place will incorrectly trigger a force, which will overwrite the value. The expected behavior is that storing a value in an unforced force place changes the place to forced state, preempting the evaluation of the delayed form. Afterward, the promise exhibits the value which was thus assigned.
The update expander is not responsible for all issues of evaluation order. A place update macro may consist of numerous places, as well as numerous value-producing forms which are not places. Each of the places can provide its registered update expander which provides code for evaluating just that place, and a means of accessing and storing the values. The place update macro must call the place expanders in the correct order, and generate any additional code in the correct order, so that the macro achieves its required documented evaluation order.
;; First, capture the update expander
;; function for (car ...) places
;; in a variable, for clarity.
(defvar car-update-expander [*place-update-expander* 'car])
;; Next, call it for the place (car [a 0]).
;; The body form specifies logic for
;; incrementing the place by one and
;; returning the new value.
(call car-update-expander 'getit 'setit '(car [a 0])
'(setit (+ (getit) 1)))
;; --> Resulting code:
(rlet ((#:g0032 [a 0]))
(macrolet ((getit nil
(append (list 'car) (list '#:g0032)))
(setit (val)
(append (list 'sys:rplaca)
(list '#:g0032) (list val))))
(setit (+ (getit) 1))))
;; Same expander call as above, with a call to expand added
;; to show the fully expanded version of the returned code,
;; in which the ;; setit and getit calls have disappeared,
;; replaced by their macro-expansions.
(expand
(call car-update-expander 'getit 'setit '(car [a 0])
'(setit (+ (getit) 1))))
;; --> Resulting code:
(let ((#:g0032 [a 0]))
(sys:rplaca #:g0032 (+ (car #:g0032) 1)))
The main noteworthy points about the generated code are:
(lambda (simple-setter-sym place-form
body-form) ...)
The clobber expander is a code-writer similar to the update expander. It takes a body-form argument, and returns a larger form which surrounds this form with additional program code.
The returned block of code has one main abstract action. It must arrange for the evaluation of body-form in a lexical environment in which a lexical macro or lexical function exists which has the name requested by the simple-setter-sym argument.
The simple setter local macro written by the clobber expander is similar to the local setter written by the update expander. It has exactly the same interface, performs the same action of storing a value into the place, and returns the new value.
The difference is that its logic may be considerably simplified by the assumption that the place is being subject to exactly one store, and no access.
A place update macro which uses a clobber expander, and calls it more than once, break the assumption; doing so may result in multiple evaluations of the place-form.
(lambda (deleter-sym place-form
body-form) ...)
The delete expander is a code-writer similar to clobber expander. It takes a body-form arguments, and returns a larger form which surrounds this form with additional program code.
The returned block of code has one main abstract action. It must arrange for the evaluation of body-form in a lexical environment in which a lexical macro or lexical function exists which has the name requested by the deleter-sym argument.
The deleter macro written by the clobber expander takes no arguments. It may be called at most once. It returns the previous value of the place, and arranges for its obliteration, whatever that means for that particular kind of place.
(with-update-expander (getter setter) place env
body-form)
The with-update-expander macro evaluates the body-form argument, whose result is expected to be a Lisp form. The macro adds additional code around this code, and the result is returned. This additional code is called the place-access code.
The getter and setter arguments must be symbols. Over the evaluation of the body-form, these symbols are bound to the names of local functions which are provided in the place-access code.
The place argument is a form which evaluates to a syntactic place. The generated place-access code is based on this place.
The env argument is a form which evaluates to a macro-expansion-time environment. The with-update-expander macro uses this environment to perform macro-expansion on the value of the place form, to obtain the correct update expander function for the fully macro-expanded place.
The place-access code is generated by calling the update expander for the expanded version of place.
The following is an implementation of the swap macro, which exchanges the contents of two places.
Two places are involved, and, correspondingly, the with-update-expander macro is used twice, to add two instances of place-update code to the macro's body.
(defmacro swap (place-0 place-1 :env env)
(with-gensyms (tmp)
(with-update-expander (getter-0 setter-0) place-0 env
(with-update-expander (getter-1 setter-1) place-1 env
^(let ((,tmp (,getter-0)))
(,setter-0 (,getter-1))
(,setter-1 ,tmp))))))
The basic logic for swapping two places is contained in the code template:
^(let ((,tmp (,getter-0)))
(,setter-0 (,getter-1))
(,setter-1 ,tmp))
The temporary variable named by the gensym symbol tmp is initialized by calling the getter function for place-0. Then the setter function of place-0 is called in order to store the value of place-1 into place-0. Finally, the setter for place-1 is invoked to store the previously saved temporary value into that place.
The name for the temporary variable is provided by the with-gensyms macro, but establishing the variable is the caller's responsibility; this is seen as an explicit let binding in the code template.
The names of the getter and setter functions are similarly provided by the with-update-expander macros. However, binding those functions is the responsibility of that macro. To achieve this, it adds the place-access code to the code generated by the ^(let ...) backquote template. In the following example macro-expansion, the additional code added around the template is seen. It takes the form of two macrolet binding blocks, each added by an invocation of with-update-expander:
(macroexpand '(swap a b))
-->
(macrolet ((#:g0036 () 'a) ;; getter macro for a
(#:g0037 (val-expr) ;; setter macro for a
(append (list 'sys:setq) (list 'a)
(list val-expr))))
(macrolet ((#:g0038 () 'b) ;; getter macro for b
(#:g0039 (val-expr) ;; setter macro for b
(append (list 'sys:setq) (list 'b)
(list val-expr))))
(let ((#:g0035 (#:g0036))) ;; temp <- a
(#:g0037 (#:g0038)) ;; a <- b
(#:g0039 #:g0035)))) ;; b <- temp
In this expansion, for example #:g0036 is the generated symbol which forms the value of the getter-0 variable in the swap macro. The getter is a macro which simply expands to a a: straightforward access to the variable a. The #:g0035 symbol is the value of the tmp variable. Thus the swap macro's ^(let ((,tmp (,getter-0))) ...) has turned into ^(let ((#:g0035 (#:g0036))) ...)
A full expansion, with the macrolet local macros expanded out:
(expand '(swap a b))
-->
(let ((#:g0035 a))
(sys:setq a b)
(sys:setq b #:g0035))
In other words, the original syntax (,getter-0) became (#:g0036) and finally just a.
Similarly, (,setter-0 (,getter-1)) became the macrolet invocations (#:g0037 (#:g0038)) which finally turned into: (sys:setq a b).
(with-clobber-expander (simple-setter) place env
body-form)
The with-clobber-expander macro evaluates body-form, whose result is expected to be a Lisp form. The macro adds additional code around this form, and the result is returned. This additional code is called the place-access code.
The simple-setter argument must be a symbol. Over the evaluation of the body-form, this symbol is bound to the name of a functions which are provided in the place-access code.
The place argument is a form which evaluates to a syntactic place. The generated place-access code is based on this place.
The env argument is a form which evaluates to a macro-expansion-time environment. The with-clobber-expander macro uses this environment to perform macro-expansion on the value of the place form, to obtain the correct update expander function for the fully macro-expanded place.
The place-access code is generated by calling the update expander for the expanded version of place.
The following implements a simple assignment statement, similar to set except that it only handles exactly two arguments:
(defmacro assign (place new-value :env env)
(with-clobber-expander (setter) place env
^(,setter ,new-value)))
Note that the correct evaluation order of place and new-value is taken care of, because with-clobber-expander generates the code which performs all the necessary evaluations of place. This evaluation occurs before the code which is generated by ^(,setter ,new-value) part is evaluated, and that code is what evaluates new-value.
Suppose that a macro were desired which allows assignment to be notated in a right to left style, as in:
(assign 42 a) ;; store 42 in variable a
Now, the new value must be evaluated prior to the place, if left-to-right evaluation order is to be maintained. The standard push macro has this property: the push value is on the left, and the place is on the right.
Now, the code has to explicitly take care of the order, like this:
;; WRONG! We can't just swap the parameters;
;; place is still evaluated first, then new-value:
(defmacro assign (new-value place :env env)
(with-clobber-expander (setter) place env
^(,setter ,new-value)))
;; Correct: arrange for evaluation of new-value first,
;; then place:
(defmacro assign (new-value place :env env)
(with-gensym (tmp)
^(let ((,tmp ,new-value))
,(with-clobber-expander (setter) place env
^(,setter ,tmp)))))
(with-delete-expander (deleter) place env
body-form)
The with-delete-expander macro evaluates body-form, whose result is expected to be a Lisp form. The macro adds additional code around this code, and the resulting code is returned. This additional code is called the place-access code.
The deleter argument must be a symbol. Over the evaluation of the body-form, this symbol is bound to the name of a functions which are provided in the place-access code.
The place argument is a form which evaluates to a syntactic place. The generated place-access code is based on this place.
The env argument is a form which evaluates to a macro-expansion-time environment. The with-delete-expander macro uses this environment to perform macro-expansion on the value of the place form, to obtain the correct update expander function for the fully macro-expanded place.
The place-access code is generated by calling the update expander for the expanded version of place.
The following implements the del macro:
(defmacro del (place :env env)
(with-delete-expander (deleter) place env
^(,deleter)))
(call-update-expander getter setter place env
body-form)
The call-update-expander function provides an alternative interface for making use of an update expander, complementary to with-update-expander.
Arguments getter and setter are symbols, provided by the caller. These are passed to the update expander function, and are used for naming local functions in the generated code which the update expander adds to body-form.
The place argument is a place which has not been subject to macro-expansion. The call-update-expander function takes on the responsibility for macro-expanding the place.
The env parameter is the macro-expansion environment object required to correctly expand place in its original environment.
The body-form argument represents the source code of a place update operation. This code makes references to the local functions whose names are given by getter and setter. Those arguments allow the update expander to write these functions with the matching names expected by body-form.
The return value is an object representing source code which incorporates the body-form, augmenting it with additional code which evaluates place to determine its location, and provides place accessor local functions expected by the body-form.
The following shows how to implement a with-update-expander macro using call-update-expander:
(defmacro with-update-expander ((getter setter)
unex-place env body)
^(with-gensyms (,getter ,setter)
(call-update-expander ,getter ,setter
,unex-place ,env ,body)))
Essentially, all that with-update-expander does is to choose the names for the local functions, and bind them to the local variable names it is given as arguments. Then it calls call-update-expander.
Implement the swap macro using call-update-expander:
(defmacro swap (place-0 place-1 :env env)
(with-gensyms (tmp getter-0 setter-0 getter-1 setter-1)
(call-update-expander getter-0 setter-0 place-0 env
(call-update-expander getter-1 setter-1 place-1 env
^(let ((,tmp (,getter-0)))
(,setter-0 (,getter-1))
(,setter-1 ,tmp))))))
(call-clobber-expander simple-setter place env
body-form)
The call-clobber-expander function provides an alternative interface for making use of a clobber expander, complementary to with-clobber-expander.
Argument simple-setter is a symbol, provided by the caller. It is passed to the clobber expander function, and is used for naming a local function in the generated code which the update expander adds to body-form.
The place argument is a place which has not been subject to macro-expansion. The call-clobber-expander function takes on the responsibility for macro-expanding the place.
The env parameter is the macro-expansion environment object required to correctly expand place in its original environment.
The body-form argument represents the source code of a place update operation. This code makes references to the local function whose name is given by simple-setter. That argument allows the update expander to write this function with the matching name expected by body-form.
The return value is an object representing source code which incorporates the body-form, augmenting it with additional code which evaluates place to determine its location, and provides the clobber local function to the body-form.
(call-delete-expander deleter place env body-form)
The call-delete-expander function provides an alternative interface for making use of a delete expander, complementary to with-delete-expander.
Argument deleter is a symbol, provided by the caller. It is passed to the delete expander function, and is used for naming a local function in the generated code which the update expander adds to body-form.
The place argument is a place which has not been subject to macro-expansion. The call-delete-expander function takes on the responsibility for macro-expanding the place.
The env parameter is the macro-expansion environment object required to correctly expand place in its original environment.
The body-form argument represents the source code of a place delete operation. This code makes references to the local function whose name is given by deleter. That argument allows the update expander to write this function with the matching name expected by body-form.
The return value is an object representing source code which incorporates the body-form, augmenting it with additional code which evaluates place to determine its location, and provides the delete local function to the body-form.
(define-modify-macro name parameter-list function-name)
The define-modify-macro macro provides a simplified way to write certain kinds of place update macros. Specifically, it provides a way to write place update macros which modify a place by retrieving the previous value, pass it through a function (perhaps together with some additional arguments), and then store the resulting value back into the place and return it.
The name parameter specifies the name for the place update macro to be written.
The function-name parameter must specify a symbol: the name of the update function.
The update macro and update function both take at least one parameter: the place to be updated, and its value, respectively.
The parameter-list specifies the additional parameters for the update function, which will also become additional parameters of the macro. Because it is a function parameter list, it cannot use the special destructuring features of macro parameter lists, or the :env or :whole special parameters. It can use optional parameters, and may be empty.
The define-modify-macro macro writes a macro called name. The leftmost parameter of this macro is a place, followed by the additional arguments specified by parameter-list. The macro will arrange for the evaluation of the place argument to determine the place location. It will then retrieve and save the prior value of the place, and evaluate the remaining arguments. The prior value of the place, and the values of the additional arguments, are all passed to function and the resulting value is then stored back into the location previously determined for place.
Some standard place update macros are implementable using define-modify-macro, such as inc.
The inc macro reads the old value of the place, then passes it through the + (plus) function, along with an extra argument: the delta value, which defaults to one. The inc macro could be written using define-modify-macro as follows:
(define-modify-macro inc (: (delta 1)) +)
Note that the argument list (: (delta 1)) doesn't specify the place, because the place is the implicit leftmost argument of the macro which isn't given a name. With the above definition in place, when (inc (car a)) is invoked, then (car a) is first reduced to a location, and that location's value is retrieved and saved. Then the delta parameter s evaluated to its value, which has defaulted to 1, since the argument was omitted. Then these two values are passed to the + function, and so 1 is added to the value previously retrieved from (car a). The resulting sum is then stored back (car a) without evaluating (car a) again.
(defplace place-destructuring-args body-sym
(getter-sym setter-sym update-body)
[(ssetter-sym clobber-body)
[(deleter-sym delete-body)]])
The defplace macro is used to introduce a new kind of syntactic place. It writes the update expander, and optionally clobber and delete expander functions, from a simpler, more compact specification, and automatically registers the resulting functions. The compact specification of a defplace call contains only code fragments for the expander functions.
The name and syntax of the place is determined by the place-destructuring-args argument, which is macro-style parameter list whose structure mimics that of the place. In particular, its leftmost symbol gives the name under which the place is registered. The defplace macro provides automatic destructuring of the syntactic place, so that the expander code fragments can refer to the components of a place by name.
The body-sym parameter must be be a symbol. This symbol will capture the body-forms parameter which is passed to the update expander, clobber expander or delete expander. The code fragments then have access to the body forms via this name.
The getter-sym, setter-sym, and update-body parenthesized triplet specify the update expander fragment. The defplace macro will bind getter-sym and setter-sym to symbols. The update-body must then specify a template of code which evaluates the syntactic place to determine its storage location, and provides a pair of local functions, using these two symbols as their name. The template must also insert the body-sym forms into the scope of these local functions, and the place determining code.
The setter-sym and clobber-body arguments similarly specify an optional clobber expander fragment, as a single optional argument. If specified, the clobber-body must generate a local function named using setter-sym wrapped around body-sym forms.
The deleter-sym and deleter-body likewise specify a delete expander fragment. If this is omitted, then the place shall not support deletion.
Implementation of the place denoting the car field of cons cells:
(defplace (car cell) body
;; the update expander fragment
(getter setter
(with-gensyms (cell-sym) ;; temporary symbol for cell
^(let ((,cell-sym ,cell)) ;; evaluate place to cell
;; getter and setter access cell via temp var
(macrolet ((,getter ()
^(car ,',cell-sym))
(,setter (val)
^(sys:rplaca ,',cell-sym ,val)))
;; insert body form from place update macro
,body))))
;; clobber expander fragment: simpler: no need
;; to evaluate cell to temporary variable.
(ssetter
^(macrolet ((,ssetter (val)
^(sys:rplaca ,',cell ,val)))
,body))
;; deleter: delegate to pop semantics:
;; (del (car a)) == (pop a).
(deleter
^(macrolet ((,deleter () ^(pop ,',cell)))
,body)))
(defset name params new-val-sym set-form)
(defset get-fun-sym set-fun-sym)
The defset macro provides a mechanism for introducing a new kind of syntactic place. It is simpler to use than defplace and more concise, but not as general.
The defset macro is designed for situations in which a function or macro which evaluates all of its arguments is required to serve as a syntactic place. It provides two flavors of syntax: the long form, indicated by giving defset five arguments, and a short form, which uses two arguments.
In the long form of defset, the syntactic place is described by name and params. The defset form expresses the request that a call to the function or operator named name be treated as a syntactic place, which has arguments described by the parameter list params.
The set-form argument specifies an expression which generates the code for storing a new value to the place.
The defset macro makes the necessary arrangements such that when an operator form named by name is treated as a syntactic place, then at macro-expansion time, code is generated to evaluate all of its argument expressions into machine-generated variables. The names of those variables are automatically bound to the corresponding symbols given in the params argument list of the defset syntax. Code is also generated to evaluate the expression which gives the new value to be stored, and that is bound to a generated variable whose name is bound to the new-val-sym symbol. Then arrangements are made to invoke the operator named by name and to evaluate the set-form in an environment in which these symbol bindings are visible. The operator named name is invoked using an altered argument list which uses temporary symbols in place of the original expressions. The task of set-form is to insert the values of the symbols from params and new-val-sym into a suitable code templates that will perform the store actions. The code generated by set-form must also take on the responsibility of yielding the new value as its result.
If params list contains optional parameters, the default value expressions of those parameters shall be evaluated in the scope of the defset definition.
The params list may specify a rest parameter. In the expansion, this parameter will capture a list of temporary symbols, corresponding to the list of variadic argument expressions. For instance if the defset parameter list for a place g is (a b . c), featuring the rest parameter c, and its set-form is ^(s ,a ,b ,*c) and the place is invoked as (g (i) (j) (k) (l)) then parameter c will be bound to a list of gensyms such as (#:g0123 #:g0124) so that the evaluation of set-form will yield syntax resembling (s #:g0121 #:g0122 #:g0123 #:g0124). Here, gensyms #:g0123 and #:g0124 are understood to be bound to the values of the expressions (k) and (l), the two trailing parameters corresponding to the rest parameter c.
Syntactic places defined by defset that have a rest parameter may be invoked with improper syntax such as (set (g x y . z) v). In this situation, that rest parameter will be bound to the name of a temporary variable which holds the value of z rather than to a list of temporary variable names holding the values of trailing expressions. The set-form must be prepared for this situation. In particular, the rest parameter's value is an atom, then it cannot be spliced in the backquote syntax, except at the last position of a list.
Although syntactic places defined by defset perform macro-parameter-like destructuring of the place form, binding unevaluated argument expressions to the parameter symbols, nested macro parameter lists are not supported: params specifies a function parameter list.
The parameter list may use parameter macros, keeping in mind that the parameter expansion is applied at the time the defset form is processed, specifying an expanded parameter list which receives unevaluated expressions. The set-form may refer to all symbols produced by parameter list expansion, other than generated symbols. For instance, if a parameter list macro :addx exists which adds the parameter symbol x to the parameter list, and this :addx is invoked in the params list of a defset, then x will be visible to the set-form.
The short, two-argument form of defset simply specifies the names of two functions or operators: get-fun-sym names the operator which accesses the place, and set-fun-sym names the operator which stores a new value into the place. It is expected that all arguments of these operators are evaluated expressions, and that the store operator takes one argument more than the access operator. The operators are otherwise assumed to be variadic: each instance of a place based on get-fun-sym individually determines how many arguments are passed to that operator and to the one named by set-fun-sym.
The definition (defset g s) means that (inc (g x y)) will generate code which ensures that x and y are evaluated exactly once, and then those two values are passed as arguments to g which returns the current value of the place. That value is then incremented by one, and stored into the place by calling the s function/operator with three arguments: the two values that were passed to g and the new value. The exact number of arguments is determined by each individual use of g as a place; the defset form doesn't specify the arity of g and s, only that s must accept one more argument relative to g.
The following equivalence holds between the short and long forms:
(defset g s) <--> (defset g (. r) n ^(g ,*r) ^(s ,*r ,n))
Note: the short form of defset is similar to the define-accessor macro.
Implementation of car as a syntactic place using a long form defset:
(defset car (cell) new
(let ((n (gensym)))
^(rlet ((,n ,new))
(progn (rplaca ,cell ,n) ,n))))
Given such a definition, the expression (inc (car (abc))) expands to code closely resembling:
(let ((#:g0048 (abc)))
(let ((#:g0050 (succ (car #:g0048))))
(rplaca #:g0048 #:g0050)
#:g0050))
The defset macro has arranged for the argument expression (abc) of car to be evaluated to a temporary variable #:g0048, a gensym. This, then, holds the cons cell being operated on. At macro-expansion time, the variable cell from the parameter list specified by the defset is bound to this symbol. The access expression (car #:0048) to retrieve the prior value is automatically generated by combining the name of the place car with the gensym to which its argument (abc) has been evaluated. The new variable was bound to the expression giving the new value, namely (succ (car #:g0048)). The set-form is careful to evaluate this only one time, storing its value into the temporary variable #:g0050, referenced by the variable n. The set-form's (rplaca ,cell ,n) fragment thus turned into (rplaca #:g0048 #:g0050) where #:g0048 references the cons cell being operated on, and #:g0050 the calculated new value to be stored into its car field. The set-form is careful to arrange for the new value #:g0050 to be returned. Those place-mutating operators which yield the new value, such as set and inc rely on this behavior.
(define-place-macro name macro-style-params
body-form*)
In some situations, an equivalence exists between two forms, only one of which is recognized as a place. The define-place-macro macro can be used to establish a form as a place in terms of a translation to an equivalent form which is already a place.
The define-place-macro has the same syntax as defmacro. It specifies a macro transformation for a compound form which has the name symbol in its leftmost position.
Place macro expansion doesn't use an environment; place macros are in a single global namespace, special to place macros. There are no lexically scoped place macros. Such an effect can be achieved by having a place macro expand to an a form which is the target of a global or local macro, as necessary.
To support place macros, forms which are used as syntactic places are subject to a modified macro-expansion algorithm:
The define-place-macro macro does not cause name to become mboundp.
There can exist both an ordinary macro and a place macro of the same name. In this situation, when the macro call appears as a place form, it is expanded as a place macro, according to the above steps. When the macro call appears as an evaluated form, not being used as a place, the form is expanded using the ordinary macro.
Implementation of first in terms of car:
(define-place-macro first (obj)
^(car ,obj))
(macroexpand-1-place form [env])
(macroexpand-place form [env])
If form is a place macro form (a form whose operator symbol has been defined as a place macro using define-place-macro) these functions expand the place macro form and return the expanded form. Otherwise, they return form.
macroexpand-1-place performs a single expansion, expanding only the place the macro that is referenced by the symbol in the first position of form, and returns the expansion. Note that if form is an ordinary macro form, this function will not expand it, even if such an expansion would reveal a place macro form.
macroexpand-place performs a full place expansion of form by the following process. If form is a place macro call, it is expanded, and the result is checked again to see whether it is a place macro, and expanded. This is repeated as many times as necessary until the result is no longer a place macro call. Then, if the resulting form is an ordinary macro invocation, it is expanded once as if by macroexpand-1. This process is iterated until a fixed point is reached.
The optional env parameter is a macro environment. Note: the macroexpand-1-place function ignores the env parameter, which could change in the future.
Given this ordinary macro definition
(defmacro leftmost (x) ^(first ,x))
the following results are obtained:
;; ordinary macro leftmost expands to first,
;; then first place macro expands to car:
(macroexpand-place '(leftmost x)) -> (car x)
;; macroexpand-1-place won't expand ordinary macro:
(macroexpand-1-place '(leftmost x)) -> (leftmost x)
;; macroexpand-1-place expands place macro
(macroexpand-1-place '(first x)) -> (car x)
(rlet ({(sym init-form)}*) body-form*)
The macro rlet is similar to the let operator. It establishes bindings for one or more syms, which are initialized using the values of init-forms.
Note that the simplified syntax for a variable which initializes to nil by default is not supported by rlet; that is to say, the syntax sym cannot be used in place of the (sym init-form) syntax when sym is to be initialized to nil.
The rlet macro differs from let in that rlet assumes that those syms whose init-forms, after macro expansion, are constant expressions (according to the constantp function) may be safely implemented as a symbol macro rather than a lexical variable.
Therefore rlet is suitable in situations in which simpler code is desired from the output of certain kinds of machine-generated code, which binds local symbols: code with fewer temporary variables.
On the other hand, rlet is not suitable in some situations when true variables are required, which are assignable, and provide temporary storage.
;; WRONG! Real storage location needed.
(rlet ((flag nil))
(flip flag)) ;; error: flag expands to nil
;; Demonstration of constant-propagation
(let ((a 42))
(rlet ((x 1)
(y a))
(+ x y))) --> 43
(expand
'(let ((a 42))
(rlet ((x 1)
(y a))
(+ x y)))) --> (let ((a 42))
(let ((y a))
(+ 1 y)))
The last example shows that the x variable has disappeared in the expansion. The rlet macro turned it into into a symacrolet denoting the constant 1, which then propagated to the use site, turning the expression (+ x y) into (+ 1 y).
(slet ({(sym init-form)}*) body-form*)
The macro slet is a stronger form of the rlet macro. Just like rlet, slet reduces bindings initialized by constant expressions to symbol macros. In addition, unlike rlet, slet also reduces to symbol macros those bindings whose initializing expressions are simple references to lexical variables.
;; reduces to let
(slet ((a (list x y)))
a)
;; b is a free variable, so this is also let
(slet ((a b))
a)
;; b is lexical, so a becomes a symbol macro
;; the (slet ...) form becomes b.
(let (b)
(slet ((a b))
a))
;; a becomes symbol macro; form transforms to 1.
(slet ((a 1))
a)
(alet ({(sym init-form)}*) body-form*)
The macro alet ("atomic" or "all") is a stronger form of the slet macro. All bindings initialized by constant expressions are turned to symbol macros. Then, if all of the remaining bindings are all initialized by lexical variables, they are also turned to symbol macros. Otherwise, none of the remaining bindings are turned to symbol macros.
The alet macro can be used even in situations when it is possible that the initializing forms the variables may have side effects through which they affect each others' evaluations. In this situation alet still propagates constants via symbol macros, and can eliminate the remaining temporaries if they can all be made symbol macros for existing lexicals: i.e. there doesn't exist any initialization form with interfering side effects.
(define-accessor get-function set-function)
The define-accessor macro is used for turning a function into an accessor, such that forms which call the function can be treated as places.
Arguments to define-accessor are two symbols, which must name functions. When the define-accessor call is evaluated, the get-function symbol is registered as a syntactic place. Stores to the place are handled via calls to set-function.
If get-function names a function which takes N arguments, set-function must name a function which takes N+1 arguments.
Moreover, in order for the accessor semantics to be correct set-function must treat its rightmost argument as the value being stored, and must also return that value.
When a function call form targeting get-function is treated as a place which is subject to an update operation (for instance an increment via the inc macro), the accessor definition created by define-accessor ensures that the arguments of get-function are evaluated only once, even though the update involves a call to get-function and set-function with the same arguments. The argument forms are evaluated to temporary variables, and these temporaries are used as the arguments in the calls.
No other assurances are provided by define-accessor.
In particular, if get-function and set-function internally each perform some redundant calculation over their arguments, this cannot be optimized. Moreover, if that calculation has a visible effect, that effect is observed multiple times in an update operation.
If further optimization or suppression of multiple effects is required, the more general defplace macro must be used to define the accessor. It may also be possible to treat the situation in a satisfactory way using a define-place-macro definition, which effectively then supplies inline code whenever a certain form is used as a place, and that code itself is treated as a place.
Note: define-accessor is similar to the short form of defset.
(read-once expression)
(set (read-once place) new-value)
When the read-once accessor is invoked as a function, it behaves like identity, simply returning the value of expression, which is not required to be a syntactic place.
If a read-once form is used as a syntactic place then its argument must also be a place. The read-once syntactic place denotes the same place as the enclosed place form, but with somewhat altered semantics, which is most useful in conjunction with placelet, and in writing place-mutating macros which make multiple accesses to a place.
Firstly, if the read-once place is evaluated, it accesses the existing value of place exactly once, even if it occurs in a place-mutating form which normally doesn't use the prior value, such as the set macro.
When read-once accesses place, it stores the value in a hidden variable. Then, within the same place-mutating form, multiple references to the same read-once form all access the value of this hidden variable. Whenever the read-once form is assigned, both the the hidden variable and the underlying place receive the new value.
Multiple references to the same read-once form can be produced using the placelet or placelet* macros, or by making multiple calls to the getter function obtained using with-update-expander in the implementation of a user-defined place-mutating operator, or user-defined place.
In both of the following two examples, there is no question that the array and i expressions are themselves evaluated only once; the issue is the access to the array itself; under the plain placelet, the array referencing takes place more times.
;; without read-once, array element [array i] is
;; accessed twice to fetch its current value: once
;; in the plusp expression, and then once again in
;; the dec expression.
(placelet ((cell [array i]))
(if (plusp cell)
(dec cell)))
;; with read-once, it is accessed once. plusp refers
;; to a hidden lexical variable to obtain the prior
;; value, and so does dec. dec stores the new value
;; through to [array i] and the hidden variable.
(placelet ((cell (read-once [array i])))
(if (plusp cell)
(dec cell)))
The following is not an example of multiple references to the same read-once form:
(defmacro inc-positive (place)
^(if (plusp (read-once ,place))
(inc (read-once ,place))))
Here, even though the read-once forms may be structurally identical, they are separate instances. The first instance isn't even a syntactic place, but a call to the read-once function. Multiple references to the same place can only be generated using placelet or else by multiple explicit calls to the same getter function or macro generated for a place by an update expander.
The following is a corrected version of inc-positive:
(defmacro inc-positive (place :env env)
(with-update-expander (getter setter) ^(read-once ,place) env
^(if (plusp (,getter))
(,setter (succ (,getter))))))
To write the macro without read-once requires that it handles the job of providing a temporary variable for the value:
(defmacro inc-positive (place :env e)
(with-update-expander (getter setter) place env
(with-gensym (value)
^(slet ((,value (,getter)))
^(if (plusp ,value)
(,setter (succ ,value)))))))
The read-once accessor wrapped around place allows inc-positive to simply make multiple references to (,getter) which will cache the value; the macro doesn't have to introduce its own hidden caching variable.
These variables hold hash tables, by means of which update expanders, clobber expanders and delete expanders are registered, as associations between symbols and functions.
If [*place-update-expander* 'sym] yields a function, then symbol sym is the basis for a syntactic place. If the expression yields nil, then forms beginning with sym are not syntactic places. (The situation of a clobber accessor or delete accessor being defined without an update expander is improper).
The *place-macro* special variable holds the hash table of associations between symbols and place macro expanders.
If the expression [*place-macro* 'sym] yields a function, then symbol sym has a binding as a place macro. If that expression yields nil, then there is no such binding: compound forms beginning with sym do not undergo place macro expansion.
TXR Lisp provides a structural pattern-matching system. Structural pattern matching is a syntax which allows for the succinct expression of code which classifies objects according to their shape and content, and which accesses the elements within objects, or both.
The central concept in structural pattern matching is the resolution of a pattern against an object. The pattern is specified as syntax which is part of the program code. The object is a run-time value of unknown type, shape and other properties. The primary pattern-matching decision is Boolean: does the object match the pattern? If the object matches the pattern, then it is possible to execute an associated body of code in a scope in which variables occurring in the pattern take on values from the corresponding parts of the object.
Structural pattern matching is available via several different macro operators, which are: when-match, if-match, match, match-case, match-cond, match-ecase, lambda-match and defun-match. Function and macro argument lists may also be augmented with pattern matching using the :match parameter macro.
The when-match macro is the simplest. It tests an object against a pattern, and if there is a match, evaluates zero or more forms in an environment in which the pattern variables have bindings to the corresponding elements of the object.
The if-match macro evaluates a single form if there is a match, in the scope of the bindings established by the pattern, otherwise an alternative form evaluated in a scope in which those bindings are absent.
The match macro tests and object against a pattern, expecting a match. If the match fails, an exception is thrown. Otherwise, it evaluates zero or more forms in the scope of the bindings established by the pattern.
The match-case macro evaluates the same object against multiple clauses, each consisting of a pattern and zero or more forms. At most one matching clause is identified and evaluated.
The match-ecase macro is similar to match-case except that if no matching case is identified, an exception is thrown.
The match-cond macro evaluates multiple clauses, each of which specifies a pattern and an object expression. If the object produced by the expression matches the pattern, the forms in the clause are evaluated in scope of the variables bound by the clause's pattern.
The lambda-match macro provides a way to express an anonymous function whose argument list is matched against multiple clauses similarly to match-case and defun-match provides a way to define a top-level function using the same concept.
Additionally, there exist each-match and while-match macro families.
TXR Lisp's structural pattern-matching notation is template-based. With the exception of structures and hash tables, objects are matched using patterns which are based on their printed notation. For instance, the pattern (1 2 @a) is a pattern matching the list (1 2 3) binding a to 3. The notation supports lists, vectors, ranges and atoms. Atoms are compared using the equal function. Thus, in the above pattern, the 1 and 2 in the pattern match the corresponding 1 and 2 atoms in the object using equal.
All parts of a pattern are static material which matches literally, except those parts introduced by the meta prefix @. This prefix denotes variables like @a as well as useful pattern-matching operators like @(all pattern) which matches a list or sublist whose elements all match pattern.
The quasiquote syntax is specially supported for expressing matching, in an alternative style. For instance the quasiquote ^(1 2 ,a) is a pattern equivalent to the (1 2 @a).
Structure objects are matched using a dedicated @(struct name ...) operator, or else in the quasiquote style using ^#S(name ...) syntax. The non-quasiquoted literal syntax #S(name ...) cannot be used for matching.
Similarly, hash objects are matched using a @(hash ...) operator, or else ^#H(...) syntax in the quasiquote style. #H(...) cannot be used.
Note: the non-quasiquoted #S and #H literals are not and cannot be used for matching because they produce structure and hash objects which lose important information about how they were specified in the syntax, and carry restrictions which are unacceptable for pattern matching. The order of sub-patterns is important in pattern syntax, but struct and hash objects do not preserve the order in which their elements were specified. A struct literal is required to specify the name of an existing struct type, and slot names which are valid for that type, otherwise it is erroneous. This is not acceptable for pattern matching, because patterns may appear in place of those elements. The pattern match for a hash may specify the same key pattern more than once, which means that the key pattern cannot be an actual key in an actual hash, which requires every key to be unique. Structure and hash quasiquotes do not have these issues; they are not actually literal structure and hash objects, but list-based syntax.
Patterns use meta-symbols for denoting variables. Variables must be either bindable symbols, or else nil, which has a special meaning: the pattern variable @nil matches any object, and binds no variable.
Pattern variables are ordinary Lisp variables. Whereas in ordinary non-pattern matching Lisp code, it is always unambiguous whether a variable is being bound or referenced, this is deliberately not the case in patterns. A variable occurring in a pattern may be a fresh variable, or a reference to an existing one. The difference between these situations is not apparent from the syntax of the pattern; it depends on the context established by the scope.
With one exception, if a pattern contains a variable which is already bound in the surrounding scope, then it refers to that binding. Otherwise, it freshly binds the variable. The exception is that pattern operator @(as) always binds a fresh variable. A variable being already bound includes as a lexical or global symbol macro (symacrolet or defsymacro).
When a pattern variable refers to an existing variable, then each occurrence of that variable must match an object which is equal to the value of that variable. For instance, the following function returns the third element of a list, if the first two elements are repetitions of the x argument, otherwise nil:
(defun x-x-y (list x)
(when-match (@x @x @y) list y))
(x-x-y '(1 1 2) 1) -> 2
(x-x-y '(1 2 3) 1) -> nil ;; no @x @x match
(x-x-y '(1 1 2 r2) 1) -> nil ;; list too long
If the variable does not exist in the scope surrounding the pattern, then the leftmost occurrence of the variable establishes a binding, taking the value from is corresponding object being matched by that occurrence of the variable. The remaining occurrences of the variable, if any, must correspond to objects which are equal to that value, or else there is no match. For instance, the pattern (@a @a) matches the list like (1 1) as follows. First @a binds to the leftmost 1 and then the second 1 matches the existing value of that a. An input such as (1 2) fails to match because the second occurrence of @a retrieves an object that is not equal to that variable's existing value.
A pattern can contain multiple occurrences of the same symbol as a variable. These may or may not refer to the same variable. Two occurrences of the same symbol refer to distinct variables if:
Any other two or more occurrences same symbol occurring in the same pattern refer to the same variable.
TXR Lisp's macro-style parameter lists, appearing in tree-bind and related macros, also provide a form of structural pattern matching. Macro-style parameter list pattern matching is limited to objects of one kind: tree structures made of cons cells. It is only useful for matching on shape, not content. For example, tree-bind cannot express the idea of matching a list whose first element is the symbol a and whose third element is 42. Moreover, every position in the tree pattern much specify a variable which captures the corresponding element of the structure or else the symbol t to indicate that no variable is to be captured. There are no other pattern matching operators.
User-defined pattern operators are possible. When the operator symbol in the @(operator argument*) syntax doesn't match any built-in operator, a search takes place to determine whether operator is a pattern macro. If so, the pattern macro is expanded, and its result of the expansion treated as a pattern to process recursively, unless it is the original macro form, in which case it is treated as a predicate pattern. User-defined pattern macros are defined using the defmatch macro.
The pattern-matching notation is documented in the following sections; a section describing the pattern-matching macros follows.
The atom is not subject to evaluation, which means that a symbolic atom stands for itself, and not the value of a variable.
;; the pattern 1 matches the object 1
(if-match 1 1 'yes 'no) --> yes
;; the object 0 does not match
(if-match 1 0 'yes 'no) --> no
;; a matches a, does not match b
(let ((sym 'a))
(list (if-match a sym 'yes 'no)
(if-match b sym 'yes 'no)))
--> (yes no)
@symbol
A meta-symbol can be used as a pattern expression. This pattern unconditionally matches an object of any kind.
The symbol is required to be a either a bindable symbol according to the bindable function, or else the symbol nil.
If symbol is a bindable symbol, which has not binding in scope, then a variable by that name is freshly bound, and takes on the corresponding object as its value.
If symbol is a bindable symbol with an existing binding, then the corresponding object must be equal to that variable's existing value, or else the match fails.
If symbol is nil, then the match succeeds unconditionally, without binding a variable.
(when-match @a 42 (list a)) -> (42)
(when-match (@a @b @c) '(1 2 3) (list c b a)) -> (3 2 1)
;; No match: list is longer than pattern
(when-match (@a @b) '(1 2 3) (list a b)) -> nil
;; Use of nil in dot position to match longer list
(when-match (@a @b . @nil) '(1 2 3) (list a b)) -> (1 2)
(pattern+)
(pattern+ . pattern)
Pattern syntax consisting of a nonempty, possibly improper list matches list structure. A pattern expression may be specified in the dotted position. If it is omitted, then there is an implicit terminating nil which constitutes an atom expression matching nil.
A list pattern matches a list of the same shape. For each pattern expressions, there must exist an item in the list.
A match occurs when every pattern matches the corresponding element of the list, including the pattern in the dotted position.
Because the dotted position pattern matches a list, it is possible for a short pattern to match a longer list.
The syntax is indicated as requiring at least one pattern because otherwise the list is empty, which corresponds to the atom pattern nil.
The syntax (. pattern) is valid, but indistinguishable from pattern. It is only a list pattern if pattern is a list pattern.
(if-match (@a @b @c . @d) '(1 2 3 . 4) (list d c b a))
--> (4 3 2 1)
;; 2 doesn't satisfy oddp
(if-match (@(oddp @a) @b @c . @d) '(2 x y z)
(list a b c d)
:no-match)
--> :no-match
;; 1 and 2 match, a takes (3 4)
(if-match (1 2 . @a) '(1 2 3 4) a) --> (3 4)
;; nesting
(if-match ((1 2 @a) @b) '((1 2 3) 4) (list a b)) -> (3 4)
#(pattern*)
A pattern match for a vector is expressed using vector notation enclosing pattern expressions. This pattern matches a vector object which contains exactly as many elements as there are patterns. Each pattern is applied against the corresponding vector element.
;; empty vector pattern matches empty vector
(if-match #() #() :yes :no) -> :yes
;; empty vector pattern fails to match nonempty vector
(if-match #() #(1) :yes :no) -> :no
;; match with nested list and vector
(if-match #((1 @a) #(3 @b)) #((1 2) #(3 4)) (list a b))
--> (2 4)
#R(from-pattern to-pattern)
A pattern match for a range can be expressed by embedding pattern expressions in the #R notation. The resulting pattern requires the corresponding object to be a range, otherwise the match fails. If the corresponding object is a range, then the from-pattern is matched against its from and the to-pattern is matched against its to part.
Note that if the range expression notation a..b is used as a pattern, that is actually a list pattern, due to that being a syntactic sugar for (rcons a b).
(if-match #R(10 20) 10..20 :yes :no) -> :yes
(if-match #R(10 20) #R(10 20) :yes :no) -> :yes
(if-match #R(10 20) #R(1 2) :yes :no) -> :no
(when-match #R(@a @b) 1..2 (list a b)) -> (1 2)
;; not a range match! rcons syntax match
(when-match @a..@b '1..2 (list a b)) -> (1 2)
;; above, de-sugared:
(when-match (rcons @a @b) '(rcons 1 2) (list a b)) -> (1 2)
`...@var...`
@`...@var...`
The quasiliteral syntax is supported as a pattern-matching operator. The corresponding object is required to be a character string, which is analyzed according to the structure of the quasiliteral pattern, and portions of the string are captured by variables. If the corresponding object isn't a string according to stringp then the match fails. The quasiliteral pattern must match the entire input string.
In order that the quasiliteral's syntactic structure is not misinterpreted as a predicate pattern, and in order to make certain situations work in quasiquoted pattern matching, a quasiliteral pattern may be specified as either `...` or @`...`. The latter form, which is structurally (sys:expr (sys:quasi ...)) is specially recognized and treated as equivalent to the unadorned quasiliteral pattern.
A quasiliteral pattern matches in a linear fashion, from left to right. Variables bound earlier in the pattern can be referenced later in the pattern as bound variables.
With one exception, bound variables denote character strings in accordance with the usual quasiliteral conversion and formatting rules. All of the modifier notations may be used. For instance, if x is a bound variable, then @{x -40} denotes the value of x converted to a string, and right-aligned in a forty-character-wide field. Consequently, the notation matches exactly such a forty-character text. The exception is that if a bound variable has a regular expression modifier, as in @{x #/re/} then it has a special meaning as a pattern. Moreover, this syntax has no meaning in a quasiliteral.
In the following description of the quasiliteral pattern-matching rules, the symbols uv, uv0 and uv1 represent to unbound variables: variables which have no apparent lexical binding and are not defined as global variables. Unless indicated otherwise, @uv refers to a plain variable syntax such as @abc or else to braced syntax without modifiers, such as @{abc}. The same remarks apply to uv0 and uv1. The symbol bv represents a bound variable: a variable which has an existing binding, which can occur in the form of the ordinary notation, or the braced notation with or without modifiers. The notation {P}, {P0}, {P1}... denotes a substring of the pattern, possibly empty.
(when-match `@a-@b` "foo-bar" (list a b)) -> ("foo" "bar")
(when-match `@{a #/\d+/}@b` "123xy" (list a b)) -> ("123" "xy")
(let ((a 42))
(when-match `[@{a -8}] @b` "[ 42] packets` b))
-> "packets"
^qq-syntax
Quasiquoting provides an alternative pattern-matching syntax. It uses a subset of the quasiquoting notation. Only specific kinds of quasiquoted objects listed in this description are supported. Within a quasiquote used for pattern-matching, unquotes indicate operators and variables instead of the @ prefix. Splicing unquote syntax plays no role; its presence produces unspecified behavior.
The quasiquote matching notation is described, understood and implementing in terms of a translation to the standard pattern-matching syntax, according to the following rules. The [X] notation used here indicates that the element enclosed in brackets is subject to a recursive translation according to the rules:
;; basic unquote: variables embedded via unquote,
;; not requiring @ prefix.
(when-match ^(,a ,b) '(1 2) (list a b))
--> (1 2)
;; operators embedded via unquote; interior of operators
;; is regular non-quasiquoting pattern syntax.
(when-match ^(,(oddp @a) ,(evenp @b)) '(1 2) (list a b))
--> (1 2)
(when-match ^#(,a ,b) #(1 2) (list a b))
--> (1 2)
(when-match ^#S(,type year ,y) #S(time year 2021)
(list (struct-type-name type) y))
--> (time 2021)
(when-match ^#H(() (x ,y) (,(symbolp @y) ,datum))
#H(() (x k) (k 42))
datum)
--> (42)
;; JSON syntax
(when-match ^#J~a 42.0 a) --> 42.0
(when-match ^#J[~a, ~b] #J[true, false] (list a b)) --> (t nil)
(when-match ^#J{"x" : ~y, ~(symbolp @y) : ~datum}
#J{"x" : true, true : 42}
datum)
--> (42.0)
(when-match ^#J{"foo" : {"x" : ~val}}
#J{"foo" : {"x" : "y"}} val)
--> "y"
@(struct name {slot-name pattern}*)
@(struct pattern {slot-name pattern}*)
The struct pattern operator matches a structure object. The operator supports two modes of matching, the choice of which depends on whether the first argument is a name or a pattern.
The first argument is considered a name if it is a bindable symbol according to the bindable function. In this situation, the operator operates in strict mode. Otherwise, the operator is in loose mode.
The name or pattern argument is followed by zero or more slot-name pattern pairs, which are not enclosed in lists, similarly to the way slots are presented in the #S struct syntax and in the argument conventions of the new macro.
In strict mode, name is assumed to be the name of an existing struct type. The object being matched is tested whether it is a subtype of this type, as if using the subtypep function. If it isn't, the match fails.
In loose mode, the object being matched is tested whether it is a structure object of any structure type. If it isn't, the match fails.
In strict mode, each slot-name pattern pair requires that the object's slot of that name contain a value which matches pattern. The operator assumes that all the slot-names are slots of the struct type indicated by name.
In loose mode, no assumption is made that the object actually has the slots specified by the slot-name arguments. The object's structure type is inquired to determine whether it has each of those slots. If it doesn't, the match fails. If the object has the required slots, then the values of those slots are matched against the patterns.
In loose mode, the pattern given in the first argument position of the syntax is matched against the object's structure type: the type itself, rather than its symbolic name.
;; extract the month from a time structure
;; that is required to have a year of 2021.
(when-match @(struct time year 2021 month @m)
#S(time year 2021 month 1)
m) -> 1
;; match any structure with name and value slots,
;; whose name is foo, and extract the value.
(defstruct widget ()
name
value)
(defstruct grommet ()
name
value)
(append-each ((obj (list (new grommet name "foo" value :grom)
(new widget name "foo" value :widg))))
(when-match @(struct @type name "foo" value @v) obj
(list (list type v))))
--> ((#<struct-type grommet> :grom)
(#<struct-type widget> :widg))
@(hash {(key-pattern [value-pattern])}*)
The hash pattern operator matches a hash-table object by means of patterns which target keys, values or both.
An important concept in the requirements governing the operation of the hash operator is that of a trivial pattern.
A pattern is nontrivial if it is a variable or operator. A quasiliteral pattern that contains variables or operators is nontrivial. A pattern is also nontrivial if it is a list, vector, range or quasiquote pattern containing at least one nontrivial pattern. Otherwise, it is trivial.
The hash operator requires the corresponding object to be a hash table, otherwise the match fails.
If the corresponding object is a hash table, then the operator matches each key-pattern and value-pattern pair against that object as described below. Each of the pairs must successfully match, otherwise the overall match fails.
The following requirements apply to key-value pattern pairs in which the value pattern is specified.
If key-pattern is a trivial pattern, then the semantics of the match is that key-pattern is taken as a literal object representing a hash key. The hash table is searched for that key. If the key is not found, the match fails. Otherwise, the value corresponding to that key is matched against the value-pattern which may be trivial or nontrivial.
If key-pattern is a simple variable pattern @sym and if sym has an existing binding, then the value of sym is looked up in the hash table. If it is not found, then the match fails, otherwise the corresponding value is matched against value-pattern, which may be trivial or nontrivial.
If key-pattern is a nontrivial pattern other than a variable pattern for a variable which has an existing binding, and if value-pattern is trivial, then value-pattern is taken as a literal object, which is used for searching the hash table for one or more keys, as if it were the value argument in a call to the hash-keys-of function, to find all keys which have a value equal to that value. If no keys are found, then the match fails. Otherwise, the key-pattern is then matched against the retrieved list of hash keys.
Finally, if both key-pattern and value-pattern are nontrivial, then an exhaustive search is performed of the hash table. Every key in the hash table is matched against key-pattern and if it matches, the value is matched against value-pattern. If both match, then the values from the matches are collected into lists. At least one matching key-value pair must be found, otherwise the overall match fails. Note: this situation can be understood as if the hash table were an association list of cons cells of the form
(
key . value)
and as if the two patterns were combined into a coll operator against this list in the following way:
@(coll (
key-pattern . value-pattern))
such that the semantics can then be understood in terms of the coll operator matching against an association list.
The following requirements apply when the value-pattern is omitted.
If key-pattern is a nontrivial pattern other than a variable pattern for a variable which has an existing binding, then the pattern is applied against the list of keys from the hash table, which are retrieved as if using the hash-keys function.
If key-pattern is a variable pattern referring to an existing binding, then that pattern is taken as a literal object. The match is successful if that object occurs as a key in the hash table.
;; First, (x @y) has a trivial key pattern so the x
;; entry from the hash table is retrieved, the
;; value being the symbol k. This k is bound to @y.
;; Because y now a bound variable the pattern (@y @datum)
;; is interpreted as search of the hash table for
;; a single entry matching the value of @y. This
;; is the k entry, whose value is 42. The @datum
;; value match takes this 42.
(when-match @(hash (x @y) (@y @datum))
#H(() (x k) (k 42)) datum)
--> 42
;; Again, (x @y) has a trivial key pattern so the x
;; entry from the hash table is retrieved, the
;; value being the symbol k. This k is bound to @y.
;; This time the second pattern has a @(symbolp)
;; predicate operator. This is not a variable, and
;; so the pattern searches the entire
;; hash table. The @y variable has a binding to k,
;; so only the (k 42) entry is matched. The 42
;; value matches @datum, and is collected into a list.
(when-match @(hash (x @y) (@(symbolp @y) @datum))
#H(() (x k) (k 42)) datum)
--> (42)
@(as name pattern)
The as pattern operator binds the corresponding object to a fresh variable given by name, similarly to the Lisp let operator. If another variable called name exists, it is shadowed; thus, no back-referencing is performed.
The name argument must be a bindable symbol, or else nil. If name is nil, then no name is bound. Thus @(as nil pattern) is equivalent to pattern. Otherwise, pattern processed in a scope in which the new name binding is already visible.
The as operator succeeds if pattern matches.
Note: in a situation when it is necessary to bind a variable to an object in parallel with one or more patterns, such that the variable can back-reference to an existing occurrence, the and pattern operator can be used.
;; w captures the entire (1 2 3) list:
(when-match @(as w (@a @b @c)) '(1 2 3) (list w a b c))
--> ((1 2 3) 1 2 3)
;; match a list which has itself as the third element
(when-match @(as a (1 2 @a 4)) '#1=(1 2 #1# 4) :yes)
--> :yes
@(with [main-pattern] {side-pattern | name} expr)
The with pattern operator matches the optional main-pattern against a corresponding object, while matching a side-pattern or name against the value of the expression expr which is embedded in the syntax.
First, if main-pattern is present in the syntax, it is matched against its corresponding object. This match must succeed, or else the with operator fails to match, in which case expr is not evaluated.
Next, if main-pattern successfully matched, or is absent, expr is evaluated in the scope of earlier pattern variables, including any which that emanate from main-pattern. It is unspecified whether later pattern variables are visible.
Finally, side-pattern is matched against the value of expr. If that succeeds, then the operator has successfully matched.
If a name is specified instead of a side-pattern, it must be a bindable symbol or else nil.
(when-match (@(with @a x 42) @b @c) '(1 2 3) (list a b c x))
--> (1 2 3 42)
(let ((o 3))
(when-match (@(evenp @x) @(with @z @(oddp y) o)) '(4 6)
(list x y z)))
--> (4 3 6)
@(require pattern condition*)
The pattern operator require applies the specified pattern to the corresponding object. If the pattern matches, the operator then imposes the additional constraints specified by zero or more condition forms. Each condition is evaluated in a scope in which the variables from pattern have already been established.
For the require operator to be a successful match, every condition must evaluate true, otherwise the match fails.
The condition forms behave as if they were the arguments of an implicit and operator, which implies left-to-right evaluation behavior, stopping evaluation on the first condition which produces nil, and defaulting to a result of t when no condition forms are specified.
;; Match a (+ a b) expression where a and b are similar:
(when-match @(require (+ @a @b) (equal a b)) '(+ z z) (list a b))
--> (z z)
;; Mismatched case
(if-match @(require (+ @a @b) (equal a b)) '(+ y z)
(list a b)
:no-match)
--> :no-match
@(all pattern)
@(all* pattern)
The all and all* pattern operators require the corresponding object to be a sequence.
The specified pattern is applied against every element of the sequence. The match is successful if pattern matches every element.
Furthermore, in the case of a successful match, each variable that is freshly bound by pattern is converted into a list of all of the objects which that variable encounters from all elements of the sequence. Those variables which already have a binding from another pattern are not converted to lists. Their existing values are merely required to match each corresponding object they encounter.
The difference between all and all* is as follows. The all operator respects the vacuous truth of the match when the sequence is empty. In that case, the match is successful, and the variables are all bound to the empty list nil. In contrast, the alternative all* operator behaves like a failed match when the sequence is empty.
;; all elements of list match the pattern (x @a @b)
;; a is bound to (1 2 3); b to (a b c)
(when-match @(all (x @a @b))
'((x 1 a) (x 2 b) (x 3 c))
(list a b))
--> ((1 2 3) (a b c))
;; Match a two element list whose second element
;; consists of nothing but zero or more repetitions
;; of the first element. x is not turned into a list
;; because it has a binding due to @x.
(when-match @(@x @(all x)) '(1 (1 1 1 1)) x) -> 1
;; no match because of the 2
(when-match @(@x @(all x)) '(1 (1 1 1 2)) x) -> nil
@(some pattern)
The some pattern operator requires the corresponding object to be a sequence. The specified pattern is applied against every element of the sequence. The match is successful if pattern matches at least one element.
Variables are extracted from the first matching which is found.
;; the second (x 2 b) element is the leftmost one
;; which matches the (x @a @b) pattern
(when-match @(some (x @a @b))
'((y 1 a) (x 2 b) (z 3 c))
(list a b))
-> (2 b)
@(coll pattern)
The coll pattern operator requires the corresponding object to be a sequence. The specified pattern is applied against every element of the sequence. The match is successful if pattern matches at least one element.
Each variable that is freshly bound by the pattern is converted into a list of all of the objects which that variable encounters from the matching elements of the sequence. Those variables which already have a binding from another pattern are not converted to lists. Their existing values are merely required to match each corresponding object they encounter.
Variables are extracted from all matching elements, and collected into parallel lists, just like with the @(all) operator.
(when-match @(coll (x @a @b))
'((y 1 a) (x 2 b) (z 3 c) (x 4 d))
(list a b))
-> ((2 4) (b d))
@(scan pattern))
@(scan-all pattern))
The scan operator matches pattern against the corresponding object. If the match fails, and the object is a cons cell, the match is tried on the cdr of the cons cell. The cdr traversal repeats until a successful match is found, or a match failure occurs against against an atom.
Thus, a list object, possibly improper, matches pattern under scan if any suffix of that object matches.
The scan-all pattern matches the object in the same way. However, instead of finding the leftmost match, it finds all matches. Every variable that occurs inside pattern is bound to a list of the matches which correspond to that variable.
;; mismatch: 1 doesn't match 2
(when-match @(scan 2) 1 t) -> t
;; simple atom match: 42 matches 42
(when-match @(scan 42) 42 t) -> t
;; (2 3) is a sublist of (1 2 3 4)
(when-match @(scan (2 3 . @nil)) '(1 2 3 4) t) -> t
;; (2 @x 4 . @nil) matches (2 3 4), binding x to 3:
(when-match @(scan (2 @x 4 . @nil)) '(1 2 3 4 5) x) -> 3
;; The entire matching suffix can be captured.
(when-match @(scan @(as sfx (2 @x 4 . @nil)))
'(1 2 3 4 5)
sfx)
-> (2 3 4 5)
;; Missing . @nil in pattern anchors search to end:
(when-match @(scan (@x 2))
'(1 2 3 2 4 2)
x)
;; Terminating atom anchors to improper end:
(when-match @(scan (@x . 4))
'(1 2 3 . 4)
x)
-> 3
;; Atom pattern matches only terminating atom
(when-match @(scan #(@x @y))
'(1 2 3 . #(4 5))
(list x y))
-> (4 5)
;; Pattern doesn't match list:
(match @(scan-all (b @x)) '(1 2 3 4 b 5 b 6 7 8) x)
-> error
;; x bound to list of items that follow b symbol:
(match @(scan-all (b @x . @nil)) '(1 2 3 4 b 5 b 6 7 8) x)
-> (5 6)
@(and pattern*)
@(or pattern*)
The and and or operators match multiple patterns in parallel, against the same object. The and operator matches if every pattern matches the object, otherwise there is no match. The or operator requires one pattern to match. It tries the patterns in left-to-right order, and stops at the first matching one, declaring failure if none match.
The and and or operators have different scoping rules. Under and, later patterns are processed in the scopes of earlier patterns, just like with other pattern operators. Duplicate variables back-reference. Under or, the patterns are processed in separate, parallel scopes. No back-referencing takes place among same-named variables introduced in separate patterns of the same or.
When the and matches, the variables from all of the patterns are bound. When the or operator matches, the variables from all of the patterns are also bound. However, only the variables from the matching pattern take on the values implied by that pattern. The variables from the nonmatching patterns that do not have the same names as variables in the matching pattern, and that have been newly introduced in the or operator, take on nil values.
(if-match @(and (@x 2 3) (1 @y 3) (1 2 @z)) '(1 2 3)
(list x y z)) -> (1 2 3)
(if-match @(or (@x 3 3) (1 @x 3) (1 2 @x)) '(1 2 3)
x) -> 2
@(not pattern)
The pattern operator not provides logical inverse semantics. It matches if and only if the pattern does not match.
Whether or not the not operator matches, no variables are bound. If the embedded pattern matches, the variables which it binds are suppressed by the not operator.
;; @a matches unconditionally, so @(not @a) always fails:
(if-match @(not @a) 1 :yes :no) -> :no
;; error: a is not bound
(if-match @(not @a) 1 :yes a) -> error
(match-case '(1 2 3)
((@(not 1) @b @c) (list :case1 b c))
((@(not 0) @b @c) (list :case2 c b)))
--> (:case2 3 2)
@(function arg*)
@(function arg* @avar arg*)
@(function arg* . @avar)
@(@rvar (function arg*))
@(@rvar (function arg* @avar arg*))
@(@rvar (function arg* . @avar))
Whenever the operator position of a pattern consists of a symbol which is neither the name of a pattern operator, nor the name of a macro, the expression denotes a predicate pattern. An expression is also a predicate pattern if it is handled by a pattern macro which declines to expand it by yielding the original expression.
An operator pattern is expected to conform to one of the first three syntactic variations above. Together, these three variations constitute the first form of the pattern predicate operator. Whenever the operator position of a pattern consists of a meta-symbol, it is also a predicate pattern, expected to conform to one of the second three syntax variations above. These three variations constitute the second form of the operator.
The first form of the predicate pattern consists of a compound form consisting of an operator and arguments. Exactly one of the arguments may be a pattern variable avar ("argument variable") which must be a bindable symbol or else nil. The pattern variable may also appear in the dot position, rather than as an argument. The role of avar and the consequences of omitting it are described below.
The second form of the predicate pattern consists of a meta-symbol rvar ("result variable") which must be a bindable symbol or else nil. This is followed by a compound form which consists of an operator symbol, followed by arguments, one of which may be a pattern avar as in the simple form. If rvar is nil, then the predicate pattern is equivalent to the first form. That is to say, the following are equivalent:
The matching of the predicate pattern is processed as follows. If the avar variable is present, then the predicate pattern first binds the corresponding object to the avar variable, performing an ordinary variable match with the potential back-referencing which that implies. If that succeeds, then the object is inserted into the compound form, substituted in the position indicated by the @avar variable, either an ordinary argument position or the dot position. This form is then evaluated. If it yields true, then the match is successful, otherwise the match fails.
If the avar variable is absent, then no initial variable matching takes place. The corresponding object is added as an extra rightmost argument into the compound form, which is evaluated. Its truth value then determines the success of the match, just like in the case with avar.
If the second form is being processed, and specifies a rvar that is not nil, and if the predicate has succeeded, then then an extra processing step takes place. A variable match is performed to bind the rvar variable to the result of the predicate, with potential back-referencing. If that match succeeds, then the predicate pattern succeeds.
The compound form may be headed by the dwim operator, and therefore the DWIM bracket notation may be used. For instance @[f @x] is equivalent to @(dwim f @x) and is processed accordingly. Similarly, @(@y [f @x]) is equivalent to @(@y (dwim f @x)).
The dot position of avar in the predicate syntax denotes function application. So that is to say, the pattern predicate form (f . @a) where @a is in the dotted position invokes the function f as if by evaluation of the form (f . x) where x is hidden temporary variable holding the object corresponding to the pattern. The form (f . x) is a standard TXR Lisp notation with the same meaning as (apply (fun f) x).
If avar is the nil symbol, then no variable is bound. The matched object is substituted into the predicate expression at the position indicated by @nil.
(when-match (@(evenp) @(oddp @x)) '(2 3) x) -> 3
(when-match @(<= 1 @x 10) 4 x) -> 4
(when-match @(@d (chr-digit @c)) #\5 (list d c)) -> (5 #\5)
(when-match @(<= 1 @x 10) 11 x) -> nil
;; use hash table as predicate:
(let ((h #H(() (a 1) (b 2))))
(when-match @[h @x] 'a x))
-> a
;; as above, also capture hash value
(let ((h #H(() (a 1) (b 2))))
(when-match @(@y [h @x]) 'a (list x y)))
-> (a 1)
;; apply (1 2 3) to < using dot position
(when-match @(@x (< . @sym)) '(1 2 3) (list x sym))
-> (t (1 2 3))
;; Match three-element list whose middle element
;; is a number in the range 10 20, without
;; binding any variables:
(when-match (@nil @(<= 10 @nil 20) @nil) obj
(prinl "obj matches"))
@(sme spat mpat epat [mvar [evar]])
The pattern macro sme (start, middle, end) is a notation defined using the defmatch macro.
The sme macro generates a complex pattern which matches three non-overlapping parts of a list object using three patterns. The spat pattern is required to match a prefix of the input list. If that match is successful, then the remainder of the list is searched for a match for mpat, using the scan operator. If that match, in turn, is successful, then the suffix of the remainder of the list is required to match epat.
The optional mvar and evar arguments must be bindable symbols, if they are specified. These symbols specify lexical variables which are bound to, respectively, the object matched by mpat and epat, using the fresh binding semantics of the as pattern operator.
The first two patterns, spat and mpat, must be possibly dotted list patterns. The last pattern, epat, may be any pattern: it may be an atom match for the terminating atom, or a possibly dotted list pattern matching the list suffix.
Important to the semantics of sme is the concept of the length of a list pattern.
The length of a pattern with a pattern variable or operator in the dotted position is the number of items before that variable or operator. The length of (1 2 . @(and a b)) is 2; likewise the length of (1 2 . @nil) is also 2. The length of a pattern which does not have a variable or operator in the dotted position is simply its list length. For instance, the pattern (1 2 3) has length 3, and so does the pattern (1 2 3 . 4). The length is determined by the list object structure of the pattern, and not the printed syntax used to express it. Thus, (1 . (2 3)) is still a length 3 pattern, because it denotes the same (1 2 3) object, using the dot notation unnecessarily.
The non-overlapping semantics of sme evolves as follows. In the following description, it is understood that a match is required at every step. If that match fails, then the entire sme operator fails:
(when-match @(sme (1 2) (3 4) (5 . 6) m e)
'(1 2 3 4 5 . 6)
(list m e))
-> ((3 4 5 . 6) (5 . 6))
(when-match @(sme (1 2) (3 4) (5 . 6) m e)
'(1 2 abc 3 4 def 5 . 6)
(list m e))
((3 4 def 5 . 6) (5 . 6))
;; backreferencing
(when-match @(sme (1 @y) (@z @x @y @z) (@x @y)) '(1 2 3 1 2 3 1 2)
(list x y z))
-> (1 2 3))
;; collect odd items starting at 3, before 7
(when-match @(and @(sme (1 @x) (3) (7) m e)
@(with @(coll @(oddp @y)) (ldiff m e)))
'(1 2 3 4 5 6 7)
(list x y))
-> (2 (3 5)))
;; no overlap
(when-match @(sme (1 2) (2 3) (3 4)) '(1 2 3 4) t) -> nil
;; The atom 5 is like a "zero-length improper list".
(when-match @(sme () () 5) 5 t) -> t
@(end pattern [var])
The pattern macro end is a notation defined using the defmatch macro, which matches pattern against the suffix of a corresponding list object, which may be an improper list or atom.
The optional argument var specifies the name of a variable which captures the matched portion of the object.
The end macro is related to the sme macro according to the following equivalence:
@(end pat var) <--> @(sme () () pat : var)
All of the requirements given for sme apply accordingly.
;; atom match
(when-match @(end 3 x) 3 x) -> 3
;; y captures (2 3)
(when-match @(end (2 @x) y)
'(1 2 3)
(list x y))
-> (3 (2 3))
;; variable in dot position
(when-match @(end (2 . @x) y)
'(1 2 . 3)
(list x y))
-> (3 (2 . 3))
;; z captures entire object
(when-match @(as z @(end (2 @x) y))
'(1 2 3)
(list x y z))
-> (3 (2 3) (1 2 3)))
(when-match pattern expr form*)
(match pattern expr form*)
(if-match pattern expr then-form [else-form])
The when-match, match and if-match macros conditionally evaluate code based on whether the value of expr matches pattern.
The when-match macro arranges for every form to be evaluated in the scope of the variables established by pattern when it matches the object produced by expr. The value of the last form is returned, or else nil if there are no forms. If the match fails, the forms are not evaluated, and nil is produced.
The match macro behaves exactly like when-match when the match is successful. When the match fails, match throws an exception of type match-error.
The if-match macro evaluates then-form in the scope of the variables established by pattern if the match is successful, and yields the value of that form. Otherwise, it evaluates else-form, which defaults to nil if it is not specified.
(match-case expr {(pattern form*)}*)
(match-ecase expr {(pattern form*)}*)
The match-case macro matches the value of expr against zero or more patterns.
Normally, the patterns are considered in left-to-right order. If the value expr matches more than one pattern, the leftmost pattern is selected and that clause is evaluated. Under certain conditions, detailed below, it is possible for match-case and match-ecase to be transformed into a casequal form. In that case, if there are multiple clauses with equivalent patterns, it is not specified which one is evaluated.
The syntax of match-case consists of an expression expr followed by zero or more clauses. Each clause is a compound expression whose first element is pattern, which is followed by zero or more forms.
First, expr is evaluated. Then, the value is matched against each pattern in succession, stopping at the first pattern which provides a successful match. If no pattern provides a successful match, then match-case terminates and returns nil.
If a pattern matches successfully, then each form associated with the pattern is evaluated in the scope of the variable bindings established by that pattern. Then match-case terminates, returning the value of the last form or else nil if there are no forms.
The match-ecase macro differs from match-case as follows. When none of the clauses match under match-case, then that form terminates with a value of nil. In the same situation, the match-ecase form throws an exception of type match-error.
An
match-ecase
form may be transformed to a
casequal
form if all the
pattern s
are trivial. A trivial pattern is either an atom, or else a vector or list
expression containing no variables.
A match-case form may be transformed to a casequal form under the same conditions as match-case. Additionally, match-case may also be transformed if it contains exactly one clause which matches any object by means of the key @nil or else a variable match such as @abc, if that clause appears last. That clause is transformed into an else-clause of the casequal form.
;; classify sequence of objects by pattern matching,
;; returning a list of the results
(collect-each ((obj (list '(1 2 3)
'(4 5)
'(3 5)
#S(time year 2021 month 1 day 1)
#(vec tor))))
(match-case obj
(@(struct time year @y) y)
(#(@x @y) (list x y))
((@nil @nil @x) x)
((4 @x) x)
((@x 5) x)))
--> (3 5 3 2021 (vec tor))
;; default case can be represented by a guaranteed match
(match-case 1
(2 :two)
(@x :default)) --> :default
(match-cond {(pattern expr form*)}*)
The match-cond macro's arguments are zero or more clauses, each of which specifies a pattern, an expression expr, and zero or more forms.
The clauses are processed in order. Successive exprs are evaluated, and matched against their corresponding pattern. If there is no match, processing continues with the next clause. If no match is found in any clause, the match-cond form terminates, returning nil.
If an expr's value matches the corresponding pattern, then every form is evaluated in scope of the variables established by the pattern. The match-form then terminates, yielding the value of the last form, or else the value of expr if there are no forms.
Note: the pattern (t t ...) is recommended for specifying an unconditionally matching clause.
(let ((x 42))
(match-cond
(`@x-73` "73-73" :a)
(`@x-@y` "42-24" y)))
--> "24"
(lambda-match {(pattern form*)}*)
The lambda-match is conceptually similar to match-case.
The arguments of lambda-match are zero or more clauses similar to those of match-case, each consisting of a compound expression headed by a pattern followed by zero or more forms.
The macro generates a lambda expression which evaluates to an anonymous function in the usual way.
When the anonymous function is called, each clause's pattern is matched against the function's actual arguments. When a match occurs, each form associated with the pattern is evaluated, and the value of the last form becomes the return value of the function. If none of the clauses match, then nil is returned.
Whenever pattern is a list-like pattern, it is not matched against a list object, as is the usual case with a list-like pattern, but against the actual arguments. For instance, the pattern (@a @b @c) expects that the function was called with exactly three arguments. If that is the case, the patterns are then matched to the arguments. The pattern @a takes the first argument, binding it to variable a and so forth.
If pattern is a dotted list-like pattern, then the dot position is matched against the remaining arguments. For instance, the pattern (@a @b . @c) requires at least two arguments. The first two are bound to a and b, respectively. The list of remaining arguments, if any, is bound to c, which will be nil if there are no remaining arguments.
Any non-list-like pattern P is analyzed as an equivalent list-like dotted pattern due to P syntax being equivalent to (. P) syntax. Such a pattern matches the list of all arguments. Thus, the following are all equivalent:
(lambda-match (@a a))
(lambda-match ((. @a) a))
(lambda a a)
(lambda (. a) a)
The characteristics of the resulting anonymous function are determined as follows.
If at least one pattern specified in a lambda-match is a dotted pattern, the function is variadic.
The arity of the resulting anonymous function is determined as follows, from the lengths of the patterns. The length of a pattern is the number of elements, not including the dotted element.
The length of the longest pattern determines the number of fixed arguments. Unless the function is variadic, it may not be called with more arguments than can be matched by the longest pattern.
The length of the shortest pattern determines the number of required arguments. The function may not be called with fewer arguments than can be matched by the shortest pattern.
If these two lengths are unequal, then the function has a number of optional arguments, equal to the difference.
Note: an anonymous function which takes one argument and matches that object against clauses using match-case can be obtained with the do operator, using the pattern: (do match @1 ...).
Note: the parameter macro :match can also define a lambda with pattern matching. Any (lambda-match clauses ...) form can be written as (lambda (:match) clauses ...). The parameter macro offers the additional ability of defining named arguments which are inserted before the implicit arguments generated from the clauses, and combining with other parameter macros.
(let ((f (lambda-match
(() (list 0 :args))
((@a) (list 1 :arg a))
((@a @b) (list 2 :args a b))
((@a @b . @c) (list* '> 2 :args a b c)))))
(list [f] [f 1] [f 1 2] [f 1 2 3]))
-->
((0 :args) (1 :arg 1) (2 :args 1 2) (> 2 :args 1 2 3))
[(lambda-match
((0 1) :zero-one)
((1 0) :one-zero)
((@x @y) :no-match)) 1 0] --> :one-zero
[(lambda-match
((0 1) :zero-one)
((1 0) :one-zero)
((@x @y) :no-match)) 1 1] --> :no-match
[(lambda-match
((0 1) :zero-one)
((1 0) :one-zero)
((@x @y) :no-match)) 1 2 3] --> ;; error
(defun-match name {(pattern form*)}*)
The defun-match macro can be used to define a top-level function in the style of lambda-match.
It produces a form which has all of the properties of defun, such as a block of the same name being established around the implicit match-case so that return-from is possible.
The (pattern form*) clauses of defun-match have exactly the same syntax and semantics as those of lambda-match.
Note: instead of defun-match, the parameter macro :match may be used. The following equivalence holds:
(defun name (:match) ...) <--> (defun-match ...)
The parameter macro offers the additional ability of defining named arguments which are inserted before the implicit arguments generated from the clauses, and combining with other parameter macros.
;; Fibonacci
(defun-match fib
((0) 1)
((1) 1)
((@x) (+ (fib (pred x)) (fib (ppred x)))))
(fib 0) -> 1
(fib 1) -> 1
(fib 2) -> 2
(fib 3) -> 3
(fib 4) -> 5
(fib 5) -> 8
;; Ackermann
(defun-match ack
((0 @n) (+ n 1))
((@m 0) (ack (- m 1) 1))
((@m @n) (ack (- m 1) (ack m (- n 1)))))
(ack 3 7) -> 1021
(ack 1 1) -> 3
(ack 2 2) -> 7
(:match left-param* [-- extra-param*]) clause*
Parameter list macro :match allows any function to be expressed in the style of lambda-match, with extra features.
The :match macro expects the body of the function to consist of lambda-match clauses, which are semantically treated in exactly the same manner as under lambda-match.
The following restrictions apply. The parameter list may not include optional parameters delimited by : (the colon keyword symbol). The parameter list may not be dotted.
The macro produces a function which the left-param parameters, if any, are inserted to the left of the implicit parameters generated by the lambda-match transformation.
Furthermore, the :match parameter macro supports integration with the :key parameter macro, or any other macro which uses a compatible -- convention for delimiting special arguments. If the parameter list includes the symbol -- then that portion of the parameter list is set aside and not included in the lambda-match transformation. Then, that list is integrated into the resulting lambda.
A complete transformation can be described by the following diagram:
(lambda (:match a b c ... -- s t u ...) clauses ...)
-->
(lambda (a b c ... m n p ... -- s t u ... . z) body ...)
In this diagram, a b c ... denote the left-param parameters. The m n p ... symbols denote the fixed parameters generated by the lambda-match transformation from the semantic analysis of clauses. The s t u ... symbols denote the original extra-param parameters. Finally, z denotes the dotted parameter generated by the lambda-match transform. If the transform produces no dotted parameter, then this is nil. The dotted parameter is thus separated from the m n p ... group to which it belongs.
When no -- and extra-params are present, the transformation reduces to:
(lambda (:match a b c ...) clauses ...)
-->
(lambda (a b c ... m n p ... . z) body ...)
Note: these requirements harmonize with the :key parameter macro. If that is present to the left of :match it removes the -- and the s t u ... keyword parameters, reuniting the z parameter with the m n p group. Furthermore, the :key macro generates code which refers to the existing z dotted parameter as the source for the keyword parameters, unless z is nil, in which case it inserts its own generated symbol.
;; Match-style cond-like macro with unreachability diagnosis.
;; Demonstrates usefulness of :match, which allows the :form
;; parameter to be promoted through to the macro definition.
(defmacro my-cond (:match :form f)
(() nil)
(((@(and @(constantp @test) @(eval))) . @rest)
(when rest
(compile-error f "unreachable code after ~s" test))
test)
(((@(and @(constantp @test) @(eval)) . @forms) . @rest)
(when (and rest)
(compile-error f "unreachable code after ~s" test))
^(progn ,*forms))
(((@test) . @rest)
^(or ,test (my-cond ,*rest)))
(((@test . @forms) . @rest)
^(if ,test (progn ,*forms)
(my-cond ,*rest)))
((@else . @rest) (compile-error f "bad syntax")))
(my-cond (3)) --> 3
(my-cond (3 4)) --> 4
(my-cond (3 4) (5)) --> ;; my-cond: unreachable code after 3
(my-cond 42) --> ;; my-cond: bad syntax
;; Keyword parameter example.
(defstruct simple-widget ()
name)
(defstruct widget (simple-widget)
frobosity
luminance)
(defstruct simple-point-widget (simple-widget)
(:static width 0)
(:static height 0))
(defstruct point-widget (widget)
(:static width 0)
(:static height 0))
(defstruct general-widget (widget)
width
height)
;; Note that in clauses with no . @rest parameter, there
;; is a mismatch if keyword arguments are present. The (0 0)
;; clause exploits this to match only when keywords are absent.
(defun make-widget (:key :match name -- frob lum)
((0 0) (new simple-point-widget name name))
((0 0 . @rest) (new point-widget name name
frobosity frob
luminance lum))
((@x @y . @rest) (new general-widget name name
width x
height x
frobosity frob
luminance lum)))
(make-widget "abc" 0 0) --> #S(simple-point-widget name "abc")
(make-widget "abc" 0 0 :frob 42)
--> #S(point-widget name "abc" frobosity 42 luminance nil)
(make-widget "abc" 0 0 :lum 9)
--> #S(point-widget name "abc" frobosity nil luminance 9)
(make-widget "abc" 0 1 :lum 9)
--> #S(general-widget name "abc" frobosity nil luminance 9
width 0 height 0)
(defmatch name macro-style-params
body-form*)
The defmatch macro allows for the definition of pattern macros: user-defined pattern operators which are implemented via expansion into existing operator syntax.
The defmatch macro has the same syntax as defmacro. It specifies a macro transformation for a compound form which has the name symbol in its leftmost position.
This macro transformation is performed when name is used as a pattern operator: an expression of the form @(name argument*) occurring in pattern-matching syntax.
The behavior is unspecified if name is the name a built-in pattern operator, or a predefined pattern macro.
The pattern macro bindings are stored in a hash table held by the variable *match-macro* whose keys are symbols, and whose values are expander functions. There are no lexically scoped pattern macros.
Pattern macros defined with defmatch may specify the special macro parameters :form and :env in their parameter lists. The values of these parameters are determined in a manner particular to defmatch.
The :form parameter captures the pattern-matching form, or a constituent thereof, in which the macro is being invoked. For instance, if the operator is being used inside a pattern given to a when-match macro invocation, then the form will be that entire when-match form.
The :env parameter captures a specially constructed macro-time environment object in which all of the variables to the left of the pattern appear as lexical variables. The parent of this environment is the surrounding macro environment. If the pattern macro needs to treat a variable which already has a binding differently from an unbound variable, it can look up the variable in this environment.
;; Create an alias called let for the @(as var pattern) operator:
;; Note that the macro produces @(as ...) and not just (as ...)
(defmatch let (var pattern)
^@(as ,var ,pattern))
;; use the macro in matching:
(when-match @(let x @(or foo bar)) 'foo x)
;; Error reporting example using :form
(defmatch foo (sym)
(unless (bindable sym)
(compile-error *match-form*
"~s: bindable symbol expected, not ~s"
'foo sym))
...)
;; Pattern macro which uses = equality to backreference
;; an existing lexical binding, or else binds the variable
;; if it has no existing lexical binding.
(defmatch var= (sym :env e)
(if (lexical-var-p e sym)
(with-gensyms (obj)
^@(require (sys:var ,obj)
(= ,sym ,obj)))
^(sys:var ,sym)))
;; example use:
(when-match (@(var= a) @(var= a)) '(1 1.0) a)
-> 1
;; no match: (equal 1 1.0) is false
(when-match (@a @a) '(1 1.0) a)
-> nil
(macroexpand-match pattern [env])
If pattern is a compound form whose operator symbol has been defined as a macro pattern using defmatch, then macroexpand-match will expand that pattern and return the expansion. Otherwise it returns the pattern argument.
In order to be recognized by macroexpand-match the pattern argument must not include the @ prefix that would normally be used to invoke it. The expansion, however, will include that syntax.
The env parameter specifies the macro-time environment for the expander. Note: pattern expanders, like built-in patterns, may use the macro environment for deciding whether a variable is an existing lexical variable, or a free variable, based on which a pattern may be expanded differently.
Given:
(defmatch point (x y)
^@(struct point x @,x y @,y))
a result similar to the following may be obtained:
(macroexpand-match '(point a b)) -> @(struct point x @a y @b)
Note that the pattern is specified plainly as
(point a b)
rather than
@(point a b),
yet the expansion is
@(struct ...).
The *match-macro* special variable holds the hash table of associations between symbols and pattern macro expanders.
If the expression [*match-macro* 'sym] yields a function, then symbol sym has a binding as a pattern macro. If that expression yields nil, then there is no such binding: pattern operator forms based on sym do not undergo place macro expansion.
The macro expanders in *match-macro* are two-parameter functions. The first argument passes the operator syntax to be expanded. The second argument is used for passing the environment object which the expander can capture using :env in its macro parameter list.
(each-match ({pattern seq-form}*) body-form*)
(each-match-product ({pattern seq-form}*) body-form*)
The each-match macro arranges for elements from multiple sequences to be visited in parallel, and each to be matched against respective patterns. For each matching tuple of parallel elements, a body of forms is evaluated in the scope of the variables bound in the patterns.
The first argument of each-match specifies a list of alternating pattern and seq-form expressions. Each pattern is associated with the sequence which results from evaluating the immediately following seq-form. Items coming from that sequence correspond with that pattern.
The remaining arguments are body-forms to evaluated for successful matches.
The body-forms are surrounded by an implicit anonymous block. If any of the forms return invoke a return out of this block, then the iteration terminates, and the result value of the block becomes the result value of the loop.
The processing takes place as follows:
;; Number all the .JPG files in the current directory.
;; For instance foo.jpg becomes foo-0001.jpg, if it is
;; the first file.
(each-match (@(as name `@base.jpg`) (glob "*.jpg")
@(@num (fmt "~,04a")) 1)
(rename-path name `@base-@num.jpg`))
;; Iterate over combinations of matching phone
;; numbers and odd integers from the (1 2 3) list
(build
(each-match-product (`(@a) @b-@c` '("x"
""
"(311) 555-5353"
"(604) 923-2323"
"133"
"4-5-6-7")
@(oddp @x) '(1 2 3))
(add (list x a b c))))
-->
((1 "311" "555" "5353") (3 "311" "555" "5353")
(1 "604" "923" "2323") (3 "604" "923" "2323")))
(append-matches ({pattern seq-form}*) body-form*)
(append-match-products ({pattern seq-form}*) body-form*)
The macro append-matches is subject to all of the requirements specified for each-match in regard to the argument conventions and semantics, and the presence of the implicit anonymous block around the body-forms.
Whereas each-match returns nil, the append-matches macro requires, in each iteration which produces a match for each pattern, that the last body-form evaluated must produce a list.
These lists are catenated together as if by the append function and returned.
It is unspecified whether the nonmatching iterations produce empty lists which are included in the append operation.
If the last tuple of items which produces a match is absolutely the the last tuple, the corresponding body-form evaluation may yield an atom which then becomes the terminator for the returned list, in keeping with the semantics of append. an atom.
The append-match-products macro differs from append-matches in that it iterates over the Cartesian product tuples of the sequences, rather than parallel tuples. The difference is exactly like that between each-match and each-match-product.
(append-matches
((:foo @y) '((:foo a) (:bar b) (:foo c) (:foo d))
(@x :bar) '((1 :bar) (2 :bar) (3 :bar) (4 :foo)))
(list x y))
--> (1 a 3 c)
(append-matches (@x '((1) (2) (3) 4)) x)
--> (1 2 3 . 4)
(append-match-products (@(oddp @x) (range 1 5)
@(evenp @y) (range 1 5))
(list x y))
--> (1 2 1 4 3 2 3 4 5 2 5 4)
(keep-matches ({pattern seq-form}*) body-form*)
(keep-match-products ({pattern seq-form}*) body-form*)
The macro keep-matches is subject to all of the requirements specified for each-match in regard to the argument conventions and semantics, and the presence of the implicit anonymous block around the body-forms.
Whereas each-match returns nil, the keep-matches macro returns a list of the values produced by all matching iterations which led to the execution of the body-forms.
The keep-match-products macro differs from keep-matches in that it iterates over the Cartesian product tuples of the sequences, rather than parallel tuples. The difference is exactly like that between each-match and each-match-product.
(keep-matches ((:foo @y) '((:foo a) (:bar b) (:foo c) (:foo d))
(@x :bar) '((1 :bar) (2 :bar) (3 :bar) (4 :foo)))
(list x y))
--> ((1 a) (3 c))
(keep-match-products (@(oddp @x) (range 1 5)
@(evenp @y) (range 1 5))
(list x y))
--> ((1 2) (1 4) (3 2) (3 4) (5 2) (5 4))
(while-match pattern expr form*)
The while-match macro evaluates expr and matches it against pattern similarly to when-match.
If the match is successful, every form is evaluated in an environment in which new bindings from pattern are visible. In this case, the process repeats: expr is evaluated again, and tested against pattern.
If the match fails, while-match terminates and produces nil as its result value.
Each iteration produces fresh bindings for any variables that are implicated for binding in pattern.
The expr and form expressions are surrounded by an anonymous block.
(while-match-case expr {(pattern form*)}*)
(while-true-match-case expr {(pattern form*)}*)
The macros while-match-case and while-true-match-case combine iteration with the semantics of match-case.
The while-match-case evaluates expr and matches it against zero or more clauses in the manner of match-case. If there is a match, this process is repeated. If there is no match, while-match-case terminates, and returns nil.
In each iteration, the matching clause produces fresh bindings for any variables implicated for binding in its respective pattern.
The expr and form expressions are surrounded by an anonymous block.
The while-true-match-case macro is identical in almost every respect to while-match-case, except that it terminates the loop if expr evaluates to nil, without attempting to match that value against the clauses.
Note: the semantics of while-true-match-case can be obtained in while-match-case by inserting a return clause. That is to say, a construct of the form
(while-true-match-case expr
...)
may be rewritten into
(while-match-case expr
(nil (return)) ;; match nil and return
...)
except that while-true-match-case isn't required to rely on performing a block return.
(qquote form)
The qquote (quasi-quote) macro operator implements a notation for convenient list construction. If form is an atom, or a list structure which does not contain any unquote or splice operators, then (qquote form) is equivalent to (qquote form).
If form, however, is a list structure which contains unquote or splice operators, then the substitutions implied by those operators are performed on form, and the qquote operator returns the resulting structure.
Note: how the qquote operator actually works is that it is compiled into code. It becomes a Lisp expression which, when evaluated, computes the resulting structure.
A qquote can contain another qquote. If an unquote or splice operator occurs within a nested qquote, it belongs to that qquote, and not to the outer one.
However, an unquote operator which occurs inside another one belongs one level higher. For instance in
(qquote (qquote (unquote (unquote x))))
the leftmost qquote belongs with the rightmost unquote, and the inner qquote and unquote belong together. When the outer qquote is evaluated, it will insert the value of x, resulting in the object (qquote (unquote [value-of-x])). If this resulting qquote value is evaluated again as Lisp syntax, then it will yield [value-of-value-of-x], the value of [value-of-x] when treated as a Lisp expression and evaluated.
(qquote a) -> a
(qquote (a b c)) -> (a b c)
(qquote (1 2 3 (unquote (+ 2 2)) (+ 2 3))) -> (1 2 3 4 (+ 2 3))
(qquote (unquote (+ 2 2))) -> 4
In the second-to-last example, the 1 2 3 and the (+ 2 3) are quoted verbatim. Whereas the (unquote (+ 2 2)) operator caused the evaluation of (+ 2 2) and the substitution of the resulting value.
The last example shows that form can itself (the entire argument of qquote) can be an unquote operator. However, note: (quote (splice form)) is not valid.
Note: a way to understand the nesting behavior is a via a possible model of quasi-quote expansion which recursively compiles any nested quasi quotes first, and then treats the result of their expansion. For instance, in the processing of
(qquote (qquote (unquote (unquote x))))
the qquote operator first encounters the embedded (qquote ...) and compiles it to code. During that recursive compilation, the syntax (unquote (unquote x)) is encountered. The inner quote processes the outer unquote which belongs to it, and the inner (unquote x) becomes material that is embedded verbatim in the compilation, which will then be found when the recursion pops back to the outer quasiquote, which will then traverse the result of the inner compilation and find the (unquote x).
In Lisp dialects which have a published quasiquoting operator syntax, there is the expectation that the quasiquote read syntax corresponds to it. That is to say, the read syntax ^(a b ,c) is expected to translate to (qquote a b (unquote c)).
In TXR Lisp, this is not true! Although ^(a b ,c) is translated to a quasiquoting macro, it is an internal one, not based on the public qquote, unquote and splice symbols being documented here.
This idea exists for hygiene. The quasiquote read syntax is not confused by the presence of the symbols qquote, unquote or splice in the template, since it doesn't treat them specially.
This also allows programmers to use the quasiquote read syntax to construct quasiquote macros. For instance
^(qquote (unquote ,x)) ;; does not mean ^^,,x !
To the quasiquote reader, the qquote and unquote symbols mean nothing special, and so this syntax simply means that if the value of x is foo, the result of evaluating this expression will be (qquote (unquote foo)).
The form's expansion is actually this:
(sys:qquote (qquote (unquote (sys:unquote x))))
the sys:qquote macro recognizes sys:unquote embedded in the form, and the other symbols not in the sys: package are just static template material.
The sys:quote macro and its associated sys:unquote and sys:splice operators work exactly like their ordinary counterparts. So in effect, TXR has two nearly identical, independent quasi-quote implementations, one of which is tied to the read syntax, and one of which isn't. This is useful for writing quasiquotes which write quasiquotes.
(qquote (... (unquote form) ...))
(qquote (unquote form))
The unquote operator is not an operator per se. The unquote symbol has no binding in the global environment. It is a special syntax that is recognized within a qquote form, to indicate forms within the quasiquote which are to be evaluated and inserted into the resulting structure.
The syntax (qquote (unquote form)) is equivalent to form: the qquote and unquote "cancel out".
(qquote (... (splice form) ...))
The splice operator is not an operator per se. The splice symbol has no binding in the global environment. It is a special syntax that is recognized within a qquote form, to indicate forms within the quasiquote which are to be evaluated and inserted into the resulting structure.
The syntax (qquote (splice form)) is not permitted and raises an exception if evaluated. The splice syntax must occur within a list, and not in the dotted position.
The splice form differs from unquote in that (splice form) requires that form must evaluate to a list. That list is integrated into the surrounding list.
The following documentation describes the behavior of the Math Library functions as they apply to the native numeric and character types.
The functions also support application-defined structure types. That feature is not described here but in the section User-Defined Arithmetic Types.
When one or more operands of a Math Library function is a user-defined arithmetic structure, no conversions are performed on the operands, and the stated restrictions do not apply. The operands are passed to the methods as described in the User-Defined Arithmetic Types section. The operands need not be numeric.
User-defined arithmetic structures can work with operands which are not numbers. If a is such a type, it is possible for an expression such as (+ a "abc") to be meaningful and correct. Similarly, it is possible for an apparent division by zero such as (/ a 0) to be meaningful and correct, since the / method of the a object decides how to handle zero.
(+ number*)
(- number number*)
(* number*)
The +, - and * functions perform addition, subtraction and multiplication, respectively. Additionally, the - function performs additive inverse.
The + function requires zero or more arguments. When called with no arguments, it produces 0 (the identity element for addition), otherwise it produces the sum over all of the arguments.
Similarly, the * function requires zero or more arguments. When called with no arguments, it produces 1 (the identity element for multiplication). Otherwise it produces the product of all the arguments.
The semantics of - changes from subtraction to additive inverse when there is only one argument. The argument is treated as a subtrahend, against an implicit minuend of zero. When there are two or more argument, the first one is the minuend, and the remaining are subtrahends.
When there are three or more operands, these operations are performed as if by binary operations, in a left-associative way. That is to say, (+ a b c) means (+ (+ a b) c). The sum of a and b is computed first, and then this is added to c. Similarly (- a b c) means (- (- a b) c). First, b is subtracted from a, and then c is subtracted from that result.
The arithmetic inverse is performed as if it were subtraction from integer 0. That is, (- x) means the same thing as (- 0 x).
The operands of +, - and * can be characters, integers (fixnum and bignum), and floats, in nearly any combination.
If two operands have different types, then one of them is converted to the type of the one with the higher rank, according to this ranking: character < integer < float. For instance if one operand is integer, and the other float, the integer is converted to a float.
Characters are not considered numbers, and participate in these operations in limited ways. Subtraction can be used to computed the displacement between the Unicode values of characters, and an integer displacement can be added to a character, or subtracted from a character. For instance (- #\9 #\0) is 9. The Unicode value of a character C can be found using (- C #\x0): the displacement from the NUL character.
The rules can be stated as a set of restrictions:
(/ divisor)
(/ dividend divisor*)
The / function performs floating-point division. Each operands is first converted to floating-point type, if necessary. In the one-argument form, the dividend argument is omitted. An implicit dividend is present, whose value is 1.0, such that the one-argument form (/ x) is equivalent to the two-argument form (/ 1.0 x).
If there are two or more arguments, explicitly or by the above equivalence, then a cumulative division is performed. The divisor value is taken into consideration, and divided by the first divisor. If another divisor follows, then that value is divided by that subsequent divisor. This process repeats until all divisors are exhausted, and the value of the last division is returned.
A division by zero throws an exception of type numeric-error.
(sum sequence [keyfun])
(prod sequence [keyfun])
The sum and prod functions operate on an effective sequence of numbers derived from sequence, which is an object suitable for iteration according to seq-begin.
If the keyfun argument is omitted, then the effective sequence is the sequence argument itself. Otherwise, the effective sequence is understood to be a projection mapping of the elements of sequence through keyfun as would be calculated by the (mapcar keyfun sequence) expression.
The sum function returns the left-associative sum of the elements of the effective sequence calculated as if using the + function. Similarly, the prod function calculates the left-associative product of the elements of the sequence as if using the * function.
If sequence is empty then sum returns 0 and prod returns 1.
If the effective sequence contains one number, then both functions return that number.
(wrap start end number)
(wrap* start end number)
The wrap and wrap* functions reduce number into the range specified by start and end.
Under wrap the range is inclusive of the end value, whereas under wrap* it is exclusive.
The following equivalence holds
(wrap a b c) <--> (wrap* a (succ b) c)
The expression (wrap* x0 x1 x) performs the following calculation:
(+ (mod (- x x0) (- x1 x0)) x0)
In other words, first start is subtracted from number. Then the result is reduced modulo the displacement between start and end. Finally, start is added back to that result, which is returned.
;; perform ROT13 on the string "nop"
[mapcar (opip (+ 13) (wrap #\a #\z)) "nop"] -> "abc"
(gcd number*)
(lcm number*)
The gcd function computes the greatest common divisor: the largest positive integer which divides each number.
The lcm function computes the lowest common multiple: the smallest positive integer which is a multiple of each number.
Each number must be an integer.
Negative integers are replaced by their absolute values, so (lcm -3 -4) is 12 and (gcd -12 -9) yields 3.
The value of (gcd) is 0 and that of (lcm) is 1 .
The value of (gcd x) and (lcm x) is (abs x).
Any arguments of gcd which are zero are effectively ignored so that (gcd 0) and (gcd 0 0 0) are both the same as (gcd) and (gcd 1 0 2 0 3) is the same as (gcd 1 2 3).
If lcm has any argument which is zero, it yields zero.
(divides d n)
The divides function tests whether integer d divides integer n. If this is true, t is returned, otherwise nil.
The integers 1 and -1 divide every other integer and themselves. By established convention, every integer, except zero, divides zero.
For other values, d divides n if division of n by d leaves no remainder.
(abs number)
The abs function computes the absolute value of number. If number is positive, it is returned. If number is negative, its additive inverse is returned: a positive number of the same type with exactly the same magnitude.
(signum number)
The signum function calculates a representation of the sign of number as a numeric value.
If number is an integer, then signum returns -1 if the integer is negative, 1 if the integer is positive, or else 0.
If number is a floating-point value then signum returns -1.0 if the value is negative, 1.0 if the value is positive or else 0.0.
(trunc dividend [divisor])
(floor dividend [divisor])
(ceil dividend [divisor])
(round dividend [divisor])
The trunc, floor, ceil and round functions perform division of the dividend by the divisor, returning an integer quotient.
If the divisor is omitted, it defaults to 1.
A zero divisor results in an exception of type numeric-error.
If both inputs are integers, the result is of type integer.
If all inputs are numbers and at least one of them is floating-point, the others are converted to floating-point and the result is floating-point.
The dividend input may be a range. In this situation, the operation is recursively distributed over the from and to fields of the range, individually matched against the divisor, and the result is a range composed of these two individual quotients.
When the quotient is a scalar value, trunc returns the closest integer, in the zero direction, from the value of the quotient. The floor function returns the highest integer which does not exceed the value of the quotient. That is to say, the division is truncated to an integer value toward negative infinity. The ceil function the lowest integer which is not below the value of the quotient. does not exceed the value of dividend. That is to say, the division is truncated to an integer value toward positive infinity. The round function returns the nearest integer to the quotient. Exact halfway cases are rounded to the integer away from zero so that (round -1 2) yields -1 and (round 1 2) yields 1.
Note that for large floating point values, due to the limited precision, the integer value corresponding to the mathematical floor or ceiling may not be available.
In ANSI Common Lisp, the round function chooses the nearest even integer, rather than rounding halfway cases away from zero. TXR's choice harmonizes with the semantics of the round function in the C language.
(mod dividend divisor)
The mod function performs a modulus operation. Firstly, the absolute value of divisor is taken to be a modulus. Then a residue of dividend with respect to modulus is calculated. The residue's sign follows that of the sign of divisor. That is, it is the smallest magnitude (closest to zero) residue of dividend with respect to the absolute value of divisor, having the same sign as divisor. If the operands are integer, the result is an integer. If either operand is of type float, then the result is a float. The modulus operation is then generalized into the floating point domain. For instance the expression (mod 0.75 0.5) yields a residue of 0.25 because 0.5 "goes into" 0.75 only once, with a "remainder" of 0.25.
If divisor is zero, mod throws an exception of type numeric-error.
(trunc-rem dividend [divisor])
(floor-rem dividend [divisor])
(ceil-rem dividend [divisor])
(round-rem dividend [divisor])
These functions, respectively, perform the same division operation as trunc, floor, ceil, and round, referred to here as the respective target functions.
If the divisor is missing, it defaults to 1.
Each function returns a list of two values: a quotient and a remainder. The quotient is exactly the same value as what would be returned by the respective target function for the same inputs.
The remainder value obeys the following identity:
(eql remainder (- dividend (*divisor quotient)))
If divisor is zero, these functions throw an exception of type numeric-error.
(sin radians)
(cos radians)
(tan radians)
(atan slope)
(atan2 y x)
(asin num)
(acos num)
These trigonometric functions convert their argument to floating point and return a float result. The sin, cos and tan functions compute the sine and cosine and tangent of the radians argument which represents an angle expressed in radians. The atan, acos and asin are their respective inverse functions. The num argument to asin and acos must be in the range -1.0 to 1.0. The atan2 function converts the rectilinear coordinates x and y to an angle in polar coordinates in the range [0, 2π).
(sinh argument)
(cosh argument)
(tanh argument)
(atanh argument)
(asinh argument)
(acosh argument)
These functions are the hyperbolic analogs of the trigonometric functions sin, cos and so forth. They convert their argument to floating point and return a float result.
(exp arg)
(log arg)
(log10 arg)
(log2 arg)
The exp function calculates the value of the transcendental number e raised to the exponent arg.
The log function calculates the base e logarithm of arg, which must be a positive value.
The log10 function calculates the base 10 logarithm of arg, which must be a positive value.
The log2 function calculates the base 2 logarithm of arg, which must be a positive value.
(expt base exponent*)
(sqrt arg)
(isqrt arg)
The expt function raises base to zero or more exponents given by the exponent arguments. (expt x) is equivalent to (expt x 1), and yields x for all x. For three or more arguments, the operation is right-associative. That is to say, (expt x y z) is equivalent to (expt x (expt y z)), similarly to the way nested exponents work in standard algebraic notation.
Exponentiation is done pairwise using a binary operation. If both operands to this binary operation are nonnegative integers, then the result is an integer.
If the exponent is negative, and the base is zero, the situation is treated as a division by zero: an exception of type numeric-error is thrown. Otherwise, a negative exponent is converted to floating-point, if it already isn't, and a floating-point exponentiation is performed.
If either operand is a float, then the other operand is converted to a float, and a floating point exponentiation is performed. Exponentiation that would produce a complex number is not supported.
If the exponent is zero, then the return value is 1.0 if at least one operand is floating-point, otherwise 1.
The sqrt function produces a floating-point square root of arg, which is converted from integer to floating-point if necessary. Negative operands are not supported.
The isqrt function computes the integer square root of arg, which must be an integer. The integer square root is a value which is the greatest integer that is no greater than the real square root of arg. The input value must be an integer.
(exptmod base exponent modulus)
The exptmod function performs modular exponentiation and accepts only integer arguments. Furthermore, exponent must be a nonnegative and modulus must be positive.
The return value is base raised to exponent, and reduced to the least positive residue modulo modulus.
(square argument)
The square function returns the product of argument with itself. The following equivalence holds, except that x is evaluated only once in the square expression:
(cum-norm-dist argument)
The cum-norm-dist function calculates an approximation to the cumulative normal distribution function: the integral, of the normal distribution function, from negative infinity to the argument.
(inv-cum-norm argument)
The inv-cum-norm function calculates an approximate to the inverse of the cumulative normal distribution function. The argument, a value expected to lie in the range [0, 1], represents the integral of the normal distribution function from negative infinity to some domain point p. The function calculates the approximate value of p. The minimum value returned is -10, and the maximum value returned is 10, regardless of how closely the argument approaches, respectively, the 0 or 1 integral endpoints. For values less than zero, or exceeding 1, the values returned, respectively, are -10 and 10.
(n-choose-k n k)
(n-perm-k n k)
The n-choose-k function computes the binomial coefficient nCk which expresses the number of combinations of k items that can be chosen from a set of n, where combinations are subsets.
The n-perm-k function computes nPk: the number of permutations of size k that can be drawn from a set of n, where permutations are sequences, whose order is significant.
The calculations only make sense when n and k are nonnegative integers, and k does not exceed n. The behavior is not specified if these conditions are not met.
(fixnump object)
(bignump object)
(integerp object)
(floatp object)
(numberp object)
These functions test the type of object, returning t if it is an object of the implied type, nil otherwise. The fixnump, bignump and floatp functions return t if the object is of the basic type fixnum, bignum or float. The function integerp returns true of object is either a fixnum or a bignum. The function numberp returns t if object is either a fixnum, bignum or float.
(arithp object)
The arithp function returns true if object is a character, integer, floating-point number, range or a user-defined arithmetic object. For a range, t is returned without examining the values of the from and to fields. A user-defined arithmetic object is identified as a struct type which implements the + method as a static slot.
(zerop number)
(nzerop number)
The zerop function tests number for equivalence to zero. The argument must be a number or character. It returns t for the integer value 0 and for the floating-point value 0.0. For other numbers, it returns nil. It returns t for the null character #\nul and nil for all other characters.
If number is a range, then zerop returns t if both of the range endpoints individually satisfy zerop.
The nzerop function is the logical inverse of zerop: it returns t for those arguments for which zerop returns nil and vice versa.
(plusp number)
(minusp number)
These functions test whether a number is positive or negative, returning t or nil, as the case may be.
The argument may also be a character. All characters other than the null character #\nul are positive. No character is negative.
(evenp integer)
(oddp integer)
The evenp and oddp functions require integer arguments. evenp returns t if integer is even (divisible by two), otherwise it returns nil. oddp returns t if integer is not divisible by two (odd), otherwise it returns nil.
(succ number)
(ssucc number)
(sssucc number)
(pred number)
(ppred number)
(pppred number)
The succ function adds 1 to its argument and returns the resulting value. If the argument is an integer, then the return value is the successor of that integer, and if it is a character, then the return value is the successor of that character according to Unicode.
The pred function subtracts 1 from its argument, and under similar considerations as above, the result represents the predecessor.
The ssucc and sssucc functions add 2 and 3, respectively. Similarly, ppred and pppred subtract 2 and 3 from their argument.
(> object object*)
(< object object*)
(>= object object*)
(<= object object*)
(= object object*)
These relational functions compare characters, numbers, ranges and sequences of characters or numbers for numeric equality or inequality. The arguments must be one or more numbers, characters, ranges, or sequences of these objects, or, recursively, of sequences.
If just one argument is given, then these functions all return t.
If two arguments are given then, they are compared as follows. First, if the numbers do not have the same type, then the one which has the lower ranking type is converted to the type of the other, according to this ranking: character < integer < float. For instance if a character and integer are compared, the character is converted to its integer character code. Then a numeric comparison is applied.
Three or more arguments may be given, in which case the comparison proceeds pairwise from left to right. For instance in (< a b c), the comparison (< a b) is performed in isolation. If the comparison is false, then nil is returned, otherwise the comparison (< b c) is performed in isolation, and if that is false, nil is returned, otherwise t is returned. Note that it is possible for b to undergo two different conversions. For instance in the (< float character integer) comparison, character will first convert to a floating-point representation of its Unicode value so that it can be compared to float, and if that comparison succeeds, then in the second comparison, character will be converted to integer so that it can be compared to integer.
Ranges may only be compared with ranges. Corresponding fields of ranges are compared for equality by = such that #R(0 1) and #R(0 1.0) are reported as equal. The inequality comparisons are lexicographic, such that the from field of the range is considered more major than the to field. For example the inequalities (< #R(1 2) #R(2 0)) and (< #R(1 2) #R(1 3)) hold.
Sequences may only be compared with sequences, but mixtures of any kinds of sequences may be compared: lists with vectors, vectors with strings, and so on.
The = function considers a pair of sequences of unequal length to be unequal, reporting nil. Sequences are equal if they have the same length and their corresponding elements are recursively equal under the = function.
The inequality functions treat sequences lexicographically. A pair of sequences is compared by comparing corresponding elements. The < function tests each successive pair of corresponding elements recursively using the < function. If this recursive comparison reports t, then the function immediately returns t without considering any more pairs of elements. Otherwise the same pair of elements is compared again using the = function. If that reports false, then the function reports false without considering any more pairs of elements. Otherwise processing continues with the next pair, if any. If all corresponding elements are equal, but the right sequence is longer, < returns t, otherwise the function reports nil. The <= function tests each successive pair of corresponding elements recursively using the <= function. If this returns nil then the function returns nil without considering any more pairs. Otherwise processing continues with the next pair, if any. If all corresponding elements satisfy the test, but the left sequence is longer, then nil is returned. Otherwise t is returned.
The inequality relations exhibit symmetry, which means that the functions > and >= functions are equivalent, respectively, to < and <= with the order of the argument values reversed. For instance, the expression (< a b c) is equivalent to (> c b a) except for the difference in evaluation order of the a, b and c operands themselves. Any semantic description of < or <= applies, respectively, also to > or >= with the appropriate adjustment for argument order reversal.
(/= number*)
The arguments to /= may be numbers or characters. The /= function returns t if no two of its arguments are numerically equal. That is to say, if there exist some a and b which are distinct arguments such that (= a b) is true, then the function returns nil. Otherwise it returns t.
(max first-arg arg*)
(min first-arg arg*)
The max and min functions determine and return the highest or lowest value from among their arguments.
If only first-arg is given, that value is returned.
These functions are type generic, since they compare arguments using the same semantics as the less function.
If two or more arguments are given, then (max a b) is equivalent to (if (less a b) b a), and (min a b) is equivalent to (if (less a b) a b). If the operands do not have the same type, then one of them is converted to the type of the other; however, the original unconverted values are returned. For instance (max 4 3.0) yields the integer 4, not 4.0.
If three or more arguments are given, max and min reduce the arguments in a left-associative manner. Thus (max a b c) means (max (max a b) c).
(clamp low high val)
The clamp function clamps value val into the range low to high.
The clamp function returns low if val is less than low. If val is greater than or equal to low, but less than high, then it returns val. Otherwise it returns high.
More precisely, (clamp a b c) is equivalent to (max a (min b c)).
(bracket value level*)
The bracket function's arguments consist of one required value followed by n level arguments. The level arguments are optional; in other words, n may be zero.
The bracket function calculates the bracket of the value argument: a zero-based positional index of the value, in relation to the level arguments.
Each of the level arguments, of which there may be none, is associated with an integer index, starting at zero, in left-to-right order. The level arguments are examined in that order. When a level argument is encountered which exceeds value, that level argument's index is returned. If value exceeds all of the level arguments, then n is returned.
Determining whether value exceeds a level is performed using the less function.
(bracket 42) -> 0
(bracket 5 10) -> 0
(bracket 15 10) -> 1
(bracket 15 10 20) -> 1
(bracket 15 10 20 30) -> 1
(bracket 20 10 20 30) -> 2
(bracket 35 10 20 30) -> 3
(bracket "a" "aardvark" "zebra") -> 0
(bracket "ant" "aardvark" "zebra") -> 1
(bracket "zebu" "aardvark" "zebra") -> 2
(int-str string [radix])
(flo-str string)
(num-str string)
These functions extract numeric values from character string string. Leading whitespace in string, if any, is skipped. If no digits can be successfully extracted, then nil is returned. Trailing material which does not contribute to the number is ignored.
The int-str function converts a string of digits in the specified radix to an integer value. If radix isn't specified, it defaults to 10. Otherwise it must be an integer in the range 2 to 36, or else the character #\c.
For radices above 10, letters of the alphabet are used for digits: A represent a digit whose value is 10, B represents 11 and so forth until Z. Uppercase and lowercase letters are recognized. Any character which is not a digit of the specified radix is regarded as the start of trailing junk at which the extraction of the digits stops.
When radix is specified as the character object #\c, this indicates that a C-language-style integer constant should be recognized. If, after any optional sign, the remainder of string begins with the character pair 0x then that pair is considered removed from the string, and it is treated as base 16 (hexadecimal). If, after any optional sign, the remainder of string begins with a leading zero not followed by x, then the radix is taken to be 8 (octal). In scanning these formats, int-str function is not otherwise constrained by C language representational limitations. Specifically, the input values are taken to be the printed representation of arbitrary-precision integers and treated accordingly.
The flo-str function converts a floating-point decimal notation to a nearby floating point value. The material which contributes to the value is the longest match for optional leading space, followed by a mantissa which consists of an optional sign followed by a mixture of at least one digit, and at most one decimal point, optionally followed by an exponent part denoted by the letter E or e, an optional sign and one or more optional exponent digits. If the value specified by string is out of range of the floating-point representation, then nil is returned.
The num-str function converts a decimal notation to either an integer as if by a radix 10 application of int-str, or to a floating point value as if by flo-str. The floating point interpretation is chosen if the possibly empty initial sequence of digits (following any whitespace and optional sign) is followed by a period, or by e or E.
(int-flo float)
(flo-int integer)
These functions perform numeric conversion between integer and floating point type. The int-flo function returns an integer by truncating toward zero. The flo-int function returns an exact floating point value corresponding to integer, if possible, otherwise an approximation using a nearby floating point value.
(tofloat value)
(toint value [radix])
These functions convert value to floating-point or integer, respectively. The value can be of several types, including string.
If a floating-point value is passed into tofloat, or an integer value into toint, then that value is simply returned.
If value is a character, then it is treated as a string of length one containing that character.
If value is a string, then it is converted by tofloat as if by the function flo-str, and by toint as if by the function int-str.
If value is an integer, then it is converted by tofloat as if by the function flo-int.
If value is a floating-point number, then it is converted by toint as if by the function int-flo.
If value is a structure, then it is expected to implement the tofloat or toint method. This method is invoked by the same-named function, and the value is returned.
These variables hold, respectively, the most negative value of the fixnum integer type, and its most positive value. Integer values from fixnum-min to fixnum-max are all of type fixnum. Integers outside of this range are bignum integers.
(tofloatz value)
(tointz value [radix])
These functions are closely related to, respectively, tofloat and toint. They differ in that these functions return a floating-point or integer zero, respectively, in some situations in which those functions would return nil or throw an error.
Whereas those functions reject a value argument of nil, for that same argument tofloatz function returns 0.0 and tointz returns 0.
Likewise, in cases when value contains a string or character which cannot be converted to a number, and tofloat and toint would return nil, these functions return 0.0 and 0, respectively.
In other situations, these functions behave exactly like tofloat and toint.
These variables hold, respectively: the smallest positive floating-point value; the largest positive floating-point value; and the difference between 1.0 and the smallest representable value greater than 1.0.
flo-min and flo-max define the floating-point range, which consists of three regions: values from (- flo-max) to (- flo-min); the value 0.0, and values from flo-min to flo-max.
This variable holds an integer representing the number of decimal digits in a decimal floating-point number such that this number can be converted to a TXR floating-point number, and back to decimal, without a change in any of the digits. This holds regardless of the value of the number, provided that it does not exceed the floating-point range.
This variable holds an integer representing the maximum number of decimal digits required to capture the value of a floating-point number such that the resulting decimal form will convert back to the same floating-point number. See also the *print-flo-precision* variable.
These variables hold an approximation of the mathematical constants π and e. To four digits of precision, π is 3.142 and e is 2.718. The %pi% and %e% approximations are accurate to flo-dig decimal digits.
(digits number [radix])
The digits function returns a list of the digits of number represented in the base given by radix.
The number argument must be a nonnegative integer, and radix must be an integer greater than one.
If radix is omitted, it defaults to 10.
The return value is a list of the digits in descending order of significance: most significant to least significant. The digits are integers. For instance, if radix is 42, then the digits are integer values in the range 0 to 41.
The returned list always contains at least one element, and includes no leading zeros, except when number is zero. In that case, a one-element list containing zero is returned.
(digits 1234) -> (1 2 3 4)
(digits 1234567 1000) -> (1 234 567)
(digits 30 2) -> (1 1 1 1 0)
(digits 0) -> (0)
(digpow number [radix])
The digpow function decomposes the number argument into a power series whose terms add up to number.
The number argument must be a nonnegative integer, and radix must be an integer greater than one.
The returned power series consists of a list of nonnegative integers. It is formed from the digits of number in the given radix, which serve as coefficients which multiply successive powers of the radix, starting at the zeroth power (one).
The terms are given in decreasing order of significance: the term corresponding to the most significant digit of number, multiplying the highest power of radix, is listed first.
The returned list always contains at least one element, and includes no leading zeros, except when number is zero. In that case, a one-element list containing zero is returned.
(digpow 1234) -> (1000 200 30 4)
(digpow 1234567 1000) -> (1000000 234000 567)
(digpow 30 2) -> (16 8 4 2 0)
(digpow 0) -> (0)
(poly arg coeffs)
(rpoly arg coeffs)
The poly and rpoly functions evaluate a polynomial, for the given numeric argument value arg and the coefficients given by coeffs, a sequence of numbers.
If coeffs is an empty sequence, it denotes the zero polynomial, whose value is zero everywhere; the functions return zero in this case.
Otherwise, the poly function considers coeffs to hold the coefficients in the conventional order, namely in order of decreasing degree of polynomial term. The first element of coeffs is the leading coefficient, and the constant term appears as the last element.
The rpoly function takes the coefficients in opposite order: the first element of coeffs gives the constant term coefficient, and the last element gives the leading coefficient.
Note: except in the case of rpoly operating on a list or list-like sequence of coefficients, Horner's method of evaluation is used: a single result accumulator is initialized with zero, and then for each successive coefficient, in order of decreasing term degree, the accumulator is multiplied by the argument, and the coefficient is added. When rpoly operates on a list or list-like sequence, it makes a single pass through the coefficients in order, thus taking them in increasing term degree. It maintains two accumulators: one for successive powers of arg and one for the resulting value. For each coefficient, the power accumulator is updated by a multiplication by arg and then this value is multiplied by the coefficient, and that value is then added to the result accumulator.
;; 2
;; evaluate x + 2x + 3 for x = 10.
(poly 10 '(1 2 3)) -> 123
;; 2
;; evaluate 3x + 2x + 1 for x = 10.
(rpoly 10 '(1 2 3)) -> 321
(bignum-len arg)
The bignum-len function reports the machine-specific bignum order of the integer or character argument arg.
If arg is a character or fixnum integer, the function returns zero.
Otherwise arg is expected to be a bignum integer, and the function returns the number of "limbs" used for its representation, a positive integer.
Note: the bignum-len function is intended to be of use in algorithms whose performance benefits from ordering the operations on multiple integer operands according to the magnitudes of those operands. The function provides an estimate of magnitude which trades accuracy for efficiency.
(quantile p [group-size [rate]])
The quantile function returns a function which estimates a specific quantile of a set of real-valued samples. The desired quantile is indicated by the p parameter, which is a number in the range 0 to 1.0. If p is specified as 0.5, then the median is estimated. The p value of 0.9 leads to the estimation of the 90th percentile: a value such that approximately 90% of the samples are below that value.
If the group-size parameter is specified, it must be a positive integer. The returned function then operates in grouped mode. The rate parameter is relevant only to grouped mode. Grouped mode is described below.
The function returned by quantile maintains internal state in relation to calculating the quantile. The function may be called with any number of arguments, including none. It expects every argument to be either a number, or a sequence of numbers. These numbers are accumulated into the quantile calculation, and a revised estimate of the quantile is then returned.
Note: the algorithm used is the P-Squared algorithm invented in 1985 by Raj Jain and Imrich Chlamtac, which avoids accumulating and sorting the entire data set, while still obtaining good quality estimates of the quantile. The algorithm requires an initial seed of five samples. Then additional samples input into the algorithm produce quantile estimates. To eliminate this special case from the abstract interface, the TXR implementation is capable of producing an estimate when five or fewer samples have been presented, including none. In this low situation, the p value is ignored in reporting the estimate. When no samples have been given, the estimate is zero. When one sample has been given, the estimate is that sample itself. When between two and five samples have been given, the estimate is their median. Using the median as the estimate ensures a smooth transition from these early estimates into the estimates produced by the P-Squared algorithm. This is because the P-Squared algorithm always reports the value of the middle height accumulator as the estimate, and that accumulator's initial value is the median of the first five samples.
The function returned by quantile, though not accumulating all of the samples passed to it, nevertheless has a limited sample capacity, because the registers it uses for tracking the sample group positions are fixed-width integers. The sample capacity is approximately 4 times the value of fixnum-max.
(defparm q (quantile 0.9)) ;; create 90-th percentile accumulator
[q] -> 0.0 ;; no samples given: estimate is 0.
[q 3.14] -> 3.14 ;; one sample: estimate is that sample
[q 13.3 7.9 5.2 6.3] -> 7.9 ;; five samples: estimate is median.
[q 6.8 7.3 9.1 4.0] ;; more than five samples; estimate now
-> 8.44651234567901 ;; from P-Square algorithm
[q #(13.1 5 2.5)] ;; vector argument
-> 9.68660493827161
[q] -> 9.68660493827161 ;; no arguments: repeat current estimate
If the group-size argument is specified, then the quantile accumulator operates in grouped mode. Grouped mode allows infinite sample operation without overflow: an unlimited number of samples can be accepted. However, old samples lose their influence over the estimated value: newer samples are considered more significant than old samples.
In grouped mode, the quantile accumulator is reset to its initial state whenever group-size samples have been accumulated, and begins freshly calculating the quantile. Prior to the reset, an estimate is obtained and retained in an internal register. Going forward, this remembered previous estimate is blended in with the newly calculated estimate values, as described below. The cycle repeats itself whenever group-size samples accumulate: the state is reset, and the current estimate is loaded into the previous estimate register, which is then blended with newly computed values.
The rate parameter, whose default value is 0.999, controls the estimate blending. It should be a value between 0 and 1.
Upon each reset, a blend value register is initialized to 1.0. Each time a new sample is accumulated, the blend register is multiplied by the rate parameter, and the product is stored back into the blend register. Thus if the rate is between 0 and 1, exclusive, then the blend register exponentially decreases as the number of samples grows. The blend register indicates the fraction of the estimate which comes from the remembered previous estimate.
For instance, if the current blend value is 0.8, then the returned estimate value is 0.8 times the remembered previous estimate, plus 0.2 times the newly computed estimate for the current sample in the new group: the previous and current estimate are blended 80:20.
The default rate value of 0.999 is chosen for a slow transition to the new estimates, which helps to conceal inaccuracies in the algorithm associated with having accumulated a small number of samples. At this rate, it requires about 290 samples before the blend value drops to 75% of the old estimate.
If rate is specified as 0, then no blending of the previous estimate value takes place, since the blend factor will drop to zero upon the first sample being received after the group reset, causing the newly calculated estimates to be returned without blending. The previous sample groups therefore have no influence over newer estimates. If rate is specified as 1, then the blend factor will stay at 1, and so the estimate will forever remain at the previous value, ignoring the calculations driven by the new samples.
Note: it is recommended that if group-size is specified, the value should be at least several hundred. Too small a group size will prevent the estimation algorithm from settling on good results. The rate parameter should not much smaller than 1. A rate too low will cause the previous estimate's contribution to the quantile value to diminish, too quickly, before the new estimation settles.
These variables hold integer values suitable as arguments to the flo-set-round-mode function, which controls the rounding mode for the results of floating-point operations. These variables are only defined on platforms which support rounding control.
Their values have the following meanings:
(flo-get-round-mode)
(flo-set-round-mode mode)
Sometimes floating-point operations produce a result which requires more bits of precision than the floating point representation can provide. A representable floating-point value must be substituted for the true result and yielded by the operation.
On platforms which support rounding control, these functions are provided for selecting the decision procedure by which the floating-point representation is taken.
The flo-get-round-mode returns the current rounding mode. The rounding mode is represented by an integer value which is either equal to one of the four variables flo-near, flo-down, flo-up and flo-zero, or else some other value specific to the host environment. Initially, the value is that of flo-near. Otherwise, the value returned is that which was stored by the most recent successful call to flo-set-round-mode.
The flo-set-round-mode function changes the rounding mode. The argument to its mode parameter may be the value of one of the above four variables, or else some other value supported by the host environment's fesetround C library function.
The flo-set-round-mode function returns t if it is successful, otherwise the return value is nil and the rounding mode is not changed.
If a value is passed to flo-set-round-mode which is not the value of one of the above four rounding mode variables, and the function succeeds anyway, then the rounding behavior of floating-point operations depends on the host environment's interpretation of that value.
The following functions are defined, if they are available from the host platform. They corresponds to same-named functions in the ISO C language standard, which appeared in the 1999 revision ("C99").
Even if some of these functions happen not to be defined, it is nevertheless possible to define them as methods in a user-defined arithmetic structure. See the section User-Defined Arithmetic Types below
(cbrt arg)
(erf arg)
(erfc arg)
(exp10 arg)
(exp2 arg)
(expm1 arg)
(gamma arg)
(j0 arg)
(j1 arg)
(lgamma arg)
(log1p arg)
(logb arg)
(nearbyint arg)
(rint arg)
(significand arg)
(tgamma arg)
(y0 arg)
(y1 arg)
These are one-argument functions, which take a numeric argument, and return a floating-point result.
(copysign arg1 arg2)
(drem arg1 arg2)
(fdim arg1 arg2)
(fmax arg1 arg2)
(fmin arg1 arg2)
(hypot arg1 arg2)
(jn arg1 arg2)
(ldexp arg1 arg2)
(nextafter arg1 arg2)
(remainder arg1 arg2)
(scalb arg1 arg2)
(scalbln arg1 arg2)
(yn arg1 arg2)
These are two-argument functions, which take numeric arguments, and return a floating-point result.
In TXR Lisp, similarly to Common Lisp, bit operations on integers are based on a concept that might be called "infinite two's complement". Under infinite two's complement, a positive number is regarded as having a binary representation prefixed by an infinite stream of zero digits (for example 1 is ...00001). A negative number in infinite two's complement is the bitwise negation of its positive counterpart, plus one: it carries an infinite prefix of 1 digits. So for instance the number -1 is represented by ...11111111: an infinite sequence of 1 bits. There is no specific sign bit; any operation which produces such an infinite sequence of 1 digits on the left gives rise to a negative number. For instance, consider the operation of computing the bitwise complement of the number 1. Since the number 1 is represented as ...0000001, its complement is ...11111110. Each one of the 0 digits in the infinite sequence is replaced by 1, And this leading sequence means that the number is negative, in fact corresponding to the two's complement representation of the value -2. Hence, the infinite digit concept corresponds to an arithmetic interpretation.
In fact TXR Lisp's bignum integers do not use a two's complement representation internally. Numbers are represented as an array which holds a pure binary number. A separate field indicates the sign: negative, or nonnegative. That negative numbers appear as two's complement under the bit operations is merely a carefully maintained illusion (which makes bit operations on negative numbers more expensive).
The logtrunc function, as well as a feature of the lognot function, allow bit manipulation code to be written which works with positive numbers only, even if complements are required. The trade off is that the application has to manage a limit on the number of bits.
(logand integer*)
(logior integer*)
(logxor int1 int2)
These operations perform the familiar bitwise and, inclusive or, and exclusive or operations, respectively. Positive values inputs are treated as pure binary numbers. Negative inputs are treated as infinite-bit two's complement.
For example (logand -2 7) produces 6. This is because -2 is ...111110 in infinite-bit two's complement. And-ing this value with 7 (or ...000111) produces 110.
The logand and logior functions are variadic, and may be called with zero, one, two, or more input values. If logand is called with no arguments, it produces the value -1 (all bits 1). If logior is called with no arguments it produces zero. In the one-argument case, the functions just return their argument value.
In the two-argument case, one of the operands may be a character, if the other operand is a fixnum integer. The character operand is taken to be an integer corresponding to the character value's Unicode code point value. The resulting value is regarded as a Unicode code point and converted to a character value accordingly.
When three or more arguments are specified, the operation's semantics is that of a left-associative reduction through two-argument invocations, so that the three-argument case (logand a b c) is equivalent to the expression (logand (logand a b) c), which features two two-argument cases.
(logtest int1 int2)
The logtest function returns true if int1 and int2 have bits in common. The following equivalence holds:
(logtest a b) <--> (not (zerop (logand a b)))
(lognot value [bits])
(logtrunc value bits)
The lognot function performs a bitwise complement of value. When the one-argument form of lognot is used, then if value is nonnegative, then the result is negative, and vice versa, according to the infinite-bit two's complement representation. For instance (lognot -2) is 1, and (lognot 1) is -2.
The two-argument form of lognot produces a truncated complement. Conceptually, a bitwise complement is first calculated, and then the resulting number is truncated to the number of bits given by bits, which must be a nonnegative integer. The following equivalence holds:
(lognot a b) <--> (logtrunc (lognot a) b)
The logtrunc function truncates the integer value to the specified number of bits. If value is negative, then the two's complement representation is truncated. The return value of logtrunc is always a nonnegative integer.
(sign-extend value bits)
The sign-extend function first truncates the infinite-bit two's complement representation of the integer value to the specified number of bits, similarly to the logtrunc function. Then, this truncated value is regarded as a bits-wide two's complement integer. The value of this integer is calculated and returned.
(sign-extend 127 8) -> 127
(sign-extend 128 8) -> -128
(sign-extend 129 8) -> -127
(sign-extend 255 8) -> -1
(sign-extend 256 8) -> 0
(sign-extend -1 8) -> -1
(sign-extend -255 8) -> 0
(ash value bits)
The ash function shifts value by the specified number of bits producing a new value. If bits is positive, then a left shift takes place. If bits is negative, then a right shift takes place. If bits is zero, then value is returned unaltered. For positive numbers, a left shift by n bits is equivalent to a multiplication by two to the power of n, or (expt 2 n). A right shift by n bits of a positive integer is equivalent to integer division by (expt 2 n), with truncation toward zero. For negative numbers, the bit shift is performed as if on the two's complement representation. Under the infinite two's complement representation, a right shift does not exhaust the infinite sequence of 1 digits which extends to the left. Thus if -4 is shifted right it becomes -2 because the bitwise representations of these values are ...111100 and ...11110.
(bit value bit)
The bit function tests whether the integer or character value has a 1 in bit position bit. The bit argument must be a nonnegative integer. A value of bit of zero indicates the least-significant-bit position of value.
The bit function has a Boolean result, returning the symbol t if bit bit of value is set, otherwise nil.
If value is negative, it is treated as if it had an infinite-bit two's complement representation. For instance, if value is -2, then the bit function returns nil for a bit value of zero, and t for all other values, since the infinite bit two's complement representation of -2 is ...11110.
(mask integer*)
The mask function takes zero or more integer arguments, and produces an integer value which corresponds to a bitmask made up of the bit positions specified by the integer arguments.
If mask is called with no arguments, then the return value is zero.
If mask is called with a single argument integer then the return value is the same as that of the expression (ash 1 <integer>): the value 1 shifted left by integer bit positions. If integer is zero, then the result is 1; if integer is 1, the result is 2 and so forth. If value is negative, then the result is zero.
If mask is called with two or more arguments, then the result is a bitwise or of the masks individually computed for each of the values.
In other words, the following equivalences hold:
(mask) <--> 0
(mask a) <--> (ash 1 a)
(mask a b c ...) <--> (logior (mask a) (mask b) (mask c) ...)
(bitset integer)
The bitset function returns a list of the positions of bits which have a value of 1 in a positive integer argument, or the positions of bits which have a value of zero in a negative integer argument. The positions are ordered from least to greatest. The least significant bit has position zero. If integer is zero, the empty list nil is returned.
A negative integer is treated as an infinite-bit two's complement representation.
The argument may be a character.
If integer x is nonnegative, the following equivalence holds:
x <--> [apply mask (bitset x)]
That is to say, the value of x may be reconstituted by applying the bit positions returned by bitset as arguments to the mask function.
The value of a negative x may be reconstituted from its bitset as follows:
x <--> (pred (- [apply mask (bitset x)]))
also, more trivially, thus:
x <--> (- [apply mask (bitset (- x))])
(width integer*)
A two's complement representation of an integer consists of a sign bit and a mantissa field. The width function computes the minimum number of bits required for the mantissa portion of the two's complement representation of the integer argument.
For a nonnegative argument, the width also corresponds to the number of bits required for a natural binary representation of that value.
Two integer values have a width of zero, namely 0 and -1. This means that these two values can be represented in a one-bit two's complement, consisting of only a sign bit: the one-bit two's complement bitfield 1 denotes -1, and 0 denotes 0.
Similarly, two integer values have a width of 1: 1 and -2. The two-bit two's complement bitfield 01 denotes 1, and 10 denotes -2.
The argument may be a character.
(logcount integer)
The logcount function considers integer to have a two's complement representation. If the integer is positive, it returns the count of bits in that representation whose value is 1. If integer is negative, it returns the count of zero bits instead. If integer is zero, the value returned is zero.
The argument may be a character.
(set-mask place integer*)
(clear-mask place integer*)
The set-mask and clear-mask macros set to 1 and 0, respectively, the bits in place corresponding to bits that are equal to 1 in the mask resulting from applying the inclusive or operation to the integer arguments. The following equivalences hold:
(set-mask place integer ...)
<--> (set place (logior place integer ...)
(clear-mask place integer ...)
<--> (set place (logand place (lognot (logior integer ...))))
TXR Lisp makes it possible for the user application program to define structure types which can participate in arithmetic operations as if they were numbers. Under most arithmetic functions, a structure object may be used instead of a number, if that structure object implements a specific method which is required by that arithmetic function.
The following paragraphs give general remarks about the method conventions. Not all arithmetic and bit manipulation functions have a corresponding method, and a small number of functions do not follow these conventions.
In the simplest case of arithmetic functions which are unary, the method takes no argument other than the object itself. Most unary arithmetic functions expect a structure argument to have a method which has the same name as that function. For instance, if x is a structure, then (cos x) will invoke x.(cos). If x has no cos method, then an error exception is thrown. A few unary methods are not named after the corresponding function. The unary case of the - function expects an object to have a method named neg; thus, (- x) invokes x.(neg). Unary division requires a method called recip; thus, (/ x), invokes x.(recip).
When a structure object is used as an argument in a two-argument (binary) arithmetic function, there are several cases to consider. If the left argument to a binary function is an object, then that object is expected to support a binary method. That method is called with two arguments: the object itself, of course, and the right argument of the arithmetic operation. In this case, the method is named after the function. For instance, if x is an object, then (+ x 3) invokes x.(+ 3). If the right argument, and only the right argument, of a binary operation is an object, then the situation falls into two cases depending on whether the operation is commutative. If the operation is commutative, then the same method is used as in the case when the object is the left argument. The arguments are merely reversed. Thus (+ 3 x) also invokes x.(+ 3). If the operation is not commutative, then the object must supply an alternative method. For most functions, that method is named by a symbol whose name begins with a r- prefix. For instance (mod x 5) invokes x.(mod 5) whereas (mod 5 x) invokes x.(r-mod 5). Note: the "r" may be remembered as indicating that the object is the right argument of the binary operation or that the arguments are reversed. Two functions do not follow the r- convention. These are - and /. For these, the methods used for the object as a right argument, respectively, are -- and //. Thus (/ 5 x) invokes x.(// 5) and (- 5 x) invokes x.(-- 5). Several binary functions do not support an object as the right argument. These are sign-extend, ash and bit.
Variadic arithmetic functions, when given three or more arguments, are regarded as performing a left-associative decimation of the arguments through a binary function. Thus for instance (- 1 x 4) is understood as (- (- 1 x) 4) where x.(-- 1) is evaluated first. If that method yields an object o then o.(- 4) is invoked.
Certain variadic arithmetic functions, if invoked with one argument, just return that argument: for instance, + and * are in this category. A special concession exists in these functions: if their one and only argument is a structure, then that structure is returned without any error checking, even if it implements no methods related to arithmetic.
The following sections describe each of the methods that must be implemented by an object for the associated arithmetic function to work with that object, either at all, or in a specific argument position, as the case may be. These methods are not provided by TXR Lisp; the application is required to provide them.
obj.(tofloat)
The tofloat method is invoked when a structure is used as the argument to the tofloat function.
If an object obj is passed to the function as (tofloat obj) then, effectively, the method call obj.(tofloat) takes place, and its return value is taken as the result of the operation.
The method should return a floating-point value. It is also permissible for the method to return nil, in which case if it is invoked via tofloatz, that function will replace the nil return with value of 0.0.
obj.(toint)
The toint method is invoked when a structure is used as the argument to the toint function.
If an object obj is passed to the function as (toint obj) then, effectively, the method call obj.(toint) takes place, and its return value is taken as the result of the operation.
The method should return an integer value. It is permissible for the method to return nil, in which case if it is invoked via tointz, that function will replace the nil return with value of 0.
obj.(+ arg)
The + method is invoked when a structure is used as an argument to the + function together with at least one other operand.
If an object obj is combined with an argument arg, either as (+ obj arg) or as (+ arg obj) then, effectively, the method call obj.(+ arg) takes place, and its return value is taken as the result of the operation.
obj.(- arg)
The - method is invoked when the structure obj is used as the left argument of the - function.
If an object obj is combined with an argument arg, as (- obj arg) then, effectively, the method call obj.(- arg) takes place, and its return value is taken as the result of the operation.
obj.(-- arg)
The -- method is invoked when the structure obj is used as the right argument of the - function.
If an object obj is combined with an argument arg, as (- arg obj) then, effectively, the method call obj.(-- arg) takes place, and its return value is taken as the result of the operation.
obj.(neg)
The neg method is invoked when the structure obj is used as the sole argument to the - function.
If an object obj is passed to the function as (- obj) then, effectively, the method call obj.(neg) takes place, and its return value is taken as the result of the operation.
obj.(* arg)
The * method is invoked when a structure is used as an argument to the * function together with at least one other operand.
If an object obj is combined with an argument arg, either as (* obj arg) or as (* arg obj) then, effectively, the method call obj.(* arg) takes place, and its return value is taken as the result of the operation.
obj.(/ arg)
The / method is invoked when the structure obj is used as the left argument of the / function.
If an object obj is combined with an argument arg, as (/ obj arg) then, effectively, the method call obj.(/ arg) takes place, and its return value is taken as the result of the operation.
obj.(// arg)
The // method is invoked when the structure obj is used as the right argument of the / function.
If an object obj is combined with an argument arg, as (/ arg obj) then, effectively, the method call obj.(// arg) takes place, and its return value is taken as the result of the operation.
obj.(recip)
The recip method is invoked when the structure obj is used as the sole argument to the / function.
If an object obj is passed to the function as (/ obj) then, effectively, the method call obj.(recip) takes place, and its return value is taken as the result of the operation.
obj.(abs)
The abs method is invoked when a structure is used as the argument to the abs function.
If an object obj is passed to the function as (abs obj) then, effectively, the method call obj.(abs) takes place, and its return value is taken as the result of the operation.
obj.(signum)
The signum method is invoked when a structure is used as the argument to the signum function.
If an object obj is passed to the function as (signum obj) then, effectively, the method call obj.(signum) takes place, and its return value is taken as the result of the operation.
obj.(trunc arg)
The trunc method is invoked when the structure obj is used as the left argument of the trunc function.
If an object obj is combined with an argument arg, as (trunc obj arg) then, effectively, the method call obj.(trunc arg) takes place, and its return value is taken as the result of the operation.
obj.(r-trunc arg)
The r-trunc method is invoked when the structure obj is used as the right argument of the trunc function.
If an object obj is combined with an argument arg, as (trunc arg obj) then, effectively, the method call obj.(r-trunc arg) takes place, and its return value is taken as the result of the operation.
obj.(trunc1)
The trunc1 method is invoked when the structure obj is used as the sole argument to the trunc function.
If an object obj is passed to the function as (trunc obj) then, effectively, the method call obj.(trunc1) takes place, and its return value is taken as the result of the operation.
obj.(mod arg)
The mod method is invoked when the structure obj is used as the left argument of the mod function.
If an object obj is combined with an argument arg, as (mod obj arg) then, effectively, the method call obj.(mod arg) takes place, and its return value is taken as the result of the operation.
obj.(r-mod arg)
The r-mod method is invoked when the structure obj is used as the right argument of the mod function.
If an object obj is combined with an argument arg, as (mod arg obj) then, effectively, the method call obj.(r-mod arg) takes place, and its return value is taken as the result of the operation.
obj.(expt arg)
The expt method is invoked when the structure obj is used as the left argument of the expt function.
If an object obj is combined with an argument arg, as (expt obj arg) then, effectively, the method call obj.(expt arg) takes place, and its return value is taken as the result of the operation.
obj.(r-expt arg)
The r-expt method is invoked when the structure obj is used as the right argument of the expt function.
If an object obj is combined with an argument arg, as (expt arg obj) then, effectively, the method call obj.(r-expt arg) takes place, and its return value is taken as the result of the operation.
obj.(exptmod arg1 arg2)
The exptmod method is invoked when the structure obj is used as the left argument of the exptmod function.
If an object obj is combined with arguments arg1 and arg2, as (exptmod obj arg1 arg2) then, effectively, the method call obj.(exptmod arg1 arg2) takes place, and its return value is taken as the result of the operation.
Note: the exptmod function doesn't support structure objects in the second and third argument positions. The exponent and base arguments must be integers.
obj.(isqrt)
The isqrt method is invoked when a structure is used as the argument to the isqrt function.
If an object obj is passed to the function as (isqrt obj) then, effectively, the method call obj.(isqrt) takes place, and its return value is taken as the result of the operation.
obj.(square)
The square method is invoked when a structure is used as the argument to the square function.
If an object obj is passed to the function as (square obj) then, effectively, the method call obj.(square) takes place, and its return value is taken as the result of the operation.
obj.(> arg)
The > method is invoked when the > function is invoked with two operands, and the structure obj is the left operand. The method is also invoked when the < function is invoked with two operands, and obj is the right operand.
If an object obj is combined with an argument arg, either as (> obj arg) or as (< arg obj) then, effectively, the method call obj.(> arg) takes place, and its return value is taken as the result of the operation.
obj.(< arg)
The < method is invoked when the < function is invoked with two operands, and the structure obj is the left operand. The method is also invoked when the > function is invoked with two operands, and obj is the right operand.
If an object obj is combined with an argument arg, either as (< obj arg) or as (> arg obj) then, effectively, the method call obj.(< arg) takes place, and its return value is taken as the result of the operation.
obj.(>= arg)
The >= method is invoked when the >= function is invoked with two operands, and the structure obj is the left operand. The method is also invoked when the <= function is invoked with two operands, and obj is the right operand.
If an object obj is combined with an argument arg, either as (>= obj arg) or as (<= arg obj) then, effectively, the method call obj.(>= arg) takes place, and its return value is taken as the result of the operation.
obj.(<= arg)
The <= method is invoked when the <= function is invoked with two operands, and the structure obj is the left operand. The method is also invoked when the >= function is invoked with two operands, and obj is the right operand.
If an object obj is combined with an argument arg, either as (<= obj arg) or as (>= arg obj) then, effectively, the method call obj.(<= arg) takes place, and its return value is taken as the result of the operation.
obj.(= arg)
The = method is invoked when a structure is used as an argument to the = function.
If an object obj is combined with an argument arg, either as (= obj arg) or as (= arg obj) then, effectively, the method call obj.(= arg) takes place, and its return value is taken as the result of the operation.
obj.(zerop)
The zerop method is invoked when a structure is used as the argument to the zerop function.
If an object obj is passed to the function as (zerop obj) then, effectively, the method call obj.(zerop) takes place, and its return value is taken as the result of the operation.
obj.(plusp)
The plusp method is invoked when a structure is used as the argument to the plusp function.
If an object obj is passed to the function as (plusp obj) then, effectively, the method call obj.(plusp) takes place, and its return value is taken as the result of the operation.
obj.(minusp)
The minusp method is invoked when a structure is used as the argument to the minusp function.
If an object obj is passed to the function as (minusp obj) then, effectively, the method call obj.(minusp) takes place, and its return value is taken as the result of the operation.
obj.(evenp)
The evenp method is invoked when a structure is used as the argument to the evenp function.
If an object obj is passed to the function as (evenp obj) then, effectively, the method call obj.(evenp) takes place, and its return value is taken as the result of the operation.
obj.(oddp)
The oddp method is invoked when a structure is used as the argument to the oddp function.
If an object obj is passed to the function as (oddp obj) then, effectively, the method call obj.(oddp) takes place, and its return value is taken as the result of the operation.
obj.(floor arg)
The floor method is invoked when the structure obj is used as the left argument of the floor function.
If an object obj is combined with an argument arg, as (floor obj arg) then, effectively, the method call obj.(floor arg) takes place, and its return value is taken as the result of the operation.
obj.(r-floor arg)
The r-floor method is invoked when the structure obj is used as the right argument of the floor function.
If an object obj is combined with an argument arg, as (floor arg obj) then, effectively, the method call obj.(r-floor arg) takes place, and its return value is taken as the result of the operation.
obj.(floor1)
The floor1 method is invoked when the structure obj is used as the sole argument to the floor function.
If an object obj is passed to the function as (floor obj) then, effectively, the method call obj.(floor1) takes place, and its return value is taken as the result of the operation.
obj.(ceil arg)
The ceil method is invoked when the structure obj is used as the left argument of the ceil function.
If an object obj is combined with an argument arg, as (ceil obj arg) then, effectively, the method call obj.(ceil arg) takes place, and its return value is taken as the result of the operation.
obj.(r-ceil arg)
The r-ceil method is invoked when the structure obj is used as the right argument of the ceil function.
If an object obj is combined with an argument arg, as (ceil arg obj) then, effectively, the method call obj.(r-ceil arg) takes place, and its return value is taken as the result of the operation.
obj.(ceil1)
The ceil1 method is invoked when the structure obj is used as the sole argument to the ceil function.
If an object obj is passed to the function as (ceil obj) then, effectively, the method call obj.(ceil1) takes place, and its return value is taken as the result of the operation.
obj.(round arg)
The round method is invoked when the structure obj is used as the left argument of the round function.
If an object obj is combined with an argument arg, as (round obj arg) then, effectively, the method call obj.(round arg) takes place, and its return value is taken as the result of the operation.
obj.(r-round arg)
The r-round method is invoked when the structure obj is used as the right argument of the round function.
If an object obj is combined with an argument arg, as (round arg obj) then, effectively, the method call obj.(r-round arg) takes place, and its return value is taken as the result of the operation.
obj.(round1)
The round1 method is invoked when the structure obj is used as the sole argument to the round function.
If an object obj is passed to the function as (round obj) then, effectively, the method call obj.(round1) takes place, and its return value is taken as the result of the operation.
obj.(sin)
The sin method is invoked when a structure is used as the argument to the sin function.
If an object obj is passed to the function as (sin obj) then, effectively, the method call obj.(sin) takes place, and its return value is taken as the result of the operation.
obj.(cos)
The cos method is invoked when a structure is used as the argument to the cos function.
If an object obj is passed to the function as (cos obj) then, effectively, the method call obj.(cos) takes place, and its return value is taken as the result of the operation.
obj.(tan)
The tan method is invoked when a structure is used as the argument to the tan function.
If an object obj is passed to the function as (tan obj) then, effectively, the method call obj.(tan) takes place, and its return value is taken as the result of the operation.
obj.(asin)
The asin method is invoked when a structure is used as the argument to the asin function.
If an object obj is passed to the function as (asin obj) then, effectively, the method call obj.(asin) takes place, and its return value is taken as the result of the operation.
obj.(acos)
The acos method is invoked when a structure is used as the argument to the acos function.
If an object obj is passed to the function as (acos obj) then, effectively, the method call obj.(acos) takes place, and its return value is taken as the result of the operation.
obj.(atan)
The atan method is invoked when a structure is used as the argument to the atan function.
If an object obj is passed to the function as (atan obj) then, effectively, the method call obj.(atan) takes place, and its return value is taken as the result of the operation.
obj.(atan2 arg)
The atan2 method is invoked when the structure obj is used as the left argument of the atan2 function.
If an object obj is combined with an argument arg, as (atan2 obj arg) then, effectively, the method call obj.(atan2 arg) takes place, and its return value is taken as the result of the operation.
obj.(r-atan2 arg)
The r-atan2 method is invoked when the structure obj is used as the right argument of the atan2 function.
If an object obj is combined with an argument arg, as (atan2 arg obj) then, effectively, the method call obj.(r-atan2 arg) takes place, and its return value is taken as the result of the operation.
obj.(sinh)
The sinh method is invoked when a structure is used as the argument to the sinh function.
If an object obj is passed to the function as (sinh obj) then, effectively, the method call obj.(sinh) takes place, and its return value is taken as the result of the operation.
obj.(cosh)
The cosh method is invoked when a structure is used as the argument to the cosh function.
If an object obj is passed to the function as (cosh obj) then, effectively, the method call obj.(cosh) takes place, and its return value is taken as the result of the operation.
obj.(tanh)
The tanh method is invoked when a structure is used as the argument to the tanh function.
If an object obj is passed to the function as (tanh obj) then, effectively, the method call obj.(tanh) takes place, and its return value is taken as the result of the operation.
obj.(asinh)
The asinh method is invoked when a structure is used as the argument to the asinh function.
If an object obj is passed to the function as (asinh obj) then, effectively, the method call obj.(asinh) takes place, and its return value is taken as the result of the operation.
obj.(acosh)
The acosh method is invoked when a structure is used as the argument to the acosh function.
If an object obj is passed to the function as (acosh obj) then, effectively, the method call obj.(acosh) takes place, and its return value is taken as the result of the operation.
obj.(atanh)
The atanh method is invoked when a structure is used as the argument to the atanh function.
If an object obj is passed to the function as (atanh obj) then, effectively, the method call obj.(atanh) takes place, and its return value is taken as the result of the operation.
obj.(log)
The log method is invoked when a structure is used as the argument to the log function.
If an object obj is passed to the function as (log obj) then, effectively, the method call obj.(log) takes place, and its return value is taken as the result of the operation.
obj.(log2)
The log2 method is invoked when a structure is used as the argument to the log2 function.
If an object obj is passed to the function as (log2 obj) then, effectively, the method call obj.(log2) takes place, and its return value is taken as the result of the operation.
obj.(log10)
The log10 method is invoked when a structure is used as the argument to the log10 function.
If an object obj is passed to the function as (log10 obj) then, effectively, the method call obj.(log10) takes place, and its return value is taken as the result of the operation.
obj.(exp)
The exp method is invoked when a structure is used as the argument to the exp function.
If an object obj is passed to the function as (exp obj) then, effectively, the method call obj.(exp) takes place, and its return value is taken as the result of the operation.
obj.(sqrt)
The sqrt method is invoked when a structure is used as the argument to the sqrt function.
If an object obj is passed to the function as (sqrt obj) then, effectively, the method call obj.(sqrt) takes place, and its return value is taken as the result of the operation.
obj.(logand arg)
The logand method is invoked when a structure is used as an argument to the logand function together with at least one other operand.
If an object obj is combined with an argument arg, either as (logand obj arg) or as (logand arg obj) then, effectively, the method call obj.(logand arg) takes place, and its return value is taken as the result of the operation.
obj.(logior arg)
The logior method is invoked when a structure is used as an argument to the logior function together with at least one other operand.
If an object obj is combined with an argument arg, either as (logior obj arg) or as (logior arg obj) then, effectively, the method call obj.(logior arg) takes place, and its return value is taken as the result of the operation.
obj.(lognot arg)
The lognot method is invoked when the structure obj is used as the left argument of the lognot function.
If an object obj is combined with an argument arg, as (lognot obj arg) then, effectively, the method call obj.(lognot arg) takes place, and its return value is taken as the result of the operation.
obj.(r-lognot arg)
The r-lognot method is invoked when the structure obj is used as the right argument of the lognot function.
If an object obj is combined with an argument arg, as (lognot arg obj) then, effectively, the method call obj.(r-lognot arg) takes place, and its return value is taken as the result of the operation.
obj.(lognot1)
The lognot1 method is invoked when the structure obj is used as the sole argument to the lognot function.
If an object obj is passed to the function as (lognot obj) then, effectively, the method call obj.(lognot1) takes place, and its return value is taken as the result of the operation.
obj.(logtrunc arg)
The logtrunc method is invoked when the structure obj is used as the left argument of the logtrunc function.
If an object obj is combined with an argument arg, as (logtrunc obj arg) then, effectively, the method call obj.(logtrunc arg) takes place, and its return value is taken as the result of the operation.
obj.(r-logtrunc arg)
The r-logtrunc method is invoked when the structure obj is used as the right argument of the logtrunc function.
If an object obj is combined with an argument arg, as (logtrunc arg obj) then, effectively, the method call obj.(r-logtrunc arg) takes place, and its return value is taken as the result of the operation.
obj.(sign-extend arg)
The sign-extend method is invoked when the structure obj is used as the left argument of the sign-extend function.
If an object obj is combined with an argument arg, as (sign-extend obj arg) then, effectively, the method call obj.(sign-extend arg) takes place, and its return value is taken as the result of the operation.
obj.(cbrt)
The cbrt method is invoked when a structure is used as the argument to the cbrt function.
If an object obj is passed to the function as (cbrt obj) then, effectively, the method call obj.(cbrt) takes place, and its return value is taken as the result of the operation.
obj.(erf)
The erf method is invoked when a structure is used as the argument to the erf function.
If an object obj is passed to the function as (erf obj) then, effectively, the method call obj.(erf) takes place, and its return value is taken as the result of the operation.
obj.(erfc)
The erfc method is invoked when a structure is used as the argument to the erfc function.
If an object obj is passed to the function as (erfc obj) then, effectively, the method call obj.(erfc) takes place, and its return value is taken as the result of the operation.
obj.(exp10)
The exp10 method is invoked when a structure is used as the argument to the exp10 function.
If an object obj is passed to the function as (exp10 obj) then, effectively, the method call obj.(exp10) takes place, and its return value is taken as the result of the operation.
obj.(exp2)
The exp2 method is invoked when a structure is used as the argument to the exp2 function.
If an object obj is passed to the function as (exp2 obj) then, effectively, the method call obj.(exp2) takes place, and its return value is taken as the result of the operation.
obj.(expm1)
The expm1 method is invoked when a structure is used as the argument to the expm1 function.
If an object obj is passed to the function as (expm1 obj) then, effectively, the method call obj.(expm1) takes place, and its return value is taken as the result of the operation.
obj.(gamma)
The gamma method is invoked when a structure is used as the argument to the gamma function.
If an object obj is passed to the function as (gamma obj) then, effectively, the method call obj.(gamma) takes place, and its return value is taken as the result of the operation.
obj.(j0)
The j0 method is invoked when a structure is used as the argument to the j0 function.
If an object obj is passed to the function as (j0 obj) then, effectively, the method call obj.(j0) takes place, and its return value is taken as the result of the operation.
obj.(j1)
The j1 method is invoked when a structure is used as the argument to the j1 function.
If an object obj is passed to the function as (j1 obj) then, effectively, the method call obj.(j1) takes place, and its return value is taken as the result of the operation.
obj.(lgamma)
The lgamma method is invoked when a structure is used as the argument to the lgamma function.
If an object obj is passed to the function as (lgamma obj) then, effectively, the method call obj.(lgamma) takes place, and its return value is taken as the result of the operation.
obj.(log1p)
The log1p method is invoked when a structure is used as the argument to the log1p function.
If an object obj is passed to the function as (log1p obj) then, effectively, the method call obj.(log1p) takes place, and its return value is taken as the result of the operation.
obj.(logb)
The logb method is invoked when a structure is used as the argument to the logb function.
If an object obj is passed to the function as (logb obj) then, effectively, the method call obj.(logb) takes place, and its return value is taken as the result of the operation.
obj.(nearbyint)
The nearbyint method is invoked when a structure is used as the argument to the nearbyint function.
If an object obj is passed to the function as (nearbyint obj) then, effectively, the method call obj.(nearbyint) takes place, and its return value is taken as the result of the operation.
obj.(rint)
The rint method is invoked when a structure is used as the argument to the rint function.
If an object obj is passed to the function as (rint obj) then, effectively, the method call obj.(rint) takes place, and its return value is taken as the result of the operation.
obj.(significand)
The significand method is invoked when a structure is used as the argument to the significand function.
If an object obj is passed to the function as (significand obj) then, effectively, the method call obj.(significand) takes place, and its return value is taken as the result of the operation.
obj.(tgamma)
The tgamma method is invoked when a structure is used as the argument to the tgamma function.
If an object obj is passed to the function as (tgamma obj) then, effectively, the method call obj.(tgamma) takes place, and its return value is taken as the result of the operation.
obj.(y0)
The y0 method is invoked when a structure is used as the argument to the y0 function.
If an object obj is passed to the function as (y0 obj) then, effectively, the method call obj.(y0) takes place, and its return value is taken as the result of the operation.
obj.(y1)
The y1 method is invoked when a structure is used as the argument to the y1 function.
If an object obj is passed to the function as (y1 obj) then, effectively, the method call obj.(y1) takes place, and its return value is taken as the result of the operation.
obj.(r-copysign arg)
The r-copysign method is invoked when the structure obj is used as the right argument of the copysign function.
If an object obj is combined with an argument arg, as (copysign arg obj) then, effectively, the method call obj.(r-copysign arg) takes place, and its return value is taken as the result of the operation.
obj.(r-drem arg)
The r-drem method is invoked when the structure obj is used as the right argument of the drem function.
If an object obj is combined with an argument arg, as (drem arg obj) then, effectively, the method call obj.(r-drem arg) takes place, and its return value is taken as the result of the operation.
obj.(r-fdim arg)
The r-fdim method is invoked when the structure obj is used as the right argument of the fdim function.
If an object obj is combined with an argument arg, as (fdim arg obj) then, effectively, the method call obj.(r-fdim arg) takes place, and its return value is taken as the result of the operation.
obj.(r-fmax arg)
The r-fmax method is invoked when the structure obj is used as the right argument of the fmax function.
If an object obj is combined with an argument arg, as (fmax arg obj) then, effectively, the method call obj.(r-fmax arg) takes place, and its return value is taken as the result of the operation.
obj.(r-fmin arg)
The r-fmin method is invoked when the structure obj is used as the right argument of the fmin function.
If an object obj is combined with an argument arg, as (fmin arg obj) then, effectively, the method call obj.(r-fmin arg) takes place, and its return value is taken as the result of the operation.
obj.(r-hypot arg)
The r-hypot method is invoked when the structure obj is used as the right argument of the hypot function.
If an object obj is combined with an argument arg, as (hypot arg obj) then, effectively, the method call obj.(r-hypot arg) takes place, and its return value is taken as the result of the operation.
obj.(r-jn arg)
The r-jn method is invoked when the structure obj is used as the right argument of the jn function.
If an object obj is combined with an argument arg, as (jn arg obj) then, effectively, the method call obj.(r-jn arg) takes place, and its return value is taken as the result of the operation.
obj.(r-ldexp arg)
The r-ldexp method is invoked when the structure obj is used as the right argument of the ldexp function.
If an object obj is combined with an argument arg, as (ldexp arg obj) then, effectively, the method call obj.(r-ldexp arg) takes place, and its return value is taken as the result of the operation.
obj.(r-nextafter arg)
The r-nextafter method is invoked when the structure obj is used as the right argument of the nextafter function.
If an object obj is combined with an argument arg, as (nextafter arg obj) then, effectively, the method call obj.(r-nextafter arg) takes place, and its return value is taken as the result of the operation.
obj.(r-remainder arg)
The r-remainder method is invoked when the structure obj is used as the right argument of the remainder function.
If an object obj is combined with an argument arg, as (remainder arg obj) then, effectively, the method call obj.(r-remainder arg) takes place, and its return value is taken as the result of the operation.
obj.(r-scalb arg)
The r-scalb method is invoked when the structure obj is used as the right argument of the scalb function.
If an object obj is combined with an argument arg, as (scalb arg obj) then, effectively, the method call obj.(r-scalb arg) takes place, and its return value is taken as the result of the operation.
obj.(r-scalbln arg)
The r-scalbln method is invoked when the structure obj is used as the right argument of the scalbln function.
If an object obj is combined with an argument arg, as (scalbln arg obj) then, effectively, the method call obj.(r-scalbln arg) takes place, and its return value is taken as the result of the operation.
obj.(r-yn arg)
The r-yn method is invoked when the structure obj is used as the right argument of the yn function.
If an object obj is combined with an argument arg, as (yn arg obj) then, effectively, the method call obj.(r-yn arg) takes place, and its return value is taken as the result of the operation.
Note: the sign-extend function doesn't support a structure as the right argument, bits, which must be an integer.
obj.(ash arg)
The ash method is invoked when the structure obj is used as the left argument of the ash function.
If an object obj is combined with an argument arg, as (ash obj arg) then, effectively, the method call obj.(ash arg) takes place, and its return value is taken as the result of the operation.
Note: the ash function doesn't support a structure as the right argument, bits, which must be an integer.
obj.(bit arg)
The bit method is invoked when the structure obj is used as the left argument of the bit function.
If an object obj is combined with an argument arg, as (bit obj arg) then, effectively, the method call obj.(bit arg) takes place, and its return value is taken as the result of the operation.
Note: the bit function doesn't support a structure as the right argument, bit, which must be an integer.
obj.(width)
The width method is invoked when a structure is used as the argument to the width function.
If an object obj is passed to the function as (width obj) then, effectively, the method call obj.(width) takes place, and its return value is taken as the result of the operation.
obj.(logcount)
The logcount method is invoked when a structure is used as the argument to the logcount function.
If an object obj is passed to the function as (logcount obj) then, effectively, the method call obj.(logcount) takes place, and its return value is taken as the result of the operation.
obj.(bitset)
The bitset method is invoked when a structure is used as the argument to the bitset function.
If an object obj is passed to the function as (bitset obj) then, effectively, the method call obj.(bitset) takes place, and its return value is taken as the result of the operation.
An exception in TXR is a special event in the execution of the program which potentially results in a transfer of control. An exception is identified by a symbol, known as the exception type, and it carries zero or more arguments, called the exception arguments.
When an exception is initiated, it is said to be thrown. This action is initiated by the following functions: throw, throwf and error, and possibly other functions which invoke these. When an exception is thrown, TXR enters into exception processing mode. Exception processing mode terminates in one of several ways:
There are two ways by which exceptions are handled: catches and handlers. Catches and handlers are similar, but different. A catch is an exit point associated with an active scope. When an exception is handled by a catch, the form which threw the exception is abandoned, and unwinding takes place to the catch site, which receives the exception type and arguments. A handler is also associated with an active scope. However, it is a function, and not a dynamic exit point. When an exception is passed to handler, unwinding does not take place; rather, the function is called. The function then either completes the exception handling by performing a nonlocal transfer, or else declines the exception by performing an ordinary return.
Catches and handlers are identified by exception type symbols. A catch or handler is eligible to process an exception if it handles a type which is a supertype of the exception which is being processed. Handles and catches are found by means of a combined search which proceeds from the innermost nesting of dynamic scope to the outermost, without performing any unwinding. When an eligible handler is encountered, its registered function is called, thereby suspending the search. If the handler function returns, the search continues from that scope to yet unvisited outer scopes. When an eligible catch is encountered rather than a handler, the search terminates and a control transfer takes place to the catch site. That control transfer then performs unwinding, which requires it to make a second pass through the same nestings of dynamic scope that had just been traversed in order to find that catch.
Because handlers execute in the dynamic context of the exception origin, without any unwinding having taken place, they expose a potential route of sandbox escape via the package system, unless special steps are taken. The threat is that code at the handler site could take advantage of the current value of the *package* and *package-alist* variables established at the exception throw site to gain inappropriate access to symbols.
For this reason, when a handler is established, the current values of *package* and *package-alist* are recorded into the handler frame. When that handler is later invoked, it executes in a dynamic environment in which those variables are bound to the previously noted values.
The catch mechanism doesn't do any such thing because the unwinding which is performed prior to the invocation of a catch implicitly restores the values of all special variables to the values they had at the time the frame was established.
Exception type symbols are arranged in an inheritance hierarchy, at whose top the symbol t is the supertype of every exception type, and the nil symbol is at the bottom, the subtype of every exception type.
Keyword symbols may be used as exception types.
Every symbol is its own supertype and subtype. Thus whenever X is known to be a subtype of Y, it is possible that X is exactly Y. The defex macro registers exception supertype/subtype relationships among symbols.
The following tree diagram shows the relationships among TXR Lisp's built-in exception symbols. Not shown is the exception symbol nil, subtype of every exception type:
t ----+--- warning
|
+--- restart ---+--- continue
| |
| +--- retry
| |
| +--- skip
|
+--- error ---+--- type-error
|
+--- internal-error
|
+--- panic
|
+--- numeric-error
|
+--- range-error
|
+--- query-error
|
+--- file-error -------+--- path-not-found
| |
| +--- path-exists
| |
| +--- path-permission
|
+--- process-error
|
+--- socket-error
|
+--- system-error
|
+--- alloc-error
|
+--- timeout-error
|
+--- assert
|
+--- syntax-error
|
+--- eval-error
|
+--- match-error
|
+--- case-error
|
+--- opt-error
Program designers are encouraged to derive new error exceptions from the error type. The restart type is intended to be the root of a hierarchy of exception types used for denoting restart points: designers are encouraged to derive restarts from this type. A catch for the continue exception should be established around constructs which can throw an error from which it is possible to recover. That exception provides the entry point into the recovery which resumes execution. A catch for retry should be provided in situations when it is possible and makes sense for a failed operation to be tried again. A catch for skip should be provided in situations when it is possible and sensible to continue with subsequent operations even though an operation has failed.
Exception handling in TXR Lisp provides capabilities similar to the condition system in ANSI Common Lisp. The implementation and terminology differ.
Most obviously, ANSI CL uses the "condition" term, whereas TXR Lisp uses "exception".
In ANSI CL, a condition is "raised", whereas a TXR Lisp exception is "thrown".
In ANSI CL, when a condition is raised, a condition object is created. Condition object are similar to class objects, but are not required to be in the Common Lisp Object System. They are related by inheritance and can have properties. TXR Lisp exceptions are unencapsulated: they consist of a symbol, plus zero or more arguments. The symbols are related by inheritance.
When a condition is raised in ANSI CL, the dynamic scope is searched for a handler, which is an ordinary function which receives the condition. No unwinding or nonlocal transfer takes place. The handler can return, in which case the search continues. Matching the condition to the handler is by inheritance. Handler functions are bound to exception type names. If a handler chooses to actually handle a condition (thereby terminating the search) it must itself perform some kind of dynamic control transfer, rather than return normally. ANSI CL provides a dynamic control mechanism known as restarts which is usually used for this purpose. A condition handler may invoke a particular restart handler. Restart handlers are similar to exception handlers: they are functions associated with symbols in the dynamic environment.
In TXR Lisp, the special behavior which occurs for exceptions derived from error and those from warning is built into the exception handling system, and tied to those types. When an error or warning exception is unhandled, the exception handling system itself reacts, so the special behaviors occur no matter how these exceptions are raised. In ANSI CL, the special behavior for unhandled error conditions (of invoking the debugger) is implemented only in the error function; error conditions signalled other than via that function are not subject to any special behavior. There is a parallel situation with regard to warnings: the ANSI CL warn function implements a special behavior for unhandled warnings (of emitting a diagnostic) but warnings not signalled via that function are not treated that way. Thus in TXR Lisp, there is no way to raise an error or warning that is simply ignored due to being unhandled.
In TXR Lisp exceptions are a unification of conditions and restarts. From an ANSI CL perspective, TXR Lisp exceptions are a lot like CL restarts, except that the symbols are arranged in an inheritance hierarchy. TXR Lisp exceptions are used both as the equivalent of ANSI CL conditions and as restarts.
In TXR Lisp the terminology "catch" and "handle" is used in a specific way. To handle an exception means to receive it without unwinding, with the possibility of declining to handle it, so that the search continues for another handler. To catch an exception means to match an exception to a catch handler, terminate the search, unwind and pass control to the handler.
TXR Lisp provides an operator called handler-bind for specifying handlers. It has a different syntax from ANSI CL's handler-bind. TXR Lisp provides a macro called handle which simplifies the use of handler-bind. This macro superficially resembles ANSI CL's handler-case, but is semantically different. The most notable difference is that the bodies of handlers established by handler-bind execute without any unwinding taking place and may return normally, thereby declining to take the exception. In other words, handle has the same semantics as handler-bind, providing only convenient syntax.
TXR Lisp provides a macro called catch which has the same syntax as handle but specifies a catch point for exceptions. If, during an exception search, a catch clause matches an exception, a dynamic control transfer takes place from the throw site to the catch site. Then the clause body is executed. The catch macro resembles ANSI CL's restart-case or possibly handler-case, depending on point of view.
TXR Lisp provides unified introspection over handler and catch frames. A program can programmatically discover what handler and catches are available in a given dynamic scope. ANSI CL provides introspection over restarts only; the standard doesn't specify any mechanism for inquiring what condition handlers are bound at a given point in the execution.
The following two examples express a similar approach implemented using ANSI Common Lisp conditions and restarts, and then using TXR Lisp exceptions.
;; Common Lisp
(define-condition foo-error (error)
((arg :initarg :arg :reader foo-error-arg)))
(defun raise-foo-error (arg)
(restart-case
(let ((c (make-condition 'foo-error :arg arg)))
(error c))
(recover (recover-arg)
(format t "recover, arg: ~s~%" recover-arg))))
(handler-bind ((foo-error
(lambda (cond)
(format t "handling foo-error, arg: ~s~%"
(foo-error-arg cond))
(invoke-restart 'recover 100))))
(raise-foo-error 200))
The output of the above is:
handling foo-error, arg: 200
recover, arg: 100
The following is possible TXR Lisp equivalent for the above Common Lisp example. It produces identical output.
(defex foo-error error)
(defex recover restart) ;; recommended practice
(defun raise-foo-error (arg)
(catch
(throw 'foo-error arg)
(recover (recover-arg)
(format t "recover, arg: ~s\n" recover-arg))))
(handle
(raise-foo-error 200)
(foo-error (arg)
(format t "handling foo-error, arg: ~s\n" arg)
(throw 'recover 100)))
To summarize the differences: exceptions serve as both conditions and restarts in TXR. The same throw function is used to initiate exception handling for foo-error and then to transfer control out of the handler to the recovery code. The handler accepts one exception by raising another.
When an exception symbol is used for restarting, it is a recommended practice to insert it into the inheritance hierarchy rooted at the restart symbol, either by inheriting directly from restart or from an exception subtype of that symbol.
(catch
(open-file "AsDf")
(error (msg)
;; the value 2 is retrieved from msg
;; 2 is the common value of ENOENT
(list (string-get-code msg) msg)))
-> (2 "error opening \"AsDf\": 2/\"No such file or directory\"")
(throw symbol arg*)
(throwf symbol format-string format-arg*)
(error format-string format-arg*)
These functions generate an exception. The throw and throwf functions generate an exception identified by symbol, whereas error throws an exception of type error. The call (error ...) can be regarded as a shorthand for (throwf 'error ...).
The throw function takes zero or more additional arguments. These arguments become the arguments of a catch handler which takes the exception. The handler will have to be capable of accepting that number of arguments.
The throwf and error functions generate an exception which has a single argument: a character string created by a formatted print to a string stream using the format string and additional arguments.
Because error throws an error exception, it does not return. If an error exception is not handled, TXR will issue diagnostic messages and terminate. Likewise, throw or throwf are used to generate an error exception, they do not return.
If the throw and throwf functions are used to generate an exception not derived from error, and no handler is found which accepts the exception, they return normally, with a value of nil.
(catch try-expression
{(symbol (arg*) body-form*)}*)
(catch* try-expression
{(symbol (type-arg arg*) body-form*)}*)
(catch** try-expression
{(symbol desc (type-arg arg*) body-form*)}*)
The catch macro establishes an exception catching block around the try-expression. The try-expression is followed by zero or more catch clauses. Each catch clause consists of a symbol which denotes an exception type, an argument list, and zero or more body forms.
If try-expression terminates normally, then the catch clauses are ignored. The catch itself terminates, and its return value is that of the try-expression.
If try-expression throws an exception which is a subtype of one or more of the type symbols given in the exception clauses, then the first (leftmost) such clause becomes the exit point where the exception is handled. The exception is converted into arguments for the clause, and the clause body is executed. When the clause body terminates, the catch terminates, and the return value of the catch is that of the clause body.
If try-expression throws an exception which is not a subtype of any of the symbols given in the clauses, then the search for an exit point for the exception continues through the enclosing forms. The catch clauses are not involved in the handling of that exception.
When a clause catches an exception, the number of arguments in the catch must match the number of elements in the exception. A catch argument list resembles a function or lambda argument list, and may be dotted. For instance the clause (foo (a . b)) catches an exception subtyped from foo, with one or more elements. The first element binds to parameter a, and the rest, if any, bind to parameter b. If there is only one element, b takes on the value nil.
The catch* macro is a variant of catch with the following difference: when catch* invokes a clause, it passes the exception symbol as the leftmost argument type-arg. Then the exception arguments follow. In contrast, only the exception arguments are passed to the clauses of catch.
The catch** macro is a further variant, which differs from catch* by requiring each catch clause to provide a description desc, an expression which evaluates to a character string. The desc expressions are evaluated in left-to-right order prior to the evaluation of try-expression.
Also see: the unwind-protect operator, and the functions throw, throwf and error, as well as the handler-bind operator and handle macro.
(unwind-protect protected-form cleanup-form*)
The unwind-protect operator evaluates protected-form in such a way that no matter how the execution of protected-form terminates, the cleanup-forms will be executed.
The cleanup-forms, however, are not protected. If a cleanup-form terminates via some nonlocal jump, the subsequent cleanup-forms are not evaluated.
cleanup-forms themselves can "hijack" a nonlocal control transfer such as an exception. If a cleanup-form is evaluated during the processing of a dynamic control transfer such as an exception, and that cleanup-form initiates its own dynamic control transfer, the original control transfer is aborted and replaced with the new one.
The exit points for dynamic control transfers are removed as unwinding takes place. That is to say, at the start of a dynamic control transfer, a search takes place for the target exit point. That search might skip other exit points which aren't targets of the control transfer. Those skipped exit points are left undisturbed and are still visible during unwinding until their individual binding forms are abandoned. Thus at the time of execution of an unwind-protect cleanup-form, all of the exit points of dynamically surrounding forms are still visible, even ones which are nearer than the targeted exit point.
(block foo
(unwind-protect
(progn (return-from foo 42)
(format t "not reached!\n"))
(format t "cleanup!\n")))
In this example, the protected progn form terminates by returning from block foo. Therefore the form does not complete and so the output "not reached!" is not produced. However, the cleanup form executes, producing the output "cleanup!".
(ignerr form*)
The ignerr macro operator evaluates each form similarly to the progn operator. If no forms are present, it returns nil. Otherwise it evaluates each form in turn, yielding the value of the last one.
If the evaluation of any form is abandoned due to an exception of type error, the code generated by the ignerr macro catches this exception. In this situation, the execution of the ignerr form terminates without evaluating the remaining forms, and yields nil.
(ignwarn form*)
The ignwarn macro resembles ignerr. It arranges for the evaluation of each form in left-to-right order. If all the forms are evaluated, then the value of the last one is returned. If no forms are present, then nil is returned.
If any form throws an exception of type warning then this exception is intercepted by a handler established by ignwarn. This handler reacts by throwing an exception of type continue.
The effect is that the warning is ignored, since the handler doesn't issue any diagnostic, and passes control to the warning's continue point.
Note: all sites within TXR which throw a warning also provide a nearby catch for a continue exception, for resuming evaluation at the point where the warning was issued.
(handler-bind function-form symbol-list body-form*)
The handler-bind operator establishes a handler for one or more exception types, and evaluates zero or more body-forms in a dynamic scope in which that handler is visible.
When the handler-bind form terminates normally, the handler is removed. The value of the last body-form is returned, or else nil if there are no forms.
The function-form argument is an expression which must evaluate to a function. The function must be capable of accepting the exception arguments. All exceptions functions require at least one argument, since the leftmost argument in an exception handler call is the exception type symbol.
The symbol-list argument is a list of symbols, not evaluated. If it is empty, then the handler isn't eligible for any exceptions. Otherwise it is eligible for any exception whose exception type is a subtype of any of the symbols.
If the evaluation of any body-form throws an exception which is not handled within that form, and the handler is eligible for that exception, then the function is invoked. It receives the exception's type symbol as the leftmost argument. If the exception has arguments, they appear as additional arguments in the function call. If the function returns normally, then the exception search continues. The handler remains established until the exception is handled in such a way that a dynamic control transfer abandons the handler-bind form.
Note: while a handler's function is executing, the handler is disabled. If the function throws an exception for which the handler is eligible, the handler will not receive that exception; it will be skipped by the exception search as if it didn't exist. When the handler function terminates, either via a normal return or a nonlocal control transfer, then the handler is reenabled.
(handle try-expression
{(symbol (arg*) body-form*)}*)
(handle* try-expression
{(symbol (type-arg arg*) body-form*)}*)
The handle macro is a syntactic sugar for the handler-bind operator. Its syntax is exactly like that of catch. The difference between handle and catch is that the clauses in handle are invoked without unwinding. That is to say, handle does not establish an exit point for an exception. When control passes to a clause, it is by means of an ordinary function call and not a dynamic control transfer. No evaluation frames are yet unwound when this takes place.
The handle macro establishes a handler, by handler-bind whose symbol-list consists of every symbol gathered from every clause.
The handler function established in the generated handler-bind is synthesized from all of the clauses, together with dispatch logic which which passes the exception and its arguments to the first eligible clause.
The try-expression is evaluated in the context of this handler.
The clause of the handle syntax can return normally, like a function, in which case the handler is understood to have declined the exception, and exception processing continues. To handle an exception, the clause of the handle macro must perform a dynamic control transfer, such returning from a block via return or throwing an exception.
The handle* macro is a variant of handle with the following difference: when handle* invokes a clause, it passes the exception symbol as the leftmost argument type-arg. Then the exception arguments follow. In contrast, only the exception arguments are passed to the clauses of handle.
(with-resources ({(sym [init-form [cleanup-form*]])}*)
body-form*)
The with-resources macro provides a sequential binding construct similar to let*. Every sym is established as a variable which is visible to the init-forms of subsequent variables, to all subsequent cleanup-forms including that of the same variable, and to the body-forms.
If no init-form is supplied, then sym is bound to the value nil.
If an init-form is supplied, but no cleanup-forms, then sym is bound to the value of the init-form.
If one or more cleanup-forms are supplied in addition to init-form, they specify forms to be executed upon the termination of the with-resources construct.
When an instance of with-resources terminates, either normally or by a nonlocal control transfer, then for each sym whose init-form had executed, thus causing that sym to be bound to a value, the cleanup-forms corresponding to sym are evaluated in the usual left-to-right order.
The syms are cleaned up in reverse (right-to-left) order. The cleanup-forms of the most recently bound sym are processed first; those of the least recently bound sym are processed last.
When the with-resources form terminates normally, the value of the last body-form is returned, or else nil if no body-forms are present.
From its inception, until TXR 265, with-resources featured an undocumented behavior. Details are given in the COMPATIBILITY section's Compatibility Version Values subsection, in the notes for compatibility value 265.
The following expression opens a text file and reads a line from it, returning that line, while ensuring that the stream is closed immediately:
(with-resources ((f (open-file "/etc/motd") (close-stream f)))
(whilet ((l (get-line f)))
(put-line l)))
Note that a better way to initialize exactly one stream resource is with the with-stream macro, which implicitly closes the stream when it terminates.
The *unhandled-hook* variable is initialized with nil by default.
It may instead be assigned a function which is capable of taking three arguments.
When an exception occurs which has no handler, this function is called, with the following arguments: the exception type symbol, the exception object, and a third value which is either nil or else the form which was being evaluated when the exception was thrown. The call occurs before any unwinding takes place.
If the variable is nil, or isn't a function, or the function returns after being called, then unwinding takes place, after which some informational messages are printed about the exception, and the process exits with a failed termination status.
In the case when the variable contains a object other than nil which isn't a function, a diagnostic message is printed on the *stderr* stream prior to unwinding.
Prior to the function being called, the *unhandled-hook* variable is reset to nil.
Note: the functions source-loc or source-loc-str may be applied to the third argument of the *unhandled-hook* function to obtain more information about the form.
(defex {symbol}*)
The macro defex records hierarchical relationships among symbols, for the purposes of the use of those symbols as exceptions. It is closely related to the @(defex) directive in the TXR pattern language, performing the same function.
All symbols are considered to be exception subtypes, and every symbol is implicitly its own exception subtype. This macro does not introduce symbols as exception types; it only introduces subtype-supertype relationships.
If defex is invoked with no arguments, it has no effect.
If arguments are present, they must be symbols.
If defex is invoked with only one symbol as its argument, it has no effect.
At least two symbols must be specified for a useful effect to take place. If exactly two symbols are specified, then, subject to error checks, defex makes the left symbol an exception subtype of the right symbol.
This behavior generalizes to three or more arguments: if three or more symbols are specified, then each symbol other than the last is registered as a subtype of the symbol which follows.
If a defex has three or more arguments, they are processed from left to right. If errors are encountered during the processing, the correct registrations already made for prior arguments remain in place.
Every symbol is implicitly considered to be its own exception subtype, therefore it is erroneous to explicitly register a symbol as its own subtype.
The symbol nil is implicitly a subtype of every exception type. Therefore, it is erroneous to attempt to specify it as a supertype in a registration. Using nil as a subtype in a registration is silently permitted, but has no effect. No explicit registration is recorded between nil and its successor in the argument list.
The symbol t is implicitly the supertype of every exception type. Therefore, it is erroneous to attempt to register it as an exception subtype. Using t as a supertype in a registration is also erroneous.
A symbol a may not be registered as a subtype of a symbol b if the reverse relationship already exists between those two symbols.
The foregoing rules allow redefinitions to take place, while forbidding cycles from being created in the exception subtype inheritance graph.
Keyword symbols may be used as exception types.
(register-exception-subtypes {symbol}*)
The register-exception-subtypes function constitutes the underlying implementation for the defex macro.
The following equivalence applies:
(defex a b ...) <--> (register-exception-subtypes 'a 'b ...)
That is, the defex macro works as if by generating a call to the function, with the arguments quoted.
The semantics of the function is precisely that of the macro.
(exception-subtype-p left-symbol right-symbol)
The exception-subtype-p function tests whether two symbols are in a relationship as exception types, such that left-symbol is a direct or indirect exception subtype of right-symbol.
If that is the case, then t is returned, otherwise nil.
(exception-subtype-map)
The exception-subtype-map function returns a tree structure which captures information about all registered exception types.
The map appears as an association list which contains an entry for every exception symbol, paired with that type's supertype path. The first element in the supertype path is the exception's immediate supertype. The next element is that type's supertype and so on. The last element in every path is the grand supertype t.
For instance, if only the types a, b and c existed in the system, and were linked according to this inheritance graph:
t ----+--- b --- a
|
+--- c
such that the supertype of b and c is t, and a has b as supertype, then the function might return:
((a b t) (b t) (c t) (t))
or any other equivalent permutation.
The returned list may share substructure, so that the (t) sublist is shared among all four entries, and (b t) between the first two.
If the program alters the tree structure returned by exception-map-p, the consequences are unspecified; this structure may be the actual object which represents the type hierarchy.
(defstruct frame nil)
(defstruct catch-frame frame types desc jump)
(defstruct handle-frame frame types fun)
The structure types frame, catch-frame and handle-frame are used by the get-frames and find-frame functions to represent information about the currently established exception catches (see the catch macro) and handlers (see handler-bind and handle).
The frame type serves as the common base for catch-frame and handle-frame.
Modifying any of the slots of these structures has no effect on the actual frame from which they are derived; the frame structures are only representation which provides information about frames. They are not the actual frames themselves.
Both catch-frame and handle-frame have a types slot. This holds the list of exception type symbols which are matched by the catch or handler.
The desc slot of a catch-frame holds a list of the descriptions produced by the catch** macro. If there are no descriptions, then this member is nil, otherwise it is a list whose elements are in correspondence with the list in the types slot.
The jump slot of a catch-frame is an opaque cptr ("C pointer") object which is related to the stack address of the catch frame. If it is altered, the catch frame object becomes invalid for the purposes of invoke-catch.
The fun slot of a handle-frame is the registered handler function. Note that all the clauses of a handle macro are compiled to a single function, which is established via handler-bind, so an instance of the handle macro corresponds to a single handle-frame.
(get-frames)
The get-frames function inquires the current dynamic environment in order to retrieve information about established exception catch and handler frames. The function returns a list, ordered from the innermost nesting level to the outermost nesting, of structure objects derived from the frame structure type. The list contains two kinds of objects: structures of type catch-frame and of type handle-frame.
These objects are not the frames themselves, but only provide information about frames. Modifying the slots in these structures has no effect on the original frames. Also, these structures have their own lifetime and can endure after the original frames have disappeared. This has implications for the use of the invoke-catch function.
The handle-frame structures have a fun slot, which holds a function. It may be invoked directly.
A catch-frame structure may be passed as an argument to the invoke-catch function.
(find-frame [exception-symbol [frame-type]])
(find-frames [exception-symbol [frame-type]])
The find-frame function locates the first (innermost) instance of a specific kind of exception frame (a catch frame or a handler frame) which is eligible for processing an exception of a specific type. If such a frame is found, it is returned. The returned frame object is of the same kind as the objects which comprise the list returned by the function get-frames. If such a frame is not found, nil is returned.
The exception-symbol argument specifies a match by exception type: the candidate frame must specify in its list of matches at least one type which is an exception supertype of exception-symbol. If this argument is omitted, it defaults to nil which finds any handler that matches at least one type. There is no way to search for handlers which match an empty set of types; the find-frame function skips such frames.
The frame-type argument specifies which frame type to find. Useful values for this argument are the structure type names catch-frame and handle-frame or the actual structure type objects which these type names denote. If any other value is specified, the function returns nil. If the argument is omitted, it defaults to the type of the catch-frame structure. That is to say, by default, the function looks for catch frames.
Thus, if find-frame is called with no arguments at all it finds the innermost catch frame, if any exists, or else returns nil.
The find-frames function is similar to find-frame except that it returns all matching frames, ordered from the innermost nesting level to the outermost nesting. If called with no arguments, it returns a list of the catch frames.
(invoke-catch catch-frame symbol argument*)
The invoke-catch function abandons the current evaluation context to perform a nonlocal control transfer directly to the catch described by the catch-frame argument, which must be a structure of type catch-frame obtained using any of the functions get-frames, find-frames or find-frame.
The control transfer is possible only if the catch frame represented by catch-frame structure is still established, and if the structure hasn't been tampered with.
If a given catch-frame structure is usable with invoke-catch, then a copy of that structure made with copy-struct is also usable, denoting the same catch frame.
The symbol argument should be an exception symbol. It is passed to the exception frame, as if it had appeared as the first argument of the throw function. Similarly, the arguments are passed to the catch frame as if they were the trailing arguments of a throw. The difference between invoke-catch and throw is that invoke-catch targets a specific catch frame as its exit point, rather than searching for a matching catch or handler frame. That specific frame receives the control. The frame receives control even if it it is not otherwise eligible for catching the exception type denoted by symbol.
(assert expr [format-string format-arg*])
The assert macro evaluates expr. If expr yields any true value, then assert terminates normally, and that value is returned.
If instead expr yields nil, then assert throws an exception of type assert. The exception carries an informative character string that contains a diagnostic detailing the expression which yielded nil, and the source location of that expression, if available.
If the format-string and possibly additional format arguments are given to assert then those arguments are used to format additional text which is appended to the diagnostic message after a separating character such as a colon.
This section describes a number of features related to the diagnosis of errors during the static processing of program code prior to evaluation. The material is of interest to developers of macros intended for broad reuse.
TXR Lisp uses exceptions of type eval-error to identify erroneous situations during both transformation of code and its evaluation. These exceptions have one argument, which is a character string. If not handled by program code, eval-error exceptions are specially recognized and treated by the built-in handling logic. The message is incorporated into diagnostic output which includes more information which is deduced.
TXR Lisp uses exceptions of type warning to identify certain situations of interest. Ordinary non-deferrable warnings have a structure identical to errors, except for the exception symbol. TXR's provides built-in "auto continue" handling for warnings. If a warning exception is not intercepted by a catch or an accepting handler, then a diagnostic is issued on the *stderr* stream, after which a continue exception is thrown with no arguments. If that continue exception is not handled, then control returns normally to the point that exception to resume the computation which generated the warning.
Callers which invoke code that may generate warning exceptions are therefore not required to handle them. However, callers which do handle warning exceptions expect to be able to throw a continue exception in order to resume the computation that triggered the warning, without allowing other handlers to see the exception.
The generation of a warning should thus conform to the following pattern:
(catch
(throw 'warning "message")
(continue ()))
TXR supports a form of diagnostic known as a deferrable warning. A deferrable warning is distinguished in two ways. Firstly, it is either of the type defr-warning or subtyped from that type. The defr-warning type itself is a direct subtype of warning.
Secondly, a deferrable warning carries an additional tag argument after the exception message. A deferrable exception is thrown according to this pattern:
(catch
(throw 'defr-warning "message" . tag)
(continue ()))
TXR's built-in exception handling logic reacts specially to the presence of the tag material in the exception. First, the global tentative definition list is searched for the presence of the tag, using equal equality. If the tag is found, then the warning is discarded. If the tag is not found, then the exception argument list is added to the global deferred warning list. In either case, the continue exception is thrown to resume the computation which threw the warning, as in the case of an ordinary non-deferrable warning.
The purpose of this mechanism is to suppress warnings which become superfluous when more of the program code is examined. For instance, a warning about a call to an undefined function is superfluous if a definition of that function is supplied later, yet before that function call is executed.
Deferred warnings accumulate in the deferred warning list from which they can be removed. The list is purged at various times such as when a top-level load completes, and the deferred warnings are released, as if by a call to the release-deferred-warnings function.
(compile-error context-obj fmt-string fmt-arg*)
(compile-warning context-obj fmt-string fmt-arg*)
The functions compile-error and compile-warning provide a convenient and uniform way for code transforming functions such as macro-expanders to generate diagnostics. The compile-error function throws an exception of type eval-error. The compile-warning function throws an exception of type warning and internally provides a catch for the continue exception which allow a warning handler to resume execution after the warning. If a handler throws a continue exception which is caught by compile-warning, then compile-warning returns nil.
Because compile-warning throws a non-error exception, it returns nil in the event that no catch is found for the exception, and no handler which accepts it.
The argument conventions are the same for both functions. The context-obj is typically a compound form to which the diagnostic applies.
The functions produce a diagnostic message which incorporates the location information and symbol obtained from context-obj and the format-style arguments fmt-string and its fmt-args.
(compile-defr-warning context-obj tag
fmt-string fmt-arg*)
The compile-defr-warning function throws an exception of type defr-warning and internally provides a catch for the continue exception needed to resume after the warning.
The function produces a diagnostic message which incorporates the location information and symbol obtained from context-obj and the format-style arguments fmt-string and its fmt-args. This diagnostic message constitutes the first argument of the exception. The tag argument is taken as the second argument.
If the exception isn't intercepted by a catch or by an accepting handler, compile-defr-warning returns nil. In also returns nil if it catches a continue exception.
(purge-deferred-warning tag)
The purge-deferred-warning removes all warnings marked with tag from the deferred list. It also removes all tags matching tag from the tentative definition list. Tags are compared using the equal function.
(register-tentative-def tag)
The register-tentative-def function adds tag to the list of tentative definitions which are used to suppress deferrable warnings.
The idea is that a definition of some construct has been seen, but not yet executed. Thus the construct is not defined, but it can reasonably be expected that it will be defined; hence, warnings about its nonexistence can be suppressed.
For example, in the following code, when the expression (foo) is being expanded and transformed, the foo function does not exist:
The function won't be defined until the progn is evaluated. Thus a warning is generated that (foo) refers to an undefined function. However, this warning is discarded, because the expander for defun registers a tentative definition tag for foo.
When the definition of foo takes place, the defun operator will call purge-deferred-warning which will remove not only all accumulated warnings related to the undefinedness of foo but also remove the tentative definition.
Note: this mechanism isn't perfect because it will still suppresses the warning in situations like
(progn (if nil (defun foo ())) (foo))
(tentative-def-exists tag)
The tentative-def-exists function checks whether tag has been registered via register-tentative-def and not yet purged by purge-deferred-warning.
(defer-warning args)
The defer-warning function attempts to register a deferred warning. The args argument corresponds to the arguments which are passed to the throw function in order to generate a warning exception, not including the exception symbol.
Args is expected to have at least two elements, the second of which is a deferred warning tag.
The defer-warning function returns nil.
Note: this function is intended for use in exception handlers. The following example shows a handler which intercepts warnings. It defers deferrable warnings, and prints ordinary warnings:
(handle
(some-form ..) ;; some code which might generate warnings
(defr-warning (msg tag) ;; catch deferrable and defer
(defer-warning (cons msg tag))
(throw 'continue)) ;; warning processed: resume execution
(warning (msg)
(put-line `warning: @msg`) ;; print non-deferrable
(throw 'continue))) ;; warning processed: resume execution
(release-deferred-warnings)
The release-deferred-warnings removes all warnings from the deferred list. Then, it issues each deferred warning as an ordinary warning.
Note: there is normally no need for user programs to use this function since deferred warnings are issued automatically.
(dump-deferred-warnings stream)
The dump-deferred-warnings empties the list of deferred warnings, and converts each one into a diagnostic message sent to sent to stream. After the diagnostics are printed, the list of pending warnings is cleared.
Note: there is normally no need for user programs to use this function since deferred warnings are issued automatically.
TXR Lisp supports delimited continuations, which are integrated with the block feature. Any named or anonymous block, including the implicit blocks created around function bodies, can be used as the delimiting prompt for the capture of a continuation.
A delimited continuation is section of a possible future of the computation, up to a delimiting prompt, reified as a first class function.
(defun receive (cont)
(format t "cont returned ~a\n" (call cont 3)))
(defun function ()
(sys:capture-cont 'abcd (fun receive)))
(block abcd
(format t "function returned ~a\n" (function))
4)
Output:
function returned 3
cont returned 4
function returned t
Evaluation begins with the block form. This form calls function which uses sys:capture-cont to capture a continuation up to the abcd prompt. The continuation is passed to the receive function as an argument.
This captured object represents the continuation of computation up to that prompt. It appears as a one-argument function which, when called, resumes the captured computation. Its argument emerges out of the sys:capture-cont call as a return value. When the computation eventually returns all the way to the delimiting prompt, the return value of that prompt will then appear as the return value of the continuation function.
In this example, the function receive immediately invokes the continuation function which it receives, passing it the argument value 3. And so, evaluation now continues in the resumed future represented by the continuation. Inside the continuation, sys:capture-cont appears to return, yielding the value 3. This bubbles up through function up to the block abcd where a message is printed: "function returned 3".
The block terminates, yielding the value 4. Thereby, the continuation ends, since it is delimited up to that block. Control now returns to the receive function which invoked the continuation, where the function call form (call cont) terminates, yielding the value 4 that was returned by the continuation's delimiting block form. The message "cont returned 4" is printed. The receive function returns normally, returning the value t which emerged from the format call. Control is now back in function where the sys:capture-cont form terminates and returns the t. This bubbles up to block which prints "function returned t".
In summary, a continuation represents, as a function, the subsequent computation that is to take place starting at some point, up to some recently established, dynamically enclosing delimiting prompt. When the continuation is captured, that future doesn't have to take place; an alternative future can carry out in which that continuation is available as a function. That alternative future can invoke the continuation at will. Invocations (resumptions) of the continuation appear as additional returns from the capture operator. A resumption of a continuation terminates when the delimiting prompt terminates, and the continuation yields the value which emerges from the prompt.
Delimited continuations are implemented by capturing a segment of the evaluation stack between the prompt and the capture point. When a continuation is resumed, this saved copy of a stack segment is inserted on top of the current stack and the procedure context is resumed such that evaluation appears to emerge from the capture operator. As the continuation runs to completion, it simply pops these inserted stack frames naturally. Eventually it pops out of the delimiting prompt, at which point control ends up at the point which invoked the continuation function.
The low-level operator for capturing a continuation is sys:capture-cont. More expressive and convenient programming with continuations is provided by the macros obtain, obtain-block, yield-from and yield, which create an abstraction which models the continuation as a suspended procedure supporting two-way communication of data. A suspend operator is provided, which is more general. It is identical to the shift operator described in various computer science literature about delimited continuations, except that it refers to a specific delimiting prompt by name.
Continuations raise the issue of what to do about unwinding. The language Scheme provides the much criticized dynamic-wind operator which can execute initialization and clean-up code as a continuation is entered and abandoned. TXR takes a simpler, albeit risky approach. It provides a non-unwinding escape operator sys:abscond-from for use with continuations. Code which has captured a continuation can use this operator to escape from the delimiting block without triggering any unwinding among the frames between the capture point and the delimiter. When the continuation is restarted, it will then do so with all of the resources associated with it frames intact. When the continuation executes normal returns within its context, the unwinding takes place then. Thus tidy, "thread-like" use of continuations is possible with a small measure of coding discipline. Unfortunately, the absconding operator is dangerous: its use breaks the language guarantee that clean-up associated with a form is done no matter how a form terminates.
Delimited continuations resemble lexical closures in some ways. Both constructs provide a way to return to some context whose evaluation has already been abandoned, and to access some aspects of that context. However, lexical closures are statically scoped. Closures capture the lexically apparent scope at a given point, and produce a function whose body has access to that scope, as well as to some arbitrary arguments. Thus, a lexical scope is reified as a first-class function. By contrast, a delimited continuation is dynamic. It captures an an entire segment of a program activation chain, up to the delimiting prompt. This segment includes scopes which are not lexically visible at the capture point: the scopes of parent functions. Moreover, the segment includes not only scopes, but also other aspects of the evaluation context, such as the possibility of returning to callers, and the (captured portion of) the original dynamic environment, such asexception handlers. That is to say, a lexical closure's body cannot return to the surrounding code or see any of its original dynamic environment; it can only inspect the environment, and then return to its own caller. Whereas a restarted delimited continuation can continue evaluation of the surrounding code, return to surrounding forms and parent functions, and access the dynamic environment. The continuation function returns to its caller when that entire restarted context terminates, whereas a closure returns to its caller as soon as the closure body terminates.
Delimited continuations in TXR expose a behavioral difference between compiled and interpreted code which mutates the values of lexical variables.
When a continuation is captured in compiled code, it captures not only the bindings of lexical variables, but also potentially their current values at the time of capture. What this means is that whenever the continuation is resumed, those variables will appear to have the captured values, regardless of any mutations that have taken place since. In other words, the captured future includes those specific values. This is because in compiled code, variables are allocated on the stack, which is copied as part of creating a continuation. Those variables are effectively newly instantiated in each resumption of the continuation, when the captured stack segment is reinstated into the stack, and take on those original values.
In contrast, interpretation of code only maintains an environment pointer on the stack; the lexical environment is a dynamically allocated object whose contents aren't included in the continuation's stack segment capture. If the captured variables are modified after the capture, the continuation will see the updated values: all resumptions of the continuation share the same instance of the captured environment among themselves, and with the original context where the capture took place.
An additional complication is that when compiled code captures lexical closures, captured variables are moved into dynamic storage and then they become shared: the semantics of the mutation of those variables is then similar to the situation in interpreted code. Therefore, the above described non-sharing capture behavior of compiled code is not required to hold.
In continuation-based code which relies on mutation of lexical variables created with let or let*, the macros hlet and hlet* can be used instead. These macros create variable bindings whose storage is always outside of the stack, and therefore the variables will exhibit consistent interpreted and compiled semantics under continuations. All contexts which capture the same lexical binding of a given hlet/hlet* variable share a single instance. The most recent assignment to the variable taking place in any context establishes its value, as seen by any other context. The resumption of a continuation will not restore such a variable to a previous value.
If the affected variables are other kinds of bindings such as function parameters or variables created with specialized binding constructs such as with-stream, additional coding changes may be required to get interpreted code working under compilation.
(sys:capture-cont name receive-fun [context-form])
The sys:capture-cont function captures a continuation, and also serves as the resume point for the resulting continuation. Which of these two situations is the case (capture or resumption) is distinguished by the use of the receive-fun argument, which must be a function capable of being called with one argument.
A block named name must be visible; the continuation is delimited by the closest enclosing block of this name.
The optional context-form argument should be a compound form. If sys:capture-cont reports an error, it reports it against this form, and uses the form's operator symbol as the name of the function which encountered the error. If the argument is omitted, sys:capture-cont uses its own name.
The sys:capture-cont function captures a continuation, represented as a function. It immediately calls receive-fun, passing it it the continuation function as an argument. If receive-fun returns normally, then sys:capture-cont returns whatever value receive-fun returns.
Resuming a continuation is done by invoking the continuation function. When this happens, the entire continuation context is restored by recreating its captured evaluation frames on top of the current stack. Inside the continuation, the sys:capture-cont function call which captured the continuation now appears to return, and yields a value. That value is precisely the value which was just passed to the continuation function moments ago.
The resumed continuation can terminate in one of three ways. Firstly, it can simply keep executing until it discards all of its evaluation frames below the delimiting block, and then allows that block to terminate naturally by evaluating the last form contained in the block. Secondly, can use return-from against its delimiting block to explicitly abandon all evaluations in between and terminate that block. Or it may perform a nonlocal control transfer past the delimited block somewhere into the evaluation frames of the caller. In the first two cases, the termination of the block turns into an ordinary return from the continuation function, and the result value of the terminated block becomes the return value of that function call. In the last case, the call of the continuation function is abandoned and unwinding continues through the caller.
If the symbol sys:cont-poison is passed to the continuation function, the continuation will be resumed in a different manner: its context will be restored as in the ordinary resume case, whereupon it will be immediately abandoned by a nonlocal exit, causing unwinding to take place across all of the continuation's evaluation frames. The function then returns nil.
If the symbol sys:cont-free is passed to the continuation function, the continuation isn't be resumed at all; rather, the buffer which holds the saved context of the continuation is released. Thereafter, an attempt to resume the continuation results in an error exception being thrown. After releasing the buffer, the function returns nil.
The continuation function may be used any time after it is produced, and may be called more than once, regardless of whether the originally captured dynamic context is still executing. The continuation object may be communicated into the resumed continuation, which can then use it to call itself, resulting in multiple nested resumptions of the same continuation. A delimited continuation is effectively a first class function.
The underlying continuation object produced by sys:capture-cont stores a copy of the captured dynamic context. Whenever the continuation function is invoked, a copy of the captured is reinstated as if it were a new context. Thus each apparent return from the sys:capture-cont inside a resumed continuation is not actually made in the original context, but in a copy of that context. That context can be resumed multiple times sequentially or recursively.
Just like lexical closures, continuations do not copy lexical environments; they capture lexical environments by reference. If a continuation modifies the values of captured lexical variables, those modifications are visible to other resumptions of the same continuation, to other continuations which capture the same environment, to lexical closures which capture the same environment and to the original context which created that environment, if it is still active.
Unlike lexical closures, continuations do capture the local bindings of special variables. That is to say, if *var* is a special variable, then a lexical closure created inside a (let ((*var* 42)) ...) form will not capture the local rebinding of *var* which holds 42. When the closure is invoked and accesses *var*, it accesses whatever value of *var* is dynamically current, as dictated by the environment which calls the closure, rather than the capturing environment.
With continuations, the behavior is different. If a continuation is captured inside a (let ((*var* 42)) ...) form then it does capture the local binding. This is regardless whether the delimited prompt of the capture is enclosed in this form, or outside of the form. The special variable has a binding in a dynamic environment. There is always a reference to a current dynamic environment associated with every evaluation context, and a continuation captures that reference. Because it is a reference, it means that the binding is shared. That is to say, all invocations of all continuations which capture the same dynamic environment in which that (let ((*var* 42)) ...) binding was made share the same binding; if *var* is modified by assignment, the modification is visible to all those views.
Inside a resumed continuation, a form which binds a special variable such as (let ((*var* 42)) ...) may terminate. As expected, this causes the binding to be removed, revealing either another local binding of *var* or the global binding. However, this unbinding only affects only that that executing continuation; it has no effect inside other instances of the same continuation or other continuations which capture the same variable. Unbinding isn't a mutation of the dynamic environment, but may be understood as merely the restoration of an earlier dynamic environment reference.
The following example shows an implementation of the suspend operator.
(defmacro suspend (:form form name var . body)
^(sys:capture-cont ',name (lambda (,var)
(sys:abscond-from ,name ,*body))
',form))
(sys:abscond-from name [value])
The sys:abscond-from operator closely resembles return-from and performs the same function: it causes an enclosing block name to terminate with value which defaults to nil.
However, unlike return-from, sys:abscond-from does not perform any unwinding.
This operator should never be used for any purpose other than implementing primitives for the use of delimited continuations. It is used by the yield-from and yield operators to escape out of a block in which a continuation has been captured. Neglecting to unwind is valid due to the expectation that control will return into a restarted copy of that context.
(sys:abscond* name [value])
The sys:abscond* function is similar to the sys:abscond-from operator, except that name is an ordinary function parameter, and so when return* is used, an argument expression must be specified which evaluates to a symbol. Thus sys:abscond* allows the target block of a return to be dynamically computed.
The following equivalence holds between the operator and function:
(sys:abscond-from a b) <--> (sys:abscond* 'a b)
Expressions used as name arguments to abscond* which do not simply quote a symbol have no equivalent in abscond-from.
(obtain forms*)
(yield-from name [form])
The obtain and yield-from macros closely interoperate.
The obtain macro treats zero or more forms as a suspendable execution context called the obtain block. It is expected that forms establish a block named name and return its result value to obtain.
Without evaluating any of the forms in the obtain block, obtain returns a function, which takes one optional argument. This argument, called the resume value, defaults to nil if it is omitted.
The function represents the suspended execution context.
The context is resumed whenever the function is called, and executes until the next yield-from statement which references the block named name. The function's reply argument is noted.
If the yield-from specifies a form argument, then the execution context suspends, and the resume function terminates and returns the value of that form. When the function is called again to resume the context, the yield-from returns the previously noted resume value (and the new resume value just passed is noted in its place).
If the yield-from specifies no form argument, then it briefly suspends the execution context only to retrieve the resume value, without producing an item. Since no item is produced, the resume function does not return. The execution context implicitly resumes.
When execution reaches the last form in the obtain block, the resume value is discarded. The execution context terminates, and the most recent call to the resume function returns the value of that last form.
The obtain macro registers a finalizer against the returned resume function. The finalizer invokes the function, passing it the symbol sys:cont-poison, thereby triggering unwinding in the most recently captured continuation. Thus, abandoned obtain blocks are subject to unwinding when they become garbage.
The yield-from macro works by capturing a continuation and performing a nonlocal exit to the nearest block called name. It passes a special yield object to that block. The obtain macro generates code which knows what to do with this special yield object.
The following example shows a function which recursively traverses a cons cell structure, yielding all the non-nil atoms it encounters. Finally, it returns the object nil. The function is invoked on a list, and the invocation is wrapped in an obtain block to convert it to a generating function.
The generating function is then called six times to retrieve the five atoms from the list, and the final nil value. These are collected into a list.
This example demonstrates the power of delimited continuations to suspend and resume a recursive procedure.
(defun yflatten (obj)
(labels ((flatten-rec (obj)
(cond
((null obj))
((atom obj) (yield-from yflatten obj))
(t (flatten-rec (car obj))
(flatten-rec (cdr obj))))))
(flatten-rec obj)
nil))
(let ((f (obtain (yflatten '(a (b (c . d)) e)))))
(list [f] [f] [f] [f] [f] [f]))
--> (a b c d e nil)
The following interactive session log exemplifies two-way communication between the main code and a suspending function.
Here, mappend is invoked on a list of symbols representing fruit and vegetable names. The objective is to return a list containing only fruits. The lambda function suspends execution and yields a question out of the map block. It then classifies the item as a fruit or not according to the reply it receives. The reply emerges as the result value of the yield-from call.
The obtain macro converts the block to a generating function. The first call to the function is made with no argument, because the argument would be ignored anyway. The function returns a question, asking whether the first item in the list, the potato, is a fruit. To answer positively or negatively, the user calls the function again, passing in t or nil, respectively. The function returns the next question, which is answered in the same manner.
When the question for the last item is answered, the function call yields the final item: the ordinary result of the block, which is the list of fruit names.
1> (obtain
(block map
(mappend (lambda (item)
(if (yield-from map `is @item a fruit?`)
(list item)))
'(potato apple banana lettuce orange carrot))))
#<interpreted fun: lambda (: reply)>
2> (call *1)
"is potato a fruit?"
3> (call *1 nil)
"is apple a fruit?"
4> (call *1 t)
"is banana a fruit?"
5> (call *1 t)
"is lettuce a fruit?"
6> (call *1 nil)
"is orange a fruit?"
7> (call *1 t)
"is carrot a fruit?"
8> (call *1 nil)
(apple banana orange)
The following example demonstrates an accumulator. Values passed to the resume function are added to a counter which is initially zero. Each call to the function returns the updated value of the accumulator. Note the use of (yield-from acc) with no arguments to receive the value passed to the first call to the resume function, without yielding an item. The first return value 1 is produced by the (yield-from acc sum) form, not by (yield-from acc). The latter only obtains the initial value 1 and uses it to establish the seed value of the accumulator. Without causing the resume function to terminate and return, control passes into the loop, which yields the first item, causing the resume function call (call *1 1) to return 1:
1> (obtain
(block acc
(let ((sum (yield-from acc)))
(while t (inc sum (yield-from acc sum))))))
#<interpreted fun: lambda (: resume-val)>
2> (call *1 1)
1
3> (call *1 2)
3
4> (call *1 3)
6
5> (call *1 4)
10
(obtain-block name forms*)
The obtain-block macro combines block and obtain into a single expression. The forms are evaluated in a block named name.
That is to say, the following equivalence holds:
(obtain-block n f ...) <--> (obtain (block n f ...))
(yield [form])
The yield macro is to yield-from as return is to return-from: it yields from an anonymous block.
It is equivalent to calling yield-from using nil as the block name.
In other words, the following equivalence holds:
(yield x) <--> (yield-from nil x)
;; Yield the integers 0 to 4 from a for loop, taking
;; advantage of its implicit anonymous block:
(defvarl f (obtain (for ((i 0)) ((< i 5)) ((inc i))
(yield i))))
[f] -> 0
[f] -> 1
[f] -> 2
[f] -> 3
[f] -> 4
[f] -> nil
[f] -> nil
(obtain* forms*)
(obtain*-block name forms*)
The obtain* and obtain*-block macros implement a useful variation of obtain and obtain-block.
The obtain* macro differs from obtain in exactly one regard: prior to returning the function, it invokes it one time, with the argument value nil.
Thus, the following equivalence holds
(obtain* forms ...) <--> (let ((f (obtain forms ...)))
(call f)
f)
In other words, the suspended block is immediately resumed, so that it executes either to completion (in which case its value is discarded), or to its first yield or yield-from call (in which case the yielded value is discarded).
Note: the obtain* macro is useful in creating suspensions which accept data rather than produce data.
The obtain*-block macro combines obtain* and block in the same manner that obtain-block combines obtain and block.
;; Pass three values into suspended block,
;; which get accumulated into list.
(let ((f (obtain*-block nil
(list (yield nil) (yield nil) (yield nil)))))
(call f 1)
(call f 2)
(call f 3)) -> (1 2 3)
;; Under obtain, extra call is required:
(let ((f (obtain-block nil
(list (yield nil) (yield nil) (yield nil)))))
(call f nil) ;; execute block to first yield
(call f 1) ;; resume first yield with 1
(call f 2)
(call f 3)) -> (1 2 3)
(suspend block-name var-name body-form*)
The suspend operator captures a continuation up to the prompt given by the symbol block-name and binds it to the variable name given by var-name, which must be a symbol suitable for binding variables with let.
Each body-form is then evaluated in the scope of the variable var-name.
When the last body-form is evaluated, a nonlocal exit takes place to the block named by block-name (using the sys:abscond-from operator, so that unwinding isn't performed).
When the continuation bound to var-name is invoked, a copy of the entire block block-name is restarted, and in that copy, the suspend call appears to return normally, yielding the value which had been passed to the continuation.
Define John McCarthy's amb function using block and suspend:
(defmacro amb-scope (. forms)
^(block amb-scope ,*forms))
(defun amb (. args)
(suspend amb-scope cont
(each ((a args))
(if a
(iflet ((r (call cont a)))
(return-from amb-scope r))))))
Use amb to bind the x and y which satisfy the predicate (eql (* x y) 8) nondeterministically:
(amb-scope
(let ((x (amb 1 2 3))
(y (amb 4 5 6)))
(amb (eql (* x y) 8))
(list x y)))
-> (2 4)
(hlet ({sym | (sym init-form)}*) body-form*)
(hlet* ({sym | (sym init-form)}*) body-form*)
The hlet and hlet* macros behave exactly like let and let*, respectively except that they guarantee that the variable bindings are allocated in storage which isn't captured by delimited continuations.
The h in the names stands for "heap", serving as a mnemonic based on the implementation concept of these bindings being "heap-allocated".
TXR provides a "pure" regular-expression implementation based on automata theory, which equates regular expressions, finite automata and sets of strings. A regular expression determines whether or not a string of input characters belongs to a set. TXR regular expressions do not support features such as "anchoring" a match to the start or end of a string, or capturing parenthesized subexpression matches into registers. Parenthesis syntax denotes only grouping, with no additional meaning.
The semantics of whether a regular expression is used for a substring search, prefix match, suffix match, string splitting and so forth comes from the functions which use regular expressions to perform these operations.
[regex [start [from-end]] string]
A regular expression is callable as a function in TXR Lisp. When used this way, it requires a string argument. It searches the string for the leftmost match for itself, and returns the matching substring, which could be empty. If no match is found, it returns nil.
A regex takes one, two, or three arguments. The required string is always the rightmost argument. This allows for convenient partial application over optional arguments using macros in the op family, and macros in which the op syntax is implicit.
The optional arguments start and from-end are treated exactly as their like-named counterparts in the search-regst function.
Keep those elements from a list of strings which match the regular expression #/a.*b/:
(keep-if #/a.*b/ '#"abracadabra zebra hat adlib adobe deer")
--> ("abracadabra" "adlib" "adobe")
(search-regex string regex [start [from-end]])
(range-regex string regex [start [from-end]])
(search-regst string regex [start [from-end]])
The search-regex function searches through string starting at position start for a match for regex.
If start is omitted, the search starts at position 0. If from-end is specified and has a non-nil value, the search proceeds in reverse, from the position just beyond the last character of string, toward start.
If start exceeds the length of the string, then search-regex returns nil.
If start is negative then it indicates positions from the end of the string, such that -1 is the last character, -2 the second last and so forth. If the value is so negative that it refers beyond the start of the string, then the starting position is deemed to be zero.
If start is equal to the length of string, and thus refers to the position one character past its length, then a match occurs at that position if regex admits such a match.
The search-regex function returns nil if no match is found, otherwise it returns a cons, whose car indicates the position of the match, and whose cdr indicates the length of the match.
If regex is capable of matching empty strings, and no other kind of match is found within string, then search regex reports a zero length match. If from-end is false, then this match is reported at start, otherwise it is reported at the position one character beyond the end of the string.
The range-regex function is similar to search-regex, except that when a match is found, it returns a position range, rather than a position and length. A range object is returned whose from field indicates the position of the match, and whose to indicates the position one element past the last character of the match. If the match is empty, the two integers are equal.
Also see the rr function, which provides an alternative argument syntax for the semantics of range-regex.
The search-regst differs from search-regex in the representation of the return value in the matching case. Rather than returning the position and length of the match, it returns the matching substring of string.
(match-regex string regex [position])
(match-regst string regex [position])
The match-regex function tests whether regex matches at position in string.
If position is not specified, it is taken to be zero. Negative values of position index from the right end of the string such that -1 refers to the last character. Excessively negative values which index before the first character cause nil to be returned.
If the regex matches, then the length of the match is returned. If it does not match, then nil is returned.
The match-regst differs from match-regex in the representation of the return value in the matching case. Rather than returning the length of the match, it returns matching substring of string.
(match-regex-right string regex [end-position])
(match-regst-right string regex [end-position])
The match-regex-right function tests whether some substring of string which terminates at the character position just before end-position matches regex.
If end-position is not specified, it defaults to the length of the string, and the function performs a right-anchored regex match.
The end-position argument can be a negative integer, in which case it denotes positions from the end of the string, such that -1 refers to the last character. If the value is excessively negative such that the position immediately before it is before the start of the string, then nil is returned.
If end-position is a positive value beyond the length of string, then, likewise, nil is returned.
If a match is found, then the length of the match is returned.
A more precise way of articulating the role of end-position is that for the purposes of matching, string is considered to terminate just before end-position: in other words, that end-position is the length of the string. The match is then anchored to the end of this effective string.
The match-regst-right differs from match-regst-right in the representation of the return value in the matching case. Rather than returning the length of the match, it returns the matching substring of string.
;; Return matching portion rather than length thereof.
(defun match-regex-right-substring (str reg : end-pos)
(set end-pos (or end-pos (length str)))
(let ((len (match-regex-right str reg end-pos)))
(if len
[str (- end-pos len)..end-pos]
nil)))
(match-regex-right-substring "abc" #/c/) -> ""
(match-regex-right-substring "acc" #/c*/) -> "cc"
;; Regex matches starting at multiple positions, but all
;; the matches extend past the limit.
(match-regex-right-substring "acc" #/c*/ 2) -> nil
;; If the above behavior is not wanted, then
;; we can extract the string up to the limiting
;; position and do the match on that.
(match-regex-right-substring ["acc" 0..2] #/c*/) -> "c"
;; Equivalent of above call
(match-regex-right-substring "ac" #/c*/) -> "c"
(regex-prefix-match regex string [position])
The regex-prefix-match determines whether the input string might be the prefix of a string which matches regular expression regex.
The result is true if the input string matches regex exactly. However, it is also true in situations in which the input string doesn't match regex, yet can be extended with one or more additional characters beyond the end such that the extended string does match.
The string argument must be a character string. The function takes the input string to be the suffix of string which starts at the character position indicated by the position argument. If that argument is omitted, then string is taken as the input in its entirety. Negative values index backwards from the end of string according to the usual conventions elsewhere in the library.
Note: this function is not to be confused for the semantics of a regex matching a prefix of a string: that capability is provided by the functions match-regex, m^, r^, f^ and fr^.
;; The empty string is not a viable prefix match for
;; a regex that matches no strings at all:
(regex-prefix-match #/~.*/ "") -> nil
(regex-prefix-match #/[]/ "") -> nil
;; The empty string is a viable prefix of any regex
;; which matches at least one string:
(regex-prefix-match #// "") -> t
(regex-prefix-match #/abc/ "") -> t
;; This string doesn't match the regex because
;; it doesn't end in b, but is a viable prefix:
(regex-prefix-match #/a*b/ "aa") -> t
(regex-prefix-match #/a*b/ "ab") -> t
(regex-prefix-match #/a*b/ "ac") -> nil
(regex-prefix-match #/a*b/ "abc") -> nil
(regsub regex replacement string)
(regsub substring replacement string)
(regsub function replacement string)
The regsub function operates in two modes, depending on whether the first argument is a regular expression, or function.
If the first argument is a regular expression or string, then regsub searches string for multiple occurrences of non-overlapping matches for that regex or substring. A new string is constructed similar to string but in which each matching region is replaced with using replacement as follows.
The replacement object may be a character or a string, in which case it is simply taken to be the replacement for each match of the regular expression.
The replacement object may be a function of one argument, in which case for every match which is found, this function is invoked, with the matching piece of text as an argument. The function's return value is then taken to be the replacement text.
If the first argument is a function, then it is called, with string as its argument. The return value must be either a range object (see the rcons function) which indicates the extent of string to be replaced, or else nil which indicates that no replacement is to take place.
;; match every lowercase e or o, and replace by filtering
;; through the upcase-str function:
[regsub #/[eo]/ upcase-str "Hello world!"] -> "HEllO wOrld!"
;; Replace Hello with Goodbye:
(regsub #/Hello/ "Goodbye" "Hello world!") -> "Goodbye world!"
;; Same, as a simple substring match, rather than regex:
(regsub "Hello" "Goodbye" "Hello world!") -> "Goodbye world!"
;; Left-anchored replacement with r^ function:
(regsub (fr^ #/H/) "J" "Hello, hello!") -> "Jello, hello!"
(regexp obj)
The regexp function returns t if obj is a compiled regular-expression object. For any other object type, it returns nil.
(trim-left {regex | prefix} string)
(trim-right {regex | suffix} string)
The trim-left and trim-right functions return a new string, equivalent to string with a leading or trailing portion removed.
If the first argument is a regular expression regex, then, respectively, trim-left and trim-right find a prefix or suffix of string which matches the regular expression. If there is no match, or if the match is empty, then string is returned. Otherwise, a copy of string is returned in which the matching characters are removed. If regex matches all of string then the empty string is returned.
If the first argument is a character string, then it is treated as an exact match for that sequence of characters. Thus, trim-left interprets that string as a prefix to be removed, and trim-right as a suffix. If string starts with prefix, then trim-left returns a copy of string with prefix removed. Otherwise, string is returned. Likewise, if string ends with suffix, then trim-right returns a copy of string with suffix removed. Otherwise, string is returned.
(regex-compile form-or-string [error-stream])
The regex-compile function takes the source code of a regular expression, expressed as a Lisp data structure representing an abstract syntax tree, or else a regular expression specified as a character string, and compiles it to a regular-expression object.
If form-or-string is a character string, it is parsed to an abstract syntax tree first, if by the regex-parse function. If the parse is successful (the result is not nil) then the resulting tree structure is compiled by a recursive call to regex-compile.
The optional error-stream argument is passed down to regex-parse as well as in the recursive call to regex-compile, if that call takes place.
If error-stream is specified, it must be a stream. Any error diagnostics are sent to that stream.
;; the equivalent of #/[a-zA-Z0-9_]/
(regex-compile '(set (#\a . #\z) (#\A . #\Z)
(#\0 . #\9) #\_))
;; the equivalent of #/.*/ and #/.+/
(regex-compile '(0+ wild))
(regex-compile '(1+ wild))
;; #/a|b|c/
(regex-compile '(or (or #\a #\b) #\c))
;; string
(regex-compile "a|b|c")
(regex-source regex)
The regex-source function returns the source code of compiled regular expression regex.
The source code isn't the textual notation, but the Lisp data structure representing the abstract syntax tree: the same representation as what is returned by regex-parse.
(regex-parse string [error-stream])
The regex-parse function parses a character string which contains a regular expression and turns it into a Lisp data structure (the abstract syntax tree representation of the regular expression).
The regular-expression syntax #/RE/ produces the same structure, but as a literal which is processed at the time TXR source code is read; the regex-parse function performs this parsing at run time.
If there are parse errors, the function returns nil.
The optional error-stream argument specifies a stream to which error messages are sent from the parser. By default, diagnostic output goes to the *stdnull* stream, which discards it. If error-stream is specified as t, then the diagnostic output goes to the *stdout* stream.
If regex-parse returns a non-nil value, that structure is then something which is suitable as input to regex-compile.
There is a small difference in the syntax accepted by regex-parse and the syntax of regular-expression literals. Any / (slash) characters occurring in any position within string are treated as ordinary characters, not as regular-expression delimiters. The call (regex-parse "/a/") matches three characters: a slash, followed by the letter "a", followed by another slash. Note that the slashes are not escaped.
Note: if a regex-parse call is written using a string literal as the string argument, then note that any backslashes which are to be processed by the regular expression must be doubled up, otherwise they belong to the string literal:
(regex-parse "\*") ;; error, invalid string literal escape
(regex-parse "\\*") ;; correct: the \* literal match for *
The double backslash in the string literal produces a single backslash in the resulting string object that is processed by regex-parse.
(regex-optimize regex-tree-syntax)
The regex-compile function accepts the source code of a regular expression, expressed as a Lisp data structure representing an abstract syntax tree, and calculates an equivalent structure in which certain simplifications have been performed, or in some cases substitutions which eliminate the dependence on derivative-based processing.
The regex-tree-syntax argument is assumed to be correct, as if it were produced by the regex-parse or regex-from-trie functions. Incorrect syntax produces unspecified results: an exception may be thrown, or some object may appear to be successfully returned.
Note: it is unnecessary to call this function to prepare the input for regex-compile because that function optimizes internally. However, the source code attached to a compiled regular-expression object is the original unoptimized syntax tree, and that is used for rendering the #/.../ notation when the object is printed. If the syntax is passed through regex-optimize before regex-compile, the resulting object will have the optimized code attached to it, and subsequently render that way in printed form.
;; a|b|c -> [abc]
(regex-optimize '(or #\a (or #\b #\c))) -> (set #\a #\b #\c)
;; (a|) -> a?
(regex-optimize '(or #\a nil)) -> (? #\a)
(read-until-match regex [stream [include-match]])
The read-until-match function reads characters from stream, accumulating them into a string, which is returned.
If an argument is not specified for stream, then the *stdin* stream is used.
The include-match argument is Boolean, indicating whether the delimiting text matched by regex is included in the returned string. It defaults to nil.
The accumulation of characters is terminated by a match on regex, the end of the stream, or an error.
This means that characters are read from the stream and accumulated while the stream has more characters available, and while its prefix does not match regex.
If regex matches the stream before any characters are accumulated, then an empty string is returned.
If the stream ends or a non-exception-throwing error occurs before any characters are accumulated, the function returns nil.
When the accumulation of characters terminates by a match on regex, the longest possible matching sequence of characters is removed from the stream. If include-match is true, that matching text is included in the returned string. Otherwise, it is discarded. The next available character in the stream is the first nonmatching character following the matched text. However, the next available character, as well as some number of subsequent characters, may originate from the stream's push-back buffer, rather than from the underlying operating system object, due to this function's internal use of the unget-char function. Therefore, the stream position, as would be reported by seek-stream, is unspecified.
(scan-until-match regex [stream])
(count-until-match regex [stream])
The functions scan-until-match and count-until-match read characters from stream until a match occurs in the stream for regular expression regex, the stream runs out of characters, or an error occurs.
If the stream runs out of characters, or a non-exception-throwing error occurs, before a match for regex is identified, these functions return nil.
If a match for regex occurs in stream, then count-until-match returns the number of characters that were read and discarded prior to encountering the first matching character. In the same situation, the scan-until-match function returns a cons cell whose car holds the count of discarded characters, that being the same value as what would be returned by count-until-match, and whose cdr holds a character string that comprises the text matched by regex. The text matched by regex is as long as possible, and is removed from the stream. The next available character in the stream is the first nonmatching character following the matched text. However, the next available character, as well as some number of subsequent characters, may originate from the stream's push-back buffer, rather than from the underlying operating system object, due to these functions' internal use of the unget-char function. Therefore, the stream position, as would be reported by seek-stream, is unspecified.
(m^$ regex [position] string)
(m^ regex [position] string)
(m$ regex [end-position] string)
These functions provide functionality similar to the match-regst and match-regst-right functions, but under alternative interfaces which are more convenient.
The ^ and $ notation used in their names are an allusion to the regular-expression search-anchoring operators found in familiar POSIX utilities such as grep.
The position argument, if omitted, defaults to zero, so that the entire string is operated upon.
The end-position argument defaults to the length of string, so that the end position coincides with the end of the string.
If the position or end-position arguments are negative, they index backwards from the length of string so that -1 denotes the last character.
A value in either parameter which is excessively negative or positive, such that it indexes before the start of the string or exceeds its length results in a failed match and consequently nil being returned.
The m^$ function tests whether the entire portion of string starting at position through to the end of the string is in the set of strings matched by regex. If this is true, then that portion of the string is returned. Otherwise nil is returned.
The m^ function tests whether the portion of the string starting at position has a prefix which matches regex. If so, then this matching prefix is returned. Otherwise nil is returned.
The m$ function tests whether the portion of string ending just before end-position has a suffix which matches regex. If so, then this matching suffix is returned. Otherwise nil is returned.
(r^$ regex [position] string)
(r^ regex [position] string)
(r$ regex [end-position] string)
(rr regex [position [from-end]] string)
The first three of these functions perform the same operations as, respectively, m^$, m^ and m$, with the same argument conventions. They differ in return value. When a match is found, they return a range value indicating the extent of the matching substring within string rather than the matching substring itself.
The rr function performs the same operation as range-regex with different conventions with regard to argument order, harmonizing with those of the other three functions above.
The position argument, if omitted, defaults to zero, so that the entire string is operated upon.
The end-position argument defaults to the length of string, so that the end position coincides with the end of the string.
With one exception, a value in either parameter which is excessively negative or positive, such that it indexes before the start of the string or exceeds its length results in a failed match and consequently nil being returned. The exception is that the rr function permits a negative position value which refers before the start of the string; this is effectively treated as zero.
The from-end argument defaults to nil.
The r^$ function tests whether the entire portion of string starting at position through to the end of the string is in the set of strings matched by regex. If this is true, then the matching range is returned, as a range object.
The r^ function tests whether the portion of the string starting at position has a prefix which matches regex. If so, then the matching range is returned, as a range object. Otherwise nil is returned.
The r$ function tests whether the portion of string ending just before end-position has a suffix which matches regex. If so, then the matching range is returned. Otherwise nil is returned.
The rr function searches string starting at position for a match for regex. If from-end is specified and true, the rightmost match is reported. If a match is found, it is reported as a range.
A regular expression which matches empty strings matches at the start position, and every other position, including the position just after the last character, coinciding with the length of string.
Except for the different argument order such that string is always the rightmost argument, the rr function is equivalent to the range-regex function, such that correspondingly named arguments have the same semantics.
(rra regex [start [end]] string)
The rra function searches string between the start and end position for matches for the regular expression regex.
The matches are returned as a list of range objects.
The start argument defaults to zero, and end defaults to the length of the string (the position one past the last character).
Negative values of start and end indicate positions from the end of the string, such that -1 denotes the last character, -2 the second-to-last and so forth.
If start is so negative that it refers before the start of string, it is treated as zero. If this situation is true of the end argument, then the function returns nil.
If start refers to a character position beyond the length of string (two characters or more beyond the end of the string), then the function returns nil. If this situation is true of end, then end is curtailed to the string length.
The rra function returns all non-overlapping matches, including zero length matches. Zero length matches may occur before the first character of the string, or after the last character. If so, these are included.
(f^$ regex [position])
(f^ regex [position])
(f$ regex [end-position])
These regular-expression functions do not directly perform regex operations. Rather, they each return a function of one argument which performs a regex operation.
The returned functions perform the same operations as, respectively, m^$, m^ and m$.
The following equivalences nearly hold, except that the functions on the right side produced by op can accept two arguments when only r is curried, whereas the functions on the left take only one argument:
[f^$ r] <--> (op m^$ r)
[f^$ r p] <--> (op m^$ r p)
[f^ r] <--> (op m^ r)
[f^ r p] <--> (op m^ r p)
[f$ r] <--> (op m$ r)
[f$ r p] <--> (op m$ r p)
That is to say, f^$ returns a function which binds regex and possibly the optional position. When this function is invoked, it must be given an argument which is a string. It performs the same operation as m^$ being called on regex and possibly position. The same holds between f^ and m^, and between f$ and m$.
;; produce list which contains only strings
;; beginning with "cat":
(keep-if (f^ #/cat/) '#"dog catalog cat fox catapult")
--> ("catalog" "cat" "catapult")
;; map all strings in a list to just their trailing
;; digits.
(mapcar (f$ #/\d*/) '#"a123 4 z bc465")
--> ("123" "4" "" "465")
;; check that all strings consist of digits after
;; the third position.
(all '#"ABC123 DFE45 12379" (f^$ #/\d*/ 3))
--> "79" ; i.e. true
(all '#"ABC123 DFE45 12379A" (f^$ #/\d*/ 3))
--> nil
(fr^$ regex [position])
(fr^ regex [position])
(fr$ regex [end-position])
(frr regex [[start-position] from-end])
These regular-expression functions do not directly perform regex operations. Rather, they each return a function of one argument which performs a regex operation.
The returned functions perform the same operations as, respectively, r^$, r^, r$ and rr.
The following equivalences nearly hold, except that some of the functions on the right side produced by op op can accept additional arguments after the input string, whereas the functions on the left produced by f^$ et al. accept only one parameter: the input string.
[fr^$ r] <--> (op m^$ r)
[fr^$ r p] <--> (op m^$ r p)
[fr^ r] <--> (op m^ r)
[fr^ r p] <--> (op m^ r p)
[fr$ r] <--> (op m$ r)
[fr$ r p] <--> (op m$ r p)
[frr r] <--> (op m$ r)
[frr r s] <--> (op m$ r s)
[frr r s fe] <--> (op m$ r s fe)
That is to say, fr^$ returns a function which binds regex and possibly the optional position. When this function is invoked, it must be given an argument which is a string. It performs the same operation as r^$ being called on regex and possibly position, and the string. The same holds between fr^ and r^, between fr$ and r$, and between frr and rr.
;; Remove leading digits from "123A456",
;; other than first digit:
(regsub (fr^ #/\d+/ 1) "" "123A456")
--> "1A456"
A hash table is an object which retains an association between pairs of objects. Each pair consists of a key and a value. Given an object which is similar to a key in the hash table, it is possible to retrieve the corresponding value. Entries in a hash table are not ordered in any way, and lookup is facilitated by hashing: quickly mapping a key object to a numeric value which is then used to index into one of many buckets where the matching key will be found (if such a key is present in the hash table).
In addition to keys and values, a hash table contains a storage location which allows it to be associated with user data.
Important to the operation of a hash table is the criterion by which keys are considered same. By default, this similarity follows the eql function. A hash table will search for a stored key which is eql to the given search key. A hash table constructed with the equal-based property compares keys using the equal function instead.
In addition to storing key-value pairs, a hash table can have a piece of information associated with it, called the user data.
TXR hash tables contain a seed value which permutes the hashing operation, at least for keys of certain types. This feature, if the seed is randomized, helps to prevent software from being susceptible to hash collision denial-of-service attacks. However, by default, the seed is not randomized. Newly created hash tables for which a seed value is not specified take their seed value from the *hash-seed* special variable, which is initialized to zero. That includes hash tables created by parsing hash literal syntax. Security-sensitive programs requiring protection against collision attacks may use gen-hash-seed to create a randomized hash seed, and, depending on their specific need, either store that value in *hash-seed*, or pass the value to hash-table constructors like make-hash, or both. Note: randomization of hash seeding isn't a default behavior because it affects program reproducibility. The seed value affects the order in which keys are traversed, which can change the output of programs whose inputs have not changed, and whose logic is is otherwise deterministic.
A hash table can be traversed to visit all of the keys and data. The order of traversal bears no relation to the order of insertion, or to any properties of the key type.
During an open traversal, new keys can be inserted into a hash table or deleted from it while a a traversal is in progress. Insertion of a new key during traversal will not cause any existing key to be visited twice or to be skipped; however, it is not specified whether the new key will be traversed. Similarly, if a key is deleted during traversal, and that key has not yet been visited, it is not specified whether it will be visited during the remainder of the traversal. These remarks apply not only to deletion via remhash or the del operator, but also to wholesale deletion of all keys via clearhash.
The garbage collection of hash tables supports weak keys and weak values. If a hash table has weak keys, this means that from the point of view of garbage collection, that table holds only weak references to the keys stored in it. Similarly, if a hash table has weak values, it means that it holds a weak reference to each value stored. A weak reference is one which does not prevent the reclamation of an object by the garbage collector. That is to say, when the garbage collector discovers that the only references to some object are weak references, then that object is considered garbage, just as if it had no references to it. The object is reclaimed, and the weak references "lapse" in some way, which depends on what kind they are. Hash-table weak references lapse by entry removal. When an object used as a key in one or more weak-key hash tables becomes unreachable, those hash entries disappear. This happens even if the values are themselves reachable. Vice versa, when an object appearing as a value in one or more weak-value hash tables becomes unreachable, those entries disappear, even if the keys are reachable. When a hash table has both weak keys and weak values, then an the behavior is one of two possible semantics. Under the or-semantics, the hash table entry is removed if either the key or the value is unreachable. Under the and-semantics, the entry is removed only if both the key and value are unreachable.
If the keys of a weak-key hash table are reachable from the values, or if the values of a weak-key hash table are reachable from the keys, then the weak semantics is defeated for the affected entries: the hash table retains those entries as if it were an ordinary table. A hash table with both weak keys and values does not have this issue, regardless of its semantics.
An open traversal of a hash table is performed by the maphash function and the dohash operator. The traversal is open because code supplied by the program is evaluated for each entry.
The functions hash-keys, hash-values, hash-pairs, and hash-alist also perform an open traversal, because they return lazy lists. The traversal isn't complete until the returned lazy list is fully instantiated. In the meanwhile, the TXR program can mutate the hash table from which the lazy list is being generated.
Certain hash operations expose access to the internal key-value association entries of a hash table, which are represented as ordinary cons cells. Modifying the car field of such a cell potentially violates the integrity of the hash table; the behavior of subsequent lookup and insertion operations becomes unspecified.
Similarly, if an object is used as a key in an equal-based hash table, and that object is mutated in such a way that its equality to other objects under the equal function is affected or its hash value under hash-equal is altered, the behavior of subsequent lookup and insertion operations on the becomes unspecified.
(make-hash weak-keys weak-vals
equal-based [hash-seed])
(hash {:weak-keys | :weak-vals | :weak-or | :weak-and
:eql-based | :equal-based |
:eq-based | :userdata obj}*)
These functions construct a new hash table.
make-hash takes three mandatory Boolean arguments. The Boolean weak-keys argument specifies whether the hash table shall have weak keys. The weak-vals argument specifies whether it shall have weak values, and equal-based specifies whether it is equal-based.
If the weak-keys argument is one of the keywords :weak-and or :weak-or then the hash table shall have both weak keys and weak values, with the semantics implied by the keyword: :weak-and specifies and-semantics and :weak-or specifies or-semantics. The weak-vals argument is then ignored.
If both weak-keys and weak-vals are true, and weak-keys is not one of the keywords :weak-and or :weak-or, then the hash table has or-semantics.
The hash function defaults all three of these properties to false, and allows them to be overridden to true by the presence of keyword arguments.
The optional hash-seed parameter must be an integer, if specified. Its value perturbs the hashing function of the hash table, which affects :equal-based hash tables, when character strings and buffers are used as keys. If hash-seed is omitted, then the value of the *hash-seed* variable is used as the seed.
It is an error to attempt to construct an equal-based hash table which has weak keys.
The hash function provides an alternative interface. It accepts optional keyword arguments. The supported keyword symbols are: :weak-keys, :weak-vals, :weak-and, :weak-or, :equal-based, :eql-based :eq-based and :userdata which can be specified in any order to turn on the corresponding properties in the newly constructed hash table.
Only one of :equal-based, :eql-based and :eq-based may be specified. If specified, then the hash table uses equal, eql or eq equality, respectively, for considering two keys to be the same key. If none of these is specified, the hash function produces an :equal-based hash table by default.
If :weak-keys, :weak-and or :weak-or is specified, then :equal-based may not be specified.
At most one of :weak-and or :weak-or may be specified. If either of these is specified, then the :weak-keys and :weak-vals keywords are redundant and unnecessary.
If :weak-keys and :weak-vals are both specified, and :weak-and isn't specified, the situation is equivalent to :weak-or.
If :userdata is present, it must be followed by an argument value; that value specifies the user data for the hash table, which can be retrieved using the hash-userdata function.
Note: there doesn't exist a keyword for specifying the seed. This omission is deliberate. These hash construction keywords may appear in the hash literal #H syntax. A seed keyword would allow literals to specify their own seed, which would allow malicious hash literals to be crafted that perpetrate a hash collision attack against the parser.
(hash-construct hash-args key-val-pairs)
(hash-from-pairs key-val-pairs hash-arg*)
(hash-from-alist alist hash-arg*)
The hash-construct function constructs a populated hash in one step. The hash-args argument specifies a list suitable as an argument list in a call to the hash function. The key-val-pairs is a sequence of pairs, which are two-element lists representing key-value pairs.
A hash is constructed as if by a call to (apply hash hash-args), then populated with the specified pairs, and returned.
The hash-from-pairs function is an alternative interface to the same semantics. The key-val-pairs argument is first, and the hash-args are passed as trailing variadic arguments, rather than a single list argument.
The hash-from-alist function is similar to hash-from-pairs, except that the alist argument specifies they keys and values as an association list. The elements of the list are cons cells, each of whose car is a key, and whose cdr is the value.
(hash-list key-list hash-arg*)
The hash-list function constructs a hash as if by a call to [apply hash hash-args], where hash-args is a list of the individual hash-arg variadic arguments.
The hash is then populated with keys taken from key-list and returned.
The value associated with each key is that key itself.
(hash-zip key-seq value-seq hash-arg*)
The hash-zip function constructs a hash as if by a call to (apply hash hash-args), where hash-args is a list of the individual hash-arg variadic arguments.
The hash is then populated with keys taken from key-seq which are paired with values taken from from value-seq, and returned.
If key-seq is longer than value-seq, then the excess keys are ignored, and vice versa.
(hash-props {key value}*)
The hash-props function constructs a populated hash table without requiring the caller to construct a list of entries. The hash table contents are specified as direct arguments.
The hash-props function requires an even number of arguments, which are interleaved key-value pairs.
The returned hash table is equal-based, and no parameters are available for customizing any of its properties, such as weakness.
(hash-map function sequence hash-arg*)
The hash-map function constructs a a hash table from a sequence of keys and a function which maps them to values.
The function argument must be a function that can be called with one argument.
The elements of sequence become the keys of the returned hash table. The value associated with each key is determined by passing that value to function fun and taking the returned value.
The remaining hash-arg arguments determine what kind of hash table is created, as if the hash function were applied to them.
If the sequence contains duplicate elements (according to the hash table equality in effect for the hash table being constructed), duplicate elements later in the sequence replace earlier elements.
(hash-update hash function)
The hash-update function replaces each value in hash, with the value of function applied to that value.
The return value is hash.
(hash-update-1 hash key function [init])
The hash-update-1 function operates on a single entry in the hash table.
If key exists in the hash table, then its corresponding value is passed into function, and the return value of function is then installed in place of the key's value. The value is then returned.
If key does not exist in the hash table, and no init argument is given, then hash-update-1 does nothing and returns nil.
If key does not exist in the hash table, and an init argument is given, then function is applied to init, and then key is inserted into hash with the value returned by function as the datum. This value is also returned.
(group-by by-fun sequence option*)
(group-map by-fun filter-fun sequence option*)
The group-by function produces a hash table from sequence. Entries of the hash table are not elements of sequence, but lists of elements of sequence. The function by-fun is applied to each element of sequence to compute a key. That key is used to determine which list the item is added to in the hash table.
The trailing arguments option* if any, consist of the same keywords that are understood by the hash function, and determine the properties of the hash.
The group-map fun extends the semantics of group-by with a filtering step. It groups the elements of sequence in exactly the same manner, using by-fun. These lists of elements are then passed to filter-fun whose return values become the values associated with the hash table keys.
The effect of group-map may be obtained by a combination of group-by and hash-update according to the following equivalence:
(group-map bf ff seq) <--> (let ((h (group-by bf seq)))
(hash-update h ff))
Group the integers from 0 to 10 into three buckets keyed on 0, 1 and 2 according to the modulo 3 congruence:
(group-by (op mod @1 3) 0..11)
-> #H(() (0 (0 3 6 9)) (1 (1 4 7 10)) (2 (2 5 8)))
Same as above, but associate the keys with the sums of the buckets:
[group-map (op mod @1 3) sum 0..11]
-> #H(() (0 18) (1 22) (2 15))
(group-reduce hash classify-fun binary-fun seq
[init-value [filter-fun]])
The group-reduce updates hash table hash by grouping and reducing sequence seq.
The function regards the hash table as being populated with keys denoting accumulator values. Missing accumulators which need to be created in the hash table are initialized with init-value which defaults to nil.
The function iterates over seq and treats each element according to the following steps:
After the above processing, one more step is performed if the filter-fun argument is present. In this case, the hash table is destructively mapped through filter-fun before being returned. That is to say, every value in the hash table is projected through filter-fun and stored back in the table under the same key, as if by an invocation the (hash-update hash filter-fun) expression.
If group-reduce is invoked on an empty hash table, its net result closely resembles a group-by operation followed by separately performing a reduce-left on each value in the hash.
Frequency histogram:
[group-reduce (hash) identity (do inc @1)
"fourscoreandsevenyearsago" 0]
--> #H(() (#\a 3) (#\c 1) (#\d 1) (#\e 4) (#\f 1)
(#\g 1) (#\n 2) (#\o 3) (#\r 3) (#\s 3)
(#\u 1) (#\v 1) (#\y 1))
Separate the integers 1–10 into even and odd, and sum these groups:
[group-reduce (hash) evenp + 1..11 0]
-> #H(() (t 30) (nil 25))
(hist-sort sequence option*)
(hist-sort-by by-fun sequence option*)
The hist-sort function produces a histogram in the form of an association list, which is sorted in descending order of frequency. The keys in the association list are elements of sequence and the values are the frequency values: positive integers indicating how many times the keys occur in sequence.
Note: for a description of association lists, see the assoc function, and the section Association Lists in which its description is contained.
The hist-sort function works by internally constructing a hash table, which is not returned. Elements of sequence serve as keys in that hash. The trailing arguments option* if any, consist of the same keywords that are understood by the hash function, and determine the properties of that hash.
The hist-sort-by function differs from hist-sort in that it requires an additional argument by-fun with the following semantics: every element of sequence is passed to by-fun such that the resulting value is used as the hash key in the resulting histogram.
Thus, an invocation of hist-sort is equivalent to an invocation of hist-sort-by where the by-fun argument is specified as the identity function.
(hist-sort nil) -> nil
(hist-sort '(3 4 5)) -> ((3 . 1) (4 . 1) (5 . 1))
(hist-sort '("a" "b" "c" "a" "b" "a" "b" "a"))
-> (("a" . 4) ("b" . 3) ("c" . 1))
(make-similar-hash hash)
(copy-hash hash)
The make-similar-hash and copy-hash functions create a new hash object based on the existing hash object.
make-similar-hash produces an empty hash table which inherits all of the attributes of hash. It uses the same kind of key equality, the same configuration of weak keys and values, and has the same user data (see the set-hash-userdata function).
The copy-hash also produces a hash table similar to hash, in the same way as make-similar-hash. However, rather than producing producing an empty hash table, it returns a duplicate table which has all the same elements as hash: it contains the same key and value objects.
(inhash hash key [init])
The inhash function searches hash table hash for key. If key is found, then it return the hash table's cons cell which represents the association between hash and key. Otherwise, it returns nil.
If argument init is specified, then the function will create an entry for key in hash whose value is that of init. The cons cell representing that association is returned.
Note: for as long as the key continues to exist inside hash. modifying the car field of the returned cons has ramifications for the logical integrity of the hash; doing so results in unspecified behavior for subsequent insertion and lookup operations.
Modifying the cdr field has the effect of updating the association with a new value.
(gethash hash key [alt])
(set (gethash hash key [alt]) new-value)
The gethash function searches hash table hash for key key. If the key is found then the associated value is returned. Otherwise, if the alt argument was specified, it is returned. If the alt argument was not specified, nil is returned.
A valid gethash form serves as a place. It denotes either an existing value in a hash table or a value that would be created by the evaluation of the form. The alt argument is meaningful when gethash is used as a place, and, if present, is always evaluated whenever the place is evaluated. In place update operations, it provides the initial value, which defaults to nil if the argument is not specified. For example (inc (gethash h k d)) will increment the value stored under key k in hash table h by one. If the key does not exist in the hash table, then the value (+ 1 d) is inserted into the table under that key. The expression d is always evaluated, whether or not its value is needed.
If a gethash place is subject to a deletion, but doesn't exist, it is not an error. The operation does nothing, and nil is considered the prior value of the place yielded by the deletion.
(sethash hash key value)
The sethash function places a value into hash table under the given key. If a similar key already exists in the hash table, then that key's value is replaced by value. Otherwise, the key and value pair is newly inserted into hash.
The sethash function returns the value argument.
(pushhash hash key element)
The pushhash function is useful when the values stored in a hash table are lists. If the given key does not already exist in hash, then a list of length one is made which contains element, and stored in hash table under key. If the key already exists in the hash table, then the corresponding value must be a list. The element value is added to the front of that list, and the extended list then becomes the new value under key.
The return value is Boolean. If true, indicates that the hash-table entry was newly created. If false, it indicates that the push took place on an existing entry.
(remhash hash key)
The remhash function searches hash for a key similar to the key. If that key is found, then that key and its corresponding value are removed from the hash table.
If the key is found and removal takes place, then the associated value is returned. Otherwise nil is returned.
(clearhash hash)
The clearhash function removes all key-value pairs from hash, causing it to be empty.
If hash is already empty prior to the operation, then nil, is returned.
Otherwise an integer is returned indicating the number of entries that were purged from hash.
(hash-count hash)
The hash-count function returns an integer representing the number of key-value pairs stored in hash.
(hash-userdata hash)
(set (hash-userdata hash) new-value)
The hash-userdata function retrieves the user data object associated with hash.
A hash table can be created with user data using the :userdata keyword in a hash-table literal or in a call to the hash function, directly, or via other hash-constructing functions which take the hash construction keywords, such as group-by. If a hash table is created without user data, its user data is initialized to nil.
Because hash-userdata is an accessor, a hash-userdata form can be used as a place. Assigning a value to this place causes the user data of hash to be replaced with that value.
(get-hash-userdata hash)
The get-hash-userdata function is a deprecated synonym for hash-userdata.
(set-hash-userdata hash object)
The set-hash-userdata replaces, with the object, the user data object associated with hash.
(hashp object)
The hashp function returns t if the object is a hash table, otherwise it returns nil.
(maphash binary-function hash)
The maphash function successively invokes binary-function for each entry stored in hash. Each entry's key and value are passed as arguments to binary-function.
The function returns nil.
(hash-revget hash value [testfun [keyfun]])
(hash-keys-of hash value [testfun [keyfun]])
The hash-revget function performs a reverse lookup on hash.
It searches through the entries stored in hash for an entry whose value matches value.
If such an entry is found, that entry's key is returned. Otherwise nil is returned.
If multiple matching entries exist, it is not specified which entry's key is returned.
The hash-keys-of function has exactly the same argument conventions, and likewise searches the hash. However, it returns a list of all keys whose values match value.
The keyfun function is applied to each value in hash and the resulting value is compared with value. The default keyfun is the identity function.
The comparison is performed using testfun.
The default testfun is the equal function.
(hash-invert hash [joinfun [unitfun hash-arg*]])
The hash-invert function calculates and returns an inversion of hash table hash. The values in hash become keys in the returned hash table. Conversely, the values in the returned hash table are derived from the keys.
The optional joinfun and unitfun arguments must be functions, if they are given. These functions determine the behavior of hash-invert with regard to duplicate values in hash which turn into duplicate keys. The joinfun function must be callable with two arguments, and joinfun must accept one argument. If joinfun is omitted, it defaults to the identity* function; unitfun defaults to identity.
The hash-invert function constructs a hash table as if by a call to the hash function, passing the hash-arg arguments which determine the properties of the newly created hash.
The new hash table is then populated by iterating over the key-value pairs of hash and inserting them as follows: The key from hash is turned into a value v1 by invoking the unitfun function on it, and taking the return value. The value from hash is used as a key to perform a lookup in the new hash table. If no entry exists, then a new entry is created, whose value is v1. Otherwise if the entry already exists, then the value v0 of that entry is combined with v1 by calling the joinfun on the arguments v0 and v1. The entry is updated with the resulting value.
The new hash table is then returned.
;; Invert simple 1 to 1 table:
(hash-invert #H(() (a 1) (b 2) (c 3)))
--> #H(() (1 a) (2 b) (3 c))
;; Invert table such that the keys of duplicate values
;; are accumulated into lists:
[hash-invert #H(() (1 a) (2 a) (3 c) (5 c) (7 d)) append list]
--> #H(() (d (7)) (c (3 5)) (a (1 2)))
;; Invert table such that keys of duplicate values are summed:
[hash-invert #H(() (1 a) (2 a) (3 c) (5 c) (7 d)) +]
--> #H(() (d 7) (c 8) (a 3))
(hash-eql object)
(hash-equal object [hash-seed])
These functions each compute an integer hash value from the internal representation of object, which satisfies the following properties. If two objects A and B are the same under the eql function, then (hash-eql A) and (hash-eql B) produce the same integer hash value. Similarly, if two objects A and B are the same under the equal function, then (hash-equal A) and (hash-equal B) each produce the same integer hash value. In all other circumstances, the hash values of two distinct objects are unrelated, and may or may not be the same.
Object of struct type may support custom hashing by way of defining an equality substitution via an equal method. See the Equality Substitution section under Structures.
The optional hash-seed value perturbs the hashing function used by hash-equal for strings and buffer objects. This seed value must be a nonnegative integer no wider than 64 bits: that is, in the range 0 to 18446744073709551615. If the value isn't specified, it defaults to zero. On systems with 32-bit addresses, only the low 32 bits of this value may be significant.
Effectively, each possible value of the significant part of the seed specifies a different hashing function. If two objects A and B are the same under the equal function, then (hash-equal A S) and (hash-equal B S) each produce the same integer hash value for any valid seed value S.
The value returned is a fixnum value, and may be negative. It may be any value in the range fixnum-min to fixnum-max.
(hash-keys hash)
(hash-values hash)
(hash-pairs hash)
(hash-alist hash)
These functions retrieve the bulk key-value data of hash table hash in various ways. hash-keys retrieves a list of the keys. hash-values retrieves a list of the values. hash-pairs retrieves a list of pairs, which are two-element lists consisting of the key, followed by the value. Finally, hash-alist retrieves the key-value pairs as a Lisp association list: a list of cons cells whose car fields are keys, and whose cdr fields are the values. Note that hash-alist returns the actual entries from the hash table, which are conses. Modifying the cdr fields of these conses constitutes modifying the hash values in the original hash table. Modifying the car fields interferes with the integrity of the hash table, resulting in unspecified behavior for subsequent hash insertion and lookup operations.
These functions all retrieve the keys and values in the same order. For example, if the keys are retrieved with hash-keys, and the values with hash-values, then the corresponding entries from each list pairwise correspond to the pairs in hash.
The list returned by each of these functions is lazy, and hence constitutes an open traversal of the hash table.
(dohash (key-var value-var hash-form [result-form])
body-form*)
The dohash operator iterates over a hash table. The hash-form expression must evaluate to an object of hash-table type. The key-var and value-var arguments must be symbols suitable for use as variable names. Bindings are established for these variables over the scope of the body-forms and the optional result-form.
For each element in the hash table, the key-var and value-var variables are set to the key and value of that entry, respectively, and each body-form, if there are any, is evaluated.
When all of the entries of the table are thus processed, the result-form is evaluated, and its return value becomes the return value of the dohash form. If there is no result-form, the return value is nil.
The result-form and body-forms are in the scope of an implicit anonymous block, which means that it is possible to terminate the execution of dohash early using (return value) or (return).
(hash-uni hash1 hash2 [joinfun [map1fun [map2fun]]])
(hash-join hash1 hash2 joinfun [hash1dfl [hash2dfl]])
(hash-diff hash1 hash2)
(hash-symdiff hash1 hash2)
(hash-isec hash1 hash2 [joinfun])
These functions perform basic set operations on hash tables in a nondestructive way, returning a new hash table without altering the inputs. The arguments hash1 and hash2 must be compatible hash tables. This means that their keys must use the same kind of equality.
The resulting hash table inherits attributes from hash1, as if created by the make-similar-hash function. If hash1 has userdata, the resulting hash table has the same userdata. If hash1 has weak keys, the resulting table has weak keys, and so forth.
The hash-uni function performs a set union. The resulting hash contains all of the keys from hash1 and all of the keys from hash2, and their corresponding values. If a key occurs both in hash1 and hash2, then it occurs only once in the resulting hash. In this case, if the joinfun argument is not given, the value associated with this key is the one from hash1. If joinfun is specified then it is called with two arguments: the respective data items from hash1 and hash2. The return value of this function is used as the value in the union hash. If map1fun is specified it must be a function that can be called with one argument. All values from hash1 are projected through this function: the function is applied to each value, and the function's return value is used in place of the original value. Similarly, if map2fun is present, specifies a function through which values from hash2 are projected. These two functions are independent of joinfun; they are applied to values without regard for whether their keys exist in both hashes or just one.
The hash-join function performs a union operation similar to, but usefully different from hash-uni. The joinfun argument is mandatory in hash-join, and is applied to all items, regardless of whether they are present in just one hash or both hashes. The arguments hash1dfl and hash2dfl specify default values used in invocations of joinfun for keys that are present only in one hash. These values default to nil. For every key that is present only in hash1, joinfun is invoked with that key's value as its left argument, and the hash2dfl value as the right argument. Conversely, for every key that is present only in hash2, joinfun is invoked with the hash1dfl value as the left argument, and that key's value as its right argument. For every key that is present in both hashes, joinfun is invoked with the values, respectively, from hash1 and hash2. The returned hash contains all the keys from both hashes, associated with the values returned by joinfun.
The hash-diff function performs a set difference. First, a copy of hash1 is made as if by the copy-hash function. Then from this copy, all keys which occur in hash2 are deleted.
The hash-symdiff function performs a symmetric difference. A new hash is returned which contains all of the keys from hash1 that are not in hash2 and vice versa: all of the keys from hash2 that are not in hash1. The keys carry their corresponding values from hash1 and hash2, respectively.
The hash-isec function performs a set intersection. The resulting hash contains only those keys which occur both in hash1 and hash2. If joinfun is not specified, the values selected for these common keys are those from hash1. If joinfun is specified, then for each key which occurs in both hash1 and hash2, it is called with two arguments: the respective data items. The return value is then used as the data item in the intersection hash.
(hash-subset hash1 hash2)
(hash-proper-subset hash1 hash2)
The hash-subset function returns t if the keys in hash1 are a subset of the keys in hash2.
The hash-proper-subset function returns t if the keys in hash1 are a proper subset of the keys in hash2. This means that hash2 has all the keys which are in hash1 and at least one which isn't.
Note: the return value may not be mathematically meaningful if hash1 and hash2 use different equality. In any case, the actual behavior may be understood as follows. The implementation of hash-subset tests whether each of the keys in hash1 occurs in hash2 using their respective equalities. The implementation of hash-proper-subset applies hash-subset first, as above. If that is true, and the two hashes have the same number of elements, the result is falsified.
(hash-begin hash)
(hash-reset hash-iter hash)
(hash-next hash-iter)
(hash-peek hash-iter)
The hash-begin function returns a an iterator object capable of retrieving the entries in stored in hash one by one.
The hash-reset function changes the state of an existing iterator, such that it becomes prepared to retrieve the entries stored in the newly given hash, which may be the same one as the previously associated hash. In addition, hash-reset may be given a hash argument of nil, which dissociates it from its hash table.
The hash-next function's hash-iter argument is a hash iterator returned by hash-begin. If unvisited entries remain in hash, then hash-next returns the next one as a cons cell whose car holds the key and whose cdr holds the value. That entry is then considered visited by the iterator. If no more entries remain to be visited, hash-next returns nil. The hash-next function also returns nil if the iterator has been dissociated from a hash table by hash-reset.
The hash-peek function returns the same value that a subsequent call to hash-next will return for the same hash-iter, without changing the state of hash-iter. That is to say, if a cell representing a hash entry is returned, that entry remains unvisited by the iterator.
(with-hash-iter (isym hash-form [ksym [vsym]])
body-form*)
The with-hash-iter macro evaluates body-forms in an environment in which a lexically scoped function is visible.
The function is named by isym which must be a symbol suitable for naming functions with flet.
The hash-form argument must be a form which evaluates to a hash-table object.
Invocations of the function retrieve successive entries of the hash table as cons-cell pairs of keys and values. The function returns nil to indicate no more entries remain.
If either of the ksym or vsym arguments are present, they must be symbols suitable as variable names. They are bound as variables visible to body-forms, initialized to the value nil.
If ksym is specified, then whenever the function isym macro is invoked and retrieves a hash-table entry, the ksym variable is set to the key. If the function returns nil then the value of ksym is set to nil.
Similarly, if vsym is specified, then the function stores the retrieved hash value in that variable, or else sets the variable to nil if there is no next value.
(copy-hash-iter iter)
The copy-hash-iter function creates and returns a duplicate of the iter object, which must be a hash iterator returned by hash-begin.
The returned object has the same state as the original; it references the same traversal position in the same hash. However, it is independent of the original. Calls to hash-next on the original have no effect on the duplicate and vice versa.
The *hash-seed* special variable is initialized with a value of zero. Whenever a new hash table is explicitly or implicitly created, it takes its seed from the value of the *hash-seed* variable in the current dynamic environment.
The only situation in which *hash-seed* is not used when creating a new hash table is when make-hash is called with an argument given for the optional hash-seed argument.
Only equal-based hash tables make use of their seed, and only for keys which are strings and buffers. The purpose of the seed is to scramble the hashing function, to make a hash table resistant to a type of denial-of-service attack, whereby a malicious input causes a hash table to be populated with a large number of keys which all map to the same hash-table chain, causing the performance to severely degrade.
The value of *hash-seed* must be a nonnegative integer, no wider than 64 bits. On systems with 32-bit addresses, only the least significant 32 bits of this value may be significant.
(gen-hash-seed)
The gen-hash-seed function returns an integer value suitable for the *hash-seed* variable, or as the hash-seed argument of the make-hash and hash-equal functions.
The value is derived from the host environment, from information such as the process ID and time of day.
TXR Lisp provides binary search trees, which are objects of type tree. Trees have a printed notation denoted by the #T prefix. A tree may be constructed by invoking the tree function.
Binary search trees differ from hashes in that they maintain items in order. They also differ from hashes in that they store only elements, not key-value pairs. Every tree is associated with three key abstraction functions: It has a key function which is applied to the elements to map each one to a key. It also has a less function and equal function for comparing keys.
If these three functions are not specified, they respectively default to identity, less and equal, which means that the tree uses its elements as keys directly, and that they are compared using less and equal. Note: these default functions work for simple elements such as character strings or numbers, and also structures implementing equality substitution.
The elements are stored inside a tree using tree nodes, which are objects of type tnode, whose printed notation is introduced by the #N prefix.
Several tree-related functions take tnode objects as arguments or return tnode objects.
Trees may store duplicate elements. The #T literal syntax may freely specify duplicate elements. The tree constructor function specifies an initial sequence of elements to be populated into the newly constructed tree. If this initial sequence contains duplicate elements, they are preserved if the optional allow-dupes argument is true, otherwise only the rightmost member of any duplicate group appears in the tree.
The insertion functions tree-insert and tree-insert-node also overwrite duplicates by default, but optionally allow them. Duplicates are ordered by insertion: most recently inserted duplicate is rightmost. However, tree lookup chooses an unspecified duplicate.
(tnode key left right)
The tnode function allocates, initializes and returns a single tree node. A tree node has three fields key, left and right, which are accessed using the functions key, left and right.
(tnodep value)
The tnodep function returns t if value is a tree node. Otherwise, it returns nil.
(key node)
(left node)
(right node)
(set (key node) new-key)
(set (left node) new-left)
(set (right node) new-right)
The key, left and right functions retrieve the corresponding fields of the node object, which must be of type tnode.
Forms based on the key, left and right symbol are defined as syntactic places. Assigning a value v to (key n) using the set operator, as in (set (key n) v), is equivalent to (set-key n v) except that the value of the expression is v rather than n. Similar statements hold true for left and right in relation to set-left and set-right.
(set-key node new-key)
(set-left node new-left)
(set-right node new-right)
The set-key, set-left and set-right functions replace the corresponding fields of node with new values.
The node argument must be of type tnode.
These functions all return node.
(copy-tnode node)
The copy-tnode function creates a new tnode object, whose key, left and right fields are copied from node.
(tree [elems
[keyfun [lessfun [equalfun [allow-dupes]]]])
The tree function constructs and returns a new tree object. All arguments are optional.
The elems argument specifies a sequence of the elements to be stored in the tree. If the argument is absent or the sequence is empty, then an empty tree is created.
The keyfun argument specifies the function which is applied to every element to produce a key. If omitted, the tree object shall behave as if the identity function were used, taking the elements themselves to be keys.
The lessfun argument specifies the function by which two keys are compared for inequality. If omitted, the less function is used. A function used as lessfun should take two arguments, produce a Boolean result, and have ordering properties similar to the less function.
The equalfun argument specifies the function by which two keys are compared for equality. The default value is the equal function. A function used as equalfun should take two arguments, produce a Boolean result, and have the properties of an equivalence relation.
These three functions are collectively referred to as the tree's key abstraction functions.
The allow-dupes argument, which defaults to nil, is relevant if an elems sequence is specified containing some elements which which appear to be duplicates, according to the tree object's equalfun function. If allow-dupes is true then duplicates are preserved: the tree will have as many nodes as there are elements in the elems sequence. Moreover, the duplicates appear in the same relative order in the tree as they appear in the original elems sequence. If allow-dupes is false, then duplicates are suppressed: if any element appears more than once in elements, then only the last occurrence of that element appears in the tree.
Note: the tree-insert and tree-insert-node functions also has an optional argument indicating whether a duplicate insertion replaces an existing element.
Note: although the order of duplicate elements is preserved, when the tree-lookup function is used look up a key which is duplicated, the element which is retrieved is unspecified, and can change when the tree is reorganized due to insertions and deletions.
(treep value)
The treep function returns t if value is a tree. Otherwise, it returns nil.
(tree-count tree)
The tree-count function returns an integer indicating the number of nodes currently inserted into tree, which must be a search tree object.
(tree-insert-node tree node [allow-dupe])
The tree-insert-node function inserts an existing node object into a search tree.
The tree object must be of type tree, and node must be of type tnode.
The key field of the node object holds the element that is being inserted. The actual search key which is associated with this element is determined by applying tree's keyfun to the node's key value.
The node object must not currently be inserted into any existing tree. The values stored in the left and right fields of node are overwritten as required by the semantics of the insertion operation. Their original values are ignored.
The allow-dupe argument, defaulting to nil, is concerned with what happens if the tree already contains one or more nodes having a key equal to the node's key. If allow-dupe is false, then node replaces an unspecified one of those existing nodes: that replaced node is deleted from the tree. Key equivalence is determined using tree's equality function (see the equalfun argument of the tree function). If allow-dupe is true, then the new node is inserted without replacing any node, and appears together with the existing duplicate or duplicates. Among the duplicates, the newly inserted node is the rightmost node in the tree order.
The tree-insert-node function returns the node argument.
(tree-insert tree elem [allow-dupe])
The tree-insert function inserts elem into tree.
The tree argument must be an object of type tree.
The elem value may be of any type which is semantically compatible with tree's key abstraction functions.
The tree-insert function allocates a new tnode as if by invoking (tnode elem nil nil) function, and inserts that tnode as if by using the tree-insert-node function.
If one or more elements equal to elem already exist in the tree, then the behavior is determined by the allow-dupe argument, which defaults to nil. The semantics of allow-dupe is as given in the description of tree-insert-node.
The tree-insert function returns the newly inserted tnode object.
(tree-lookup-node tree key)
The tree-lookup-node searches tree for an element which matches key.
The tree argument must be an object of type tree.
The key argument may be a value of any type.
An element inside tree matches key if the tree's keyfun applied to that element produces a key value which is equal to key under the tree's equalfun function.
If such an element is found, then tree-lookup-node returns the tree node which contains that element as its key field.
If no such element is found, then tree-lookup-node returns nil.
If multiple nodes exist in the tree which have a matching key, it is unspecified which one of those nodes is retrieved.
(tree-lookup tree key)
The tree-lookup function finds an element inside tree which matches the given key.
If the element is found, it is returned. Otherwise, nil is returned.
Note: the semantics of the tree-lookup function can be understood in terms of tree-lookup-node. A possible implementation is this:
(defun tree-lookup (tree key)
(iflet ((node (tree-lookup-node tree key)))
(key node)))
If the tree contains multiple elements which match key, it is unspecified which element is retrieved.
(tree-delete-node tree key)
The tree-delete-node function searches tree for an element which matches key.
The tree argument must be an object of type tree.
The key argument may be a value of any type which is semantically compatible with tree's key abstraction functions.
If the matching element is found, then its node is removed from the tree, and returned.
Otherwise, if a matching element is not found, then nil is returned.
If more than one element exists inside tree which matches key, it is unspecified which node is deleted and returned.
(tree-delete tree key)
The tree-delete function tries to removes from tree the element which matches key.
If successful, it returns that element, otherwise it returns nil.
If more than one element exists inside tree which matches key, it is unspecified which one is deleted.
Note: the semantics of the tree-delete function can be understood in terms of tree-delete-node. A possible implementation is this:
(defun tree-delete (tree key)
(iflet ((node (tree-delete-node tree key)))
(key node)))
(tree-delete-specific-node tree node)
The tree-delete-specific-node function searches tree to find the specific node given by the node argument. If node is inserted into the tree, then it is deleted, and returned.
If node is not found in the tree, then the tree is unchanged, and nil is returned.
Note: the search for node is informed by node's key, for efficiency. However, if the tree contains duplicates of that key, then a linear search takes place among the duplicates.
(tree-min-node tree)
(tree-min tree)
The tree-min-node function returns the node in tree which holds the lowest element. If the tree is empty, it returns nil.
The tree-min function returns the lowest element, or else nil if the tree is empty.
(tree-del-min-node tree)
(tree-del-min tree)
The tree-del-min-node function returns the node in tree which has the lowest key, and removes that node from the tree. If the tree is empty, it returns nil.
The tree-del-min function returns the lowest element and removes it from the tree, or else nil if the tree is empty.
The following equivalence holds:
(tree-del-min tr) <--> (iflet ((node (tree-del-min-node tr)))
(key node))
Note: tree-insert together with tree-del-min provide the basis for using a tree as a priority queue. Elements are inserted into the queue using tree-insert and then removed in priority order using tree-del-min.
(tree-root tree)
The tree-root function returns the root node of tree, which must be a tree object.
If tree is empty, then nil is returned.
(tree-clear tree)
The tree-clear function deletes all elements from tree, which must be a tree object.
If tree is already empty, then the function returns nil, otherwise it returns an integer which gives the count of the number of deleted nodes.
(copy-search-tree tree)
The copy-search-tree returns a new tree object which is a copy of tree.
The tree argument must be an object of type tree.
The returned object has the same key abstraction functions as tree and contains the same elements.
The nodes held inside the new tree are freshly allocated, but their key objects are shared with the original tree.
(make-similar-tree tree)
The copy-search-tree returns a new, empty search tree object.
The tree argument must be an object of type tree.
The returned object has the same key abstraction functions as tree.
(tree-begin tree [low-key [high-key]])
The tree-begin function returns a new object of type tree-iter which provides in-order traversal of nodes stored in tree.
The tree argument must be an object of type tree.
If the low-key argument is specified, then nodes with keys lesser than low-key are omitted from the traversal.
If the high-key argument is specified, then nodes with keys equal to or greater than high-key are omitted from the traversal.
The nodes are traversed by applying the tree-next function to the returned tree-iter object.
A tree-iter object is iterable.
(collect-each ((el (tree-begin #T(() 1 2 3 4 5)
2 5)))
(* 10 el))
--> (20 30 40)
(tree-reset iter tree [low-key [high-key]])
The tree-reset functions is closely analogous to tree-begin.
The iter argument must be an existing tree-iter object, previously returned by a call to tree-begin.
Regardless of its current state, the iter object is re-initialized to traverse the specified tree with the specified parameters, and is then returned.
The tree-reset function prepares iter to traverse in the same manner as would new iterator returned by tree-begin for the specified tree, low-key and high-key arguments.
(tree-next iter)
(tree-peek iter)
The tree-next and tree-peek function returns the next node in sequence from the tree iterator iter. The iterator must be an object of type tree-iter, returned by the tree-begin function.
If there are no more nodes to be visited, these functions nil.
If, during the traversal of a tree, nodes are inserted or deleted, the behavior of tree-next and tree-peek on tree-iter objects that were obtained prior to the insertion or deletion is not specified. An attempt to complete the iteration may not successfully visit all keys that should be visited.
The tree-next function changes the state of the iterator. If tree-next is invoked repeatedly on the same iterator, it returns successive nodes of the tree.
If tree-peek is invoked more than once on the same iterator without any intervening calls to tree-next, it returns the same node; it does not appear to change the state of the iterator and therefore does not advance through successive nodes.
(sub-tree tree [from-key [to-key]])
The sub-tree function selects elements from tree, which must be a search tree.
If from-key is specified, then elements lesser than from-key are omitted from the selection.
If to-key is specified, the elements greater than or equal to to-key are omitted from the selection.
A list of the selected elements is returned, in which the elements appear in the same order as they do in tree.
(copy-tree-iter iter)
The copy-tree-iter function creates and returns a duplicate of the iter object, which must be a tree iterator returned by tree-begin.
The returned object has the same state as the original; it references the same traversal position in the same tree. However, it is independent of the original. Calls to tree-next on the original have no effect on the duplicate and vice versa.
(replace-tree-iter dest-iter src-iter)
The replace-tree-iter function causes the tree iterator dest-iter to be in the same state as src-iter.
Both dest-iter and src-iter must be tree iterator objects returned by tree-begin.
The contents of dest-iter are updated such that it now references the same tree as src-iter, at the same position.
The dest-iter argument is returned.
The *tree-fun-whitelist* variable holds a list of function names that may be used in the #T tree literal syntax as the keyfun, lessfun or equalfun operations of a tree. The initial value of this variable is a list which holds at least the following three symbols: identity, less and equal.
The application may change the value of this variable, or dynamically bind it, in order to allow #T literals to be processed which specify functions other than these three.
(op form+)
(do oper form*)
Like the lambda operator, the op macro denotes an anonymous function. Unlike lambda, the arguments of the function are implicit, or optionally specified within the expression, rather than as a formal parameter list which precedes a body.
The form arguments of op are implicitly turned into a DWIM expression, which means that argument evaluation follows Lisp-1 rules. (See the dwim operator).
The argument forms of op are arbitrary expressions, within which special conventions are permitted regarding the use of certain implicit variables:
Functions generated by op are always variadic; they always take additional arguments after any required ones, whether or not the @rest syntax is used.
If the body does not contain any @num or @rest syntax, then @rest is implicitly inserted. What this means is that, for example, since the form (op foo) does not contain any implicit positional arguments like @1, and does not contain @rest, it is actually a shorthand for (op foo . @rest): a function which applies foo to all of its arguments. If the body does contain at least one @num or @rest, then @rest isn't implicitly inserted. The notation (op foo @1) denotes a function which takes any number of arguments, and ignores all but the first one, which is passed to foo.
The do operator is similar to op, with the following three differences:
The actions of op and do can be understood by the following examples, which convey how the syntax is rewritten to lambda. However, note that the real translator uses generated symbols for the arguments, which are not equal to any symbols in the program.
(op) -> invalid
(op +) -> (lambda rest [+ . rest])
(op + foo) -> (lambda rest [+ foo . rest])
(op @1 . @rest) -> (lambda (arg1 . rest) [arg1 . @rest])
(op @1 @rest) -> (lambda (arg1 . rest) [arg1 @rest])
(op @1 @2) -> (lambda (arg1 arg2 . rest) [arg1 arg2])
(op foo @1 (@2) (bar @3)) -> (lambda (arg1 arg2 arg3 . rest)
[foo arg1 (arg2) (bar arg3)])
(op foo @rest @1) -> (lambda (arg1 . rest) [foo rest arg1])
(do + foo) -> (lambda (arg1 . rest) (+ foo arg1))
(do @1 @2) -> (lambda (arg1 arg2 . rest) (@1 arg2)) ;; invalid!
(do foo @rest @1) -> (lambda (arg1 . rest) (foo rest arg1))
Note that if argument @n appears in the syntax, it is not necessary for arguments @1 through @n-1 to appear. The function will have n arguments:
(op @3) -> (lambda (arg1 arg2 arg3 . rest) [arg3])
The op and do operators can be nested, in any combination. This raises the question: if an expression like @1, @rest or @rec occurs in an op that is nested within an op, what is the meaning?
An expression with a single @ always belongs with the innermost op or do operator. So for instance (op (op @1)) means that an (op @1) expression is nested within an outer op expression that contains no references to its implicit variables. The @1 belongs to the inner op.
There is a way for an inner op to refer to the implicit variables of an outer one. This is expressed by adding an extra @ prefix for every level of escape. For example in (op (op @@1)) the @@1 belongs to the outer op: it is the same as @1 appearing in the outer op. That is to say, in the expression (op @1 (op @@1)), the @1 and @@1 are the same thing: both are parameter 1 of the lambda function generated by the outer op. By contrast, in the expression (op @1 (op @1)) there are two different parameters: the first @1 is argument of the outer function, and the second @1 is the first argument of the inner function. If there are three levels of nesting, then three @ meta-prefixes are needed to insert a parameter from the outermost op into the innermost op.
Note that the implicit variables belonging to an op can be used in the dot position of a function call, such as:
[(op list 1 . @1) 2] -> (1 . 2)
This is a consequence of the special transformations described in the paragraph Dot Position in Function Calls in the subsection Additional Syntax of the TXR Lisp section.
The op syntax works in conjunction with quasiliterals which are nested within it. The metanumber notation as well as @rest are recognized without requiring an additional @ escape, which is effectively optional:
(apply (op list `@1-@rest`) '(1 2 3)) -> "1-2 3"
(apply (op list `@@1-@@rest`) '(1 2 3)) -> "1-2 3"
Though they produce the same result, the above two examples differ in that @rest embeds a metasymbol into the quasiliteral structure, whereas @@rest embeds the Lisp expression @rest into the quasiliteral. Either way, in the scope of op, @rest undergoes the macro-expansion which renames it to the machine-generated function argument symbol of the implicit function denoted by the op macro form.
This convenient omission of the @ character isn't supported for reaching the arguments of an outer op from a quasiliteral within a nested op:
;; To reach @@1, @@@1 must be written.
;; @@1 Lisp expression introduced by @.
(op ... (op ... `@@@1`))
Because the do macro may be applied to operators, it is possible to apply it to itself, as well as to op, as in the following example:
[[[[(do do do op list) 1] 2] 3] 4] -> (1 2 3 4)
The chained application associates right-to-left: the rightmost do is applied to op; the second rightmost do is applied to the rightmost one and so on. The effect is that partial application has been achieved. The value 1 is passed to the resulting function, which returns another function which takes the next argument. Finally, all these chained argument values are passed to list.
Each do/op level is processed independently. The following examples show how the list may be permuted into several different orders by referring to an implicit argument at various levels of nesting, making it the first argument of list. The unmentioned arguments implicitly follow, in order. This works because mentioning the argument explicitly means that its corresponding do operator no longer inserts its argument implicitly into body of the function which it generates:
[[[[(do do do op list @1) 1] 2] 3] 4] -> (4 1 2 3)
[[[[(do do do op list @@1) 1] 2] 3] 4] ->(3 1 2 4)
[[[[(do do do op list @@@1) 1] 2] 3] 4] -> (2 1 3 4)
[[[[(do do do op list @@@@1) 1] 2] 3] 4] -> (1 2 3 4))
The following example mentions all arguments at every do/op nesting level, thereby explicitly establishing the order in which they are passed to list:
[[[[(do do do op list @1 @@1 @@@1 @@@@1) 1] 2] 3] 4] -> (4 3 2 1)
(let ((c 0))
(mapcar (op cons (inc c)) '(a b c)))
--> ((1 . a) (2 . b) (3 . c))
(reduce-left (op + (* 10 @1) @2) '(1 2 3)) --> 123
(lop form+)
The lop macro is variant of op with special semantics.
The form arguments support the same notation as those of the op operator.
If only one form is given then lop is equivalent to op.
If two or more form arguments are present, then lop generates a variadic function which inserts all of its trailing arguments between the first and second forms.
That is to say, trailing arguments coming into the anonymous function become the left arguments of the function or function-like object denoted by the first form and the remaining forms give additional arguments. Hence the name lop, which stands for "left-inserting op".
This left insertion of the trailing arguments takes place regardless of whether @rest occurs in any form.
The form syntax determines the number of required arguments of the generated function, according to the highest-valued meta-number. The trailing arguments which are inserted into the left position are any arguments in excess of the required arguments.
The lop macro's expansion can be understood via the following equivalences, except that in the real implementation, the symbols rest and arg1 through arg3 are replaced with hygienic, unique symbols.
(lop f) <--> (op f) <--> (lambda (. rest) [f . rest])
(lop f x y) <--> (lambda (. rest)
[apply f (append rest [list x y])])
(lop f x @3 y) <--> (lambda (arg1 arg2 arg3 . rest)
[apply f
(append rest
[list x arg3 y])])
(mapcar (lop list 3) '(a b c)) --> ((a 3) (b 3) (c 3))
(mapcar (lop list @1) '(a b c)) --> ((a) (b) (c))
(mapcar (lop list @1) '(a b c) '(d e f))
--> ((d a) (e b) (f c))
(ldo oper form*)
The ldo macro provides a shorthand notation for uses of the do macro which inserts the first argument of the anonymous function as the leftmost argument of the specified operator.
The ldo syntax can be understood in terms of these equivalences:
(ldo f) <--> (do f @1)
(ldo f x) <--> (do f @1 x)
(ldo f x y) <--> (do f @1 x y)
(ldo f x @2 y) <--> (do f @1 x @2 y)
The implicit argument @1 is always inserted as the leftmost argument of the operator specified by the first form.
;; push elements of l1 onto l2.
(let ((l1 '(a b c)) l2)
(mapdo (ldo push l2) l1)
l2)
--> (c b a)
(ap form+)
(ip form+)
(ado form+)
(ido form+)
The ap macro is based on the op macro and has identical argument conventions.
The ap macro analyzes its arguments and produces a function f, in exactly the same same way as the op macro. However, instead of returning f, directly, it returns a different function g, which is a one-argument function that accepts a list. The list specifies arguments to which g applies f, and then returns the resulting value.
In other words, the following equivalence holds:
(ap form ...) <--> (apf (op form ...))
The ap macro nests properly with op and do, in any combination, in regard to the ...@@n notation.
The ip macro is similar to the ap macro, except that it is based on the semantics of the function iapply rather than apply, according to the following equivalence:
(ip form ...) <--> (ipf (op form ...))
The ado and ido macros are related to do macro in the same way that ap and ip are related to op. They produce a one-argument function which works as if by applying the function generated by do to its its own arguments, according to the following equivalence:
(ado form ...) <--> (apf (do form ...))
(ido form ...) <--> (ipf (do form ...))
See also: the apf and ipf functions.
;; Take a list of pairs and produce a list in which those pairs
;; are reversed.
(mapcar (ap list @2 @1) '((1 2) (a b))) -> ((2 1) (b a))
(opip clause*)
(oand clause*)
(lopip clause*)
(loand clause*)
The opip and oand operators make it possible to chain together functions which are expressed using the op syntax. (See the op operator for more information).
Both macros perform the same transformation except that opip translates its arguments to a call to the chain function, whereas oand translates its arguments in the same way to a call to the chand function.
More precisely, these macros perform the following rewrites:
(opip arg1 arg2 ... argn) -> [chain {arg1} {arg2} ... {argn}]
(oand arg1 arg2 ... argn) -> [chand {arg1} {arg2} ... {argn}]
where the above {arg} notation denotes the following transformation applied to each argument:
;; these specific form patterns are left untransformed:
(dwim ...) -> (dwim ...)
[...] -> [...]
(qref ...) -> (qref ...)
(uref ...) -> (uref ...)
(op ...) -> (op ...)
(do ...) -> (do ...)
(lop ...) -> (lop ...)
(ldo ...) -> (ldo ...)
(ap ...) -> (ap ...)
(ip ...) -> (ip ...)
(ado ...) -> (ado ...)
(ido ...) -> (ido ...)
(ret ...) -> (ret ...)
(aret ...) -> (aret ...)
.slot -> .slot
.(method ...) -> .(method ...)
atom -> atom
;; forms headed by let are treated specially
(let sym) -> ;; described below
(let (s0 i0)
(s1 i1)
....) -> ;; described below
(let ((s0 i0)
(s1 i1)) -> ;; described below
body)
;; other compound forms are transformed like this:
(function ...) -> (op function ...)
(operator ...) -> (do operator ...)
(macro ...) -> (do macro ...)
In other words, compound forms whose leftmost symbol is a macro or operator are translated to the do notation. Compound forms denoting function calls are translated to the op notation. Compound forms which are dwim invocations, either explicit or via the DWIM brackets notation, are used without transformation. Used without transformation also are forms denoting struct slot access, either explicitly using uref or qref or the respective dot notations, forms which invoke any of the do family of operators, as well as any atom forms.
The lopip and loand operators are similar to, respectively, opip and oand, except that they insert the implicit argument as the leftmost argument. For these macros, the above specification of what transformations are applied to the arguments is modified as follows:
;; other compound forms are transformed like this:
(function ...) -> (lop function ...)
(operator ...) -> (ldo operator ...)
(macro ...) -> (ldo macro ...)
When a let or let* expression occurs in opip syntax, it denotes a special syntax which is treated as follows.
Note: an opip form with no arguments specifies a function which returns nil, which follows from a documented property of the chain function.
Take each element from the list (1 2 3 4) and multiply it by three, then add 1. If the result is odd, collect that into the resulting list:
(mappend (opip (* 3)
(+ 1)
[iff oddp list])
(range 1 4))
The above is equivalent to:
(mappend (chain (op * 3)
(op + 1)
[iff oddp list])
(range 1 4))
The (* 3) and (+ 1) terms are rewritten to (op * 3) and (op + 1), respectively, whereas [iff oddp list] is passed through untransformed.
The following demonstrates the single variable let:
(let ((pipe (opip (+ 1) (let x)
(+ 2) (let y)
(+ 3)
(list x y))))
[pipe 1])
-> (2 4 7)
The x variable intercepts the value coming from (+ 1) and binds x to that value. When the opip function is invoked with the argument 1, that value is 2. That value also continues to the (+ 2) element which yields 4, which is similarly captured by variable y. The final list expression lists the values of x and y, as well as, implicitly, the value @1 coming from the previous element,
(opf function clause*)
(lopf function clause*)
The opf and lopf function make available the opip-style functional arguments in conjunction with an arbitrary function.
The clause arguments of opf and lopf are processed exactly like those of opip and lopip.
The syntax
(opf f c1 c2 c3 ...)
is converted into a function call of the form:
[f {c1} {c2} {c3} ...]
where every argument
{cN}
is converted to a form denoting a function, in exactly the same manner
as the arguments of
opip.
The same remarks apply to code
lopf
in relation to
lopip.
Thus, it is possible to express opip using opf by choosing chain as the function argument, according to this equivalence:
(opip c1 c2 c3 ...) <--> (opf chain c1 c2 c3 ...)
;; Remove values greater than 10 or less than five from list
(remove-if (lopf orf (> 10) (< 5)) (range 0 20)) -> (5 6 7 8 9 10))
;; Note: could be expressed as
(remove-if (orf (lop > 10) (lop < 5)) (range 0 20))
As the example shows, the opf and lopf macros provide a way to avoid repeating the op and lop syntax in every argument of a functional combinator of a function.
(flow form opip-arg*)
(lflow form lopip-arg*)
The flow macro passes the value of form through the processing stages described by the opip-arg arguments, yielding the resulting value.
The opip-arg arguments follow the semantics of the opip macro.
The same requirements apply to lflow, except that it is related to the lopip macro which inserts the implicit argument into the leftmost position.
The following equivalences hold:
(flow x ...) <--> [(opip ...) x]
(lflow x ...) <--> [(lopip ...) x]
That is to say, flow is equivalent to the application of an opip-generated function to the value of form, and likewise lflow is equivalent to the application of a lopip-generated function.
Note: if there are no opip-arg or lopip-arg arguments, then flow evaluates the x argument and returns nil; which follows from the behavior of the opip and lopip macros, when those are invoked with no arguments.
(flow 1 (+ 2) (* 3) (cons 0)) -> (0 . 9)
(flow "abc" (upcase-str) (regsub #/B/ "ZTE")) -> "AZTEC"
(flow 1 (- 10)) -> 9
(lflow 10 (- 1)) -> 9
(ret form)
The ret macro's form argument is treated similarly to the second and subsequent arguments of the op operator.
The ret macro produces a function which takes any number of arguments, and returns the value specified by form.
form can contain op meta syntax like @n and @rest.
The following equivalence holds:
(ret x) <--> (op identity* x))
Thus the expression (ret @2) returns a function similar to (lambda (x y . z) y), and the expression (ret 42) returns a function similar to (lambda (. rest) 42).
(aret form)
The aret macro's form argument is treated similarly to the second and subsequent arguments of the op operator.
The aret macro produces a function which takes any number of arguments, and returns the value specified by form.
form can contain ap meta syntax like @n and @rest.
The following equivalence holds:
(aret x) <--> (ap identity* x))
Thus the expression (aret @2) returns a function similar to (lambda (. rest) (second rest)), and the expression (aret 42) returns a function similar to (lambda (. rest) 42).
(tap arg+)
The tap macro is intended for use in conjunction with opip, flow and other macros in that family. It is a short-hand for writing a pipeline element which performs a side-effecting operation, but unconditionally returns the original input value.
The exact expansion of tap is unspecified, but the following equivalence indicates a possible expansion strategy:
(tap ...) <--> (prog1 @1 (...))
Assuming that expansion strategy, the expression (tap put-line `foo: @1`) would expand to (prog1 @1 (put-line `foo: @1`)).
Note: tap, in addition to being useful for inserting necessary side effects into pipelines, is also useful for inserting temporary debug print forms. For that purpose, inserting the `prinl` function is often enough:
Here, the pipeline will calculate (* 4 (+ 2 10)) with the side effect of the value of (+ 2 10) being printed. With tap, the output can be customized, allowing multiple output points to be distinguished.
(flow 10
(tap put-line `input: @1`)
(+ 2)
(tap put-line `+ 2: @1`)
(* 4)
(tap put-line `* 4: @1`))
-> 48
Output produced:
input: 10
+ 2: 12
* 4: 48
(dup func)
The dup function returns a one-argument function which calls the two-argument function func by duplicating its argument.
;; square the elements of a list
(mapcar [dup *] '(1 2 3)) -> (1 4 9)
(flipargs func)
The flipargs function returns a two-argument function which calls the two-argument function func with reversed arguments.
(chain func*)
(chand func*)
The chain function accepts zero or more functions as arguments, and returns a single function, called the chained function, which represents the chained application of those functions, in left-to-right order.
If chain is given no arguments, then it returns a variadic function which ignores all of its arguments and returns nil.
Otherwise, the first function may accept any number of arguments. The second and subsequent functions, if any, must accept one argument.
The chained function can be called with an argument list which is acceptable to the first function. Those arguments are in fact passed to the first function. The return value of that call is then passed to the second function, and the return value of that call is passed to the third function and so on. The final return value is returned to the caller.
The chand function is similar, except that it combines the functionality of andf into chaining. The difference between chain and chand is that chand immediately terminates and returns nil whenever any of the functions returns nil, without calling the remaining functions.
(call [chain + (op * 2)] 3 4) -> 14
In this example, a two-element chain is formed from the + function and the function produced by (op * 2) which is a one-argument function that returns the value of its argument multiplied by two. (See the definition of the op operator).
The chained function is invoked using the call function, with the arguments 3 and 4. The chained evaluation begins by passing 3 and 4 to +, which yields 7. This 7 is then passed to the (op * 2) doubling function, resulting in 14.
A way to write the above example without the use of the DWIM brackets and the op operator is this:
(call (chain (fun +) (lambda (x) (* 2 x))) 3 4)
(juxt func*)
The juxt function accepts a variable number of arguments which are functions. It combines these into a single function which, when invoked, passes its arguments to each of these functions, and collects the results into a list.
Note: the juxt function can be understood in terms of the following reference implementation:
(defun juxt (funcs)
(lambda (. args)
(mapcar (lambda (fun)
(apply fun args))
funcs)))
The callf function generalizes juxt by allowing the combining function to be specified.
;; separate list (1 2 3 4 5 6) into lists of evens and odds,
;; which end up juxtaposed in the output list:
[(op [juxt keep-if remove-if] evenp)
'(1 2 3 4 5 6)] -> ((2 4 6) (1 3 5))
;; call several functions on 1, collecting their results:
[[juxt (op + 1) (op - 1) evenp sin cos] 1]'
-> (2 0 nil 0.841470984807897 0.54030230586814)
(andf func*)
(orf func*)
The andf and orf functions are the functional equivalent of the and and or operators. These functions accept multiple functions and return a new function which represents the logical combination of those functions.
The input functions should have the same arity. Failing that, there should exist some common argument arity with which each of these can be invoked. The resulting combined function is then callable with that many arguments.
The andf function returns a function which combines the input functions with a short-circuiting logical conjunction. The resulting function passes its arguments to the input functions successively, in left-to-right order. As soon as any of the functions returns nil, then nil is returned and the remaining functions are not called. If none of the functions return nil, then the value returned by the last function is returned. If the list of functions is empty, then t is returned. That is, (andf) returns a function which accepts any arguments and returns t.
The orf function returns a function which combines the input functions with a short-circuiting logical disjunction. The resulting function passes its arguments to the input functions successively, in left-to-right order. As soon as any of the functions returns a non-nil value, that value is returned and the remaining functions are not called. If all of the functions return nil, then nil is returned. If the list of functions is empty, then nil is returned. That is, (orf) returns a function which accepts any arguments and returns nil.
(notf function)
The notf function returns a function which is the Boolean negation of function.
The returned function takes a variable number of arguments. When invoked, it passes all of these arguments to function and then inverts the result as if by application of not.
(nandf func*)
(norf func*)
The nandf and norf functions are the logical negation of the andf and orf functions. They are related according to the following equivalences:
[nandf f0 f1 f2 ...] <--> (notf [andf f0 f1 f2 ...])
[norf f0 f1 f2 ...] <--> (notf [orf f0 f1 f2 ...])
(iff condfun [thenfun [elsefun]])
(iffi condfun thenfun [elsefun])
The iff function is the functional equivalent of the if operator. It accepts functional arguments and returns a function.
The resulting function passes its arguments to condfun. If condfun yields true, then the arguments are passed to thenfun and the resulting value is returned. Otherwise the arguments are passed to elsefun and the resulting value is returned.
If thenfun is omitted then identity is used as default. This omission is not permitted by iffi, only iff.
If elsefun needs to be called, but is omitted, then nil is returned.
The iffi function differs from iff only in the defaulting behavior with respect to the elsefun argument. If elsefun is omitted in a call to iffi then the default function is identity. This is useful in situations when one value is to be replaced with another one when the condition is true, otherwise preserved.
The following equivalences hold between iffi and iff:
(iffi a b c) <--> (iff a b c)
(iffi a b) <--> (iff a b identity)
[iffi a b nilf] <--> [iff a b]
[iffi a identity nilf] <--> [iff a]
The following equivalences illustrate iff with both optional arguments omitted:
[iff a] <---> [iff a identity nilf] <---> a
(tf arg*)
(nilf arg*)
(ignore arg*)
The tf and nilf functions take zero or more arguments, and ignore them. The tf function returns t, and the nilf function returns nil.
The ignore function is a synonym of nilf.
Note: the following equivalences hold between these functions and the ret operator, and retf function.
(fun tf) <--> (ret t) <--> (retf t)
(fun nilf) <--> (ret nil) <--> (ret) <--> (retf nil)
In Lisp-1-style code, tf and nilf behave like constants which can replace uses of (ret t) and (ret nil):
[mapcar (ret nil) list] <--> [mapcar nilf list]
Note: the ignore function can be used for suppressing unused variable warnings.
;; tf and nilf are useful when functions are chained together.
;; test whether (trunc n 2) is odd.
(defun trunc-n-2-odd (n)
[[chain (op trunc @1 2) [iff oddp tf nilf]] n])
In this example, two functions are chained together, and n is passed through the chain such that it is first divided by two via the function denoted by (op trunc @1 2) and then the result is passed into the function denoted by [iff oddp tf nilf]. The iff function passes its argument into oddp, and if oddp yields true, it passes the same argument to tf. Here tf proves its utility by ignoring that value and returning t. If the argument (the divided value) passed into iff is even, then iff passes it into the nilf function, which ignores the value and returns nil.
The following example shows how ignore may be used to suppress compiler warnings about unused parameters or other variables:
(retf value)
The retf function returns a function. That function can take zero or more arguments. When called, it ignores its arguments and returns value.
See also: the ret macro.
;; the function returned by (retf 42)
;; ignores 1 2 3 and returns 42.
(call (retf 42) 1 2 3) -> 42
(apf function arg*)
(ipf function arg*)
The apf function returns a one-argument function whose argument conventions are similar to those of the apply function: it accepts one or more arguments, the last of which should be a list. When that function is called, it applies function to these arguments to as if by apply. It then returns whatever function returns.
If one or more additional args are passed to apf, then these are stored in the function which is returned. When that function is invoked, it prepends all of the stored arguments to the passed arguments, and applies the function. to the resulting combined argument list. Thus the args become the leftmost arguments of function.
The ipf function is similar to apf, except that the argument conventions and application semantics of the function returned by ipf are based on iapply rather than apply.
See also: the ap macro.
;; Function returned by [apf +] accepts the
;; (1 2 3) list and applies + to it, as
;; if (+ 1 2 3) were called.
(call [apf +] '(1 2 3)) -> 6
(callf main-function arg-function*)
The callf function returns a function which applies each arg-function to its arguments, juxtaposing the return values of these calls to form arguments to which main-function is then applied. The return value of main-function is returned.
The following equivalence holds, except for the order of evaluation of arguments:
(callf fm f0 f1 f2 ...) <--> (chain (juxt f0 f1 f2 ...)
(apf fm))
;; Keep those pairs which are two of a kind
(keep-if [callf eql first second] '((1 1) (2 3) (4 4) (5 6)))
-> ((1 1) (4 4))
The following equivalence holds between juxt and callf:
[juxt f0 f1 f2 ...] <--> [callf list f0 f1 f2 ...]:w
Thus, juxt may be regarded as a specialization of callf in which the main function is implicitly list.
(mapf main-function arg-function*)
The mapf function returns a function which distributes its arguments into the arg-functions. That is to say, each successive argument of the returned function is associated with a successive arg-function.
Each arg-function is called, passed the corresponding argument. The return values of these functions are then passed as arguments to main-function and the resulting value is returned.
If the returned function is called with fewer arguments than there are arg-functions, then only that many functions are used. Conversely, if the function is called with more arguments than there are arg-functions, then those arguments are ignored.
The following equivalence holds:
(mapf fm f0 f1 ...) <--> (lambda (. rest)
[apply fm [mapcar call
(list f0 f1 ...)
rest]])
;; Add the squares of 2 and 3
[[mapf + [dup *] [dup *]] 2 3] -> 13
TXR Lisp supports input and output streams of various kinds, with generic operations that work across the stream types.
In general, I/O errors are usually turned into exceptions. When the description of error reporting is omitted from the description of a function, it can be assumed that it throws an error.
These variables hold predefined stream objects. The *stdin*, *stdout* and *stderr* streams closely correspond to the underlying operating system streams. Various I/O functions require stream objects as arguments.
The *stddebug* stream goes to the same destination as *stdout*, but is a separate object which can be redirected independently, allowing debugging output to be separated from normal output.
The *stdnull* stream is a special kind of stream called a null stream. To read operations, the stream appears empty, like a stream open on an empty file. To write operations, it appears as a data sink of infinite capacity which consumes data and discards it. This stream is similar to the /dev/null device on Unix, and in fact has a relationship to it. If an attempt is made to obtain the underlying file descriptor of *stdnull* using the fileno function, then the /dev/null device is open, if the host platform supports it. The resulting file descriptor number is returned, and also retained in the *stdnull* device. When close-stream is invoked on *stdnull*, that descriptor is closed. This feature of *stdnull* allows it to be useful for establishing redirections around the execution of external utilities.
;; redirect output of ls *.txt command to /dev/null
(let ((*stderr* *stdnull*))
(sh "ls *.txt"))
The *print-flo-format* variable determines the conversion format which is applied when a floating-point value is converted to decimal text by the functions print, prinl, and tostring.
The default value is ~s.
The related variable *pprint-flo-format* similarly determines the conversion format applied to floating-point values by the functions pprint, pprinl, and tostringp.
The default value is ~a.
The format string in either variable must specify the consumption of exactly one format argument.
The conversion string may use embedded width and precision values: for instance, ~3,4f is a valid value for *print-flo-format* or *pprint-flo-format*.
The *print-flo-precision* special variable specifies the default floating-point printing precision which is used when the ~a or ~s conversion specifier of the format function is used for printing a floating-point value, and no precision is specified.
Note that since the default value of the variable *print-flo-format* is the string ~s, the *printf-flo-precision* variable, by default, also determines the precision which applies when floating-point values are converted to decimal text by the functions print, pprint, prinl, pprinl, tostring and tostringp.
The default value of *print-flo-precision* is that of the flo-dig variable.
Note: to print floating-point values in such a way that their values can be precisely recovered from the printed representation, it is recommended to override *print-flo-precision* to the value of the flo-max-dig variable.
The *print-flo-precision* special variable specifies the default floating-point printing precision which is used when the ~f or ~e conversion specifier of the format function is used for printing a floating-point value, and no precision is specified.
Its default value is 3.
The *print-base* variable controls the base (radix) used for printing integer values. It applies when the functions print, pprint, prinl, pprinl, tostring and tostringp process an integer value. It also applies when the ~a and ~s conversion specifiers of the format function are used for printing an integer value.
The default value of the variable is 10.
Meaningful values are: 2, 8, 10 and 16.
When base 16 is selected, hexadecimal digits are printed as uppercase characters.
The *print-circle* variable is a Boolean which controls whether the circle notation is in effect for printing aggregate objects: conses, ranges, vectors, hash tables and structs. The initial value of this variable is nil: circle notation printing is disabled.
The circle notation works for structs also, including structs which have user-defined print methods. When a print method calls functions which print objects, such as print, pprinl or format on the same stream, the detection of circularity and substructure sharing continues in these recursive invocations.
However, there are limitations in the degree of support for circle notation printing across print methods. Namely, a print method of a struct S must not procure and submit for printing objects which are not part of the ordinary structure that is reachable from the (static or instance) slots of S, if those objects have already been printed prior to invoking the print method, and have been printed without a #= circle notation label. The "ordinary structure that is reachable from the slots" denotes structure that is directly reachable by traversing conses, ranges, vectors, hashes and struct slots: all printable aggregate objects.
The *read-unknown-structs* variable controls the behavior of the parser upon encountering structure literal #S syntax which specifies an unknown structure type.
If this variable's value is nil then such a literal is erroneous; an exception is thrown. Otherwise, such a structure is converted not into a structure object, which is impossible, but into a list object whose first element is the symbol sys:struct-lit. The remaining elements are taken from the #S syntax.
(format stream-designator format-string format-arg*)
The format function performs output to a stream given by stream-designator, by interpreting the actions implicit in a format-string, incorporating material pulled from additional arguments given by format-arg*. Though the function is simple to invoke, there is complexity in format string language, which is documented below.
The stream-designator argument can be a stream object, or one of the values t or nil. The value t serves as a shorthand for *stdout*. The value nil means that the function will send output into a newly instantiated string output stream, and then return the resulting string.
Within format-string, most characters represent themselves. Those characters are simply output. The character ~ (tilde) introduces formatting directives, which are denoted by a single character, usually a letter.
The special sequence ~~ (tilde-tilde) encodes a single tilde. Nothing is permitted between the two tildes.
The syntax of a directive is generally as follows:
~[width] [,precision] letter
In other words, the ~ (tilde) character, followed by a width specifier, a precision specifier introduced by a comma, and a letter, such that width and precision are independently optional: either or both may be omitted. No whitespace is allowed between these elements.
The letter is a single alphabetic character which determines the general action of the directive. The optional width and precision are specified as follows:
If the leading < character is present, then the printing will be left-adjusted within this field. If the ^ character is present, the printing will be centered within the field. Otherwise it will be right-adjusted by default.
The width can be specified as a decimal integer with an optional leading minus sign, or as the character *. The * notation means that instead of digits, the value of the next argument is consumed, and expected to be an integer which specifies the width. If the width, specified either way, is negative, then the field will be left-adjusted. If the value is positive, but either the < or ^ prefix character is present in the width specifier, then the field is adjusted according to that character.
The padding calculations for alignment and centering take into account character display width, as defined by the display-width function. For instance, a character string containing four Chinese characters (kanji) has a display width of 8, not 4.
The width specification does not restrict the printed portion of a datum. Rather, for some kinds of conversions, it is the precision specification that performs such truncation. A datum's display width (or that of its printed portion, after such truncation is applied) can equal or exceed the specified field width. In this situation it overflows the field: the printed portion is rendered in its entirety without any padding applied on either side for alignment or centering.
The precision specifier may begin with these optional characters, whose effect
The precision value influences the printing of values of all types. The precision options apply only when the value being printed is a number; otherwise they are ignored.
If the +, - or space are multiply specified, the rightmost one takes precedence.
The precision specifier itself follows: it must be either a decimal integer or the * character indicating that the precision value comes from an integer argument.
The leading zero option is only active if accompanied by a precision value, either coming from additional digits in the formatting directive, or from an argument indicated by *. If no precision specifier is present, then the leading zero option is interpreted as a specifier indicating a precision value of zero, rather than requesting leading zeros. To request zero padding together with zero precision, either two or more zero digits are required, or else the leading zero indicator must be given together with the * specifier.
For non-numeric values, the precision specifies the maximum number of print positions to occupy, taking into account the display width of each character of the printed representation of the object, as according to the display-width function. The object's printed representation is truncated, if necessary, to the maximum number of characters which will not exceed the specified number of print positions.
A numeric argument is formatted into the field in two distinct steps, both of which involve the precision value in a different role. The details of the first of these steps, and the role payed by precision, depend on which conversion directive is used, as well as whether the argument is integer or floating-point. That first step prepares the printed representation of a number which is then fitted into the field by the second step, and also calculates the effective precision value, which is based on the original width and precision. The second step works with the effective precision rather than the original precision. Its description follows.
First, the length of the printed representation of the number, not including its sign, is calculated. If this part of the number is shorter than the effective precision, then it is padded on the left with spaces or leading zeros so that the resulting string is equal to the precision.
Next, if the number is negative, or else if adding a positive sign has been requested, then the sign is added. It is added to the left of the padding zeros, or else to the right of padding spaces, whichever the case may be.
At this stage, if the number is not yet adorned with a sign, and either the - or space precision option had been given, then the appropriate character, the digit 0 or a space, is added in the place where the sign would go. This is done only if the result will not overflow the field width, but without regard for whether the character will overflow the effective precision.
Finally, the resulting number is rendered into the field, using the requested left, right or center adjustment, as if it were a character string. If it overflows the field, it is reproduced in its entirety without any adjustment being performed.
When the a specifier is used for numbers, the formatting is performed in two distinct steps: the printed representation of the number is calculated first, and then that representation is set into the field. At the same time, an effective precision is calculated, based on the precision and width, and that effective precision is used in the second step.
In the first step, the rendering of a floating-point number to its printed representation, the precision specifies the maximum number of total significant figures, which do not include any digits in the exponent, if one is printed. Numbers are printed in E notation if their magnitude is small, or else if their exponent exceeds their precision. If the precision is not specified, then it is obtained from the *print-flo-precision* special variable, whose default value is the same as that of the flo-dig variable. The effective precision for the second step is then taken from the original precision, or one less than the width, whichever of the these two values is smaller, but no lower than zero. If the width is unspecified, it is taken as zero.
Floating point values which are integers are printed without a trailing .0 (point zero). The + flag in the precision is honored for rendering an explicit + sign on nonnegative values. If a leading zero is specified in the precision, and a nonzero width is specified, then the printed value's integer part will be padded with leading zeros up to one less than the field width. These zeros are placed before the sign. A precision value of zero imposed on floating-point values is equivalent to a value of one; it is not possible to request zero significant figures.
Integers are not affected by the precision value in the conversion to text; all of the digits of the integer are taken into the second step. In the case of integers, The effective precision for the second step is then taken from the original precision, or one less than the width, whichever of the these two values is smaller. However, if the width is not specified, or given as zero, then the unmodified precision value is taken as the effective precision. Thus, in the zero width or missing width case, integers are always padded with spaces or leading zeros due to the precision value, even if such padding overflows the field width.
Rationale: the purpose of the elaborate rules for calculating the effective precision is to both obtain consistency in the printing of integers and floating-point values that are integers, as well as to break that consistency when the width is omitted or zero. This break in consistency has two benefits. The common situation of adding leading spaces or zeros to integers can be specified without specifying the width. For instance "~,8a" will format an integer right-justified in an eight-character extent, without width having to be used in order to specify a field to accommodate that padding. The effective padding amount going into the second step is 8, exceeding the zero width, and thus allowing the padding to overflow the field. In the case of floating-point, precision alone can express the common requirement for limiting the number of digits can be expressed by the precision, without causing unwanted padding when there are fewer digits. If the above "~,8a" is used to format a floating-point value, it will be limited to 8 digits of precision, regardless of its magnitude and the position of its decimal point, or whether or not exponential notation is used. The effective precision for field placement shall then be zero in the second step, so that no padding is generated. However, if a nonzero width is used, then formatting becomes consistent between floating-point and integer so that, for instance, the format directive "~8,8a" produces the same output for the argument values 42 and 42.0, namely an eight-character-wide field in which the digits "42" appear right-aligned.
The formatting performed by f is performed in two distinct steps: the printed representation of the number is calculated first, and then that representation is set into the field. The precision parameter coming from the directive is only involved in the first step.
In the first step, the precision specifier gives the number of digits past the decimal point. The number is rounded off to the specified precision, if necessary. Furthermore, that many digits are always printed, regardless of the actual precision of the number or its type. If it is omitted, then the value is obtained from the special variable *print-flo-digits*, whose default value is three: three digits past the decimal point. A precision of zero means no digits past the decimal point, and in this case the decimal point is suppressed (regardless of whether the numeric argument is floating-point or integer).
No limit is placed on the number of significant figures in the number by either the precision or width value.
When the resulting textual number passes to the second formatting step, the precision value, for the purposes of that step, is calculated by taking one less than the field width, or else zero if the field width is zero. This value is not related to the precision that had been used to determine the number of places past the decimal point.
The formatting performed by e is performed in two distinct steps: the printed representation of the number is calculated first, and then that representation is set into the field. The precision parameter coming from the directive is only involved in the first step.
In the first step, the precision specifier gives the number of digits past the decimal point printed in the E notation, not counting the digits in the exponent. Exactly that many digits are printed, regardless of the precision of the number. If the precision is omitted, then the number of digits after the decimal point is obtained from the value of the special variable *print-flo-digits*, whose default value is three. If the precision is zero, then a decimal portion is truncated off entirely, including the decimal point.
When the resulting textual number passes to the second formatting step, the precision value, for the purposes of that step, is calculated by taking one less than the field width, or else zero if the field width is zero. This value is not related to the precision that had been used to determine the number of places past the decimal point.
The indentation mode and indentation column are automatically restored to their previous values when format function terminates, naturally or via an exception or nonlocal jump.
The effect of a precision field (even if zero) combined with the ! directive is currently not specified, and reserved for future extension. The precision field is processed syntactically, and no error occurs, however.
(fmt format-string format-arg*)
The fmt function provides a shorthand for formatting to a string, according to the following equivalence which holds between fmt and format:
(fmt s arg ...) <--> (format nil s arg ...)
(pic format-string format-arg*)
The pic macro ("picture based formatting") provides a notation for constructing a character string under the control of format-string which indicates the insertion of zero or more format-arg argument values.
Like the fmt function or quasiliteral syntax, the pic macro returns a character string.
The pic macro's format-string notation is different from quasiliterals or from fmt.
The pic format-string argument isn't an evaluated expression, but syntax. It must be either a string literal or else a string quasiliteral. No other syntax is permitted.
If pic is a string, is scanned left to right in search of pic patterns. Any characters not belonging to a pic pattern are copied into the output string verbatim. When a pic pattern is found, it is removed from format-string and applied to the next successive format-arg to perform a conversion and formatting of that value to text. The resulting text is appended to the output string, and the process continues in search of the next pic pattern. When the format-string is exhausted, the constructed string is returned.
If format-string is a quasiliteral, then all of the text strings embedded within the quasiliteral are examined in the same way, in left to right order. Each such string is transformed into an expression which produces a character string according to the semantics of the pic patterns it contains, and the resulting expressions are substituted into the original quasiliteral to produce a transformed quasiliteral.
There must be exactly as many format-arg arguments as there are pic patterns in format-string.
The pic macro arranges for the left-to-right evaluation of the format-arg expressions. If format-string is a quasiliteral, the evaluation of these expressions is interleaved into the quasiliterals expressions and variables, in the order implied by the placement of the corresponding pic patterns relative to the quasiliteral elements. For instance, if format-string is `@(abc)<<<@(xyz)` then the function abc is called first, then the format-argument is evaluated which produces a value for the <<< pic pattern, after which the xyz function is called.
There are two kinds of pic patterns: alignment patterns, numeric patterns and escape patterns.
Escape patterns consist of a two-character sequence introduced by the ~ (tilde) character, which is followed by one of the characters that are special in pic pattern syntax:
< > | + - 0 # . ! ~ , ( )
An escape pattern produces the second character as its output. For instance ~~ encoded a single ~ character, and ~# encodes a literal # character that is not part of any pattern.
Alignment patterns are described next.
[sign] [0] {#}+ [point {#}+ | !]
or else if they contain commas, the placement of these commas is governed by the more complicated rule:
[sign] [0 [,]] {#}+ {,{#}+}* [point {#}+ {,{#}+}* | !]
Commas may be placed anywhere within the pattern of hash characters, except at the beginning or end, or adjacent to the decimal point. If the leading zero is present, a comma may appear immediately after it, before the first hash.
A second form of both of the above patterns is supported, for specifying that negative numbers be shown in parentheses. Instead of the sign, an opening parenthesis may appear, which must be matched by a closing parenthesis which follows a valid pattern interior:
( [0] {#}+ [point {#}+ | !] )
With embedded commas:
( [0 [,]] {#}+ {,{#}+}* [point {#}+ {,{#}+}* | !] )
The pattern consists of an optional sign which is one of the characters + (plus) or - (minus), or else it may optionally begin with an opening parenthesis, indicating one of the two alternative forms.
This is followed by an optional leading zero. After this comes a sequence of one or more # (hash) characters, which may contain exactly one point element, which is defined as one of the characters . (period) or ! (exclamation mark). This point element may appear at most once, and must not be the first or last character, unless it is the exclamation mark, in which case it may appear last.
Except if ending in the exclamation mark, a numeric pattern specifies a field width which is equal to the number of characters occurring in the pattern itself. For instance, the patterns ####, +### and 0#.# all specify a field width of four. If the numeric pattern ends in an exclamation mark, that character is not counted toward the field width that it specifies. Thus the pattern ###! specifies a field width of three.
If the leading sign is present, it has the following meanings:
If the leading zero is present, it specifies that the number is padded with zeros on the left. In combination with the - sign, this shall not cause the leading space before a positive value to be overwritten with a zero; leading zeros, if any, begin after that space.
The remainder of the pattern specifies the number of digits of the fractional part which is indicated by number of # characters after the point. The number is rounded to that many fractional digits, which are all rendered, even if there are trailing zeros. If no point is not specified, then the number of fractional digits is zero. The same is true if point is specified as ! as the last character. In both cases, the numeric argument is rounded to integer, and rendered without any decimal point or fractional part.
There is a difference between point being specified using the ordinary decimal point character . versus the ! character. The ! character specifies that if the conversion of the numeric argument overflows the field, then instead of showing any digits, the field is filled with a pattern consisting of # (hash) characters, and possibly an embedded decimal point. In contrast, the . character permits the field's width to increase to accommodate overflowing output. If overflow takes place and the ! character appears other than as the rightmost character of the pattern, then the decimal point character . character appears at the position indicated by that ! character. If the ! character is the rightmost character of the pattern, then, just as in the case of normal, non-overflowing output, it doesn't contribute to the width of the hash fill, and only hash characters appear.
If commas appear in the numeric pattern according to the more complex syntactic rule, they count toward the field width and specify the insertion of digit-separating commas at the indicated locations. Digit separators may be specified on either side of the decimal point, but not adjacent to it. In the output, a digit separating comma shall not appear if it would be immediately preceded by a + or - sign or space. In this situations, the sign character or space appears in place of the digit separator. A digit separator that appears in a position occupied by a space is also suppressed in favor of the space. Digit separators are included among leading zeros. It is not logically possible for a digit separator to appear as the first character of a pattern's output, because it may not be the first character of a pattern. However, if a numeric pattern is preceded or followed by a comma, those commas are ordinary characters which are copied to the output.
When, due to the presence of !, an overflowing field is handled by the generation of a the hash character fill, the hash characters are treated as digits for the purpose of digit separation.
When the pattern uses parentheses to specify that negative numbers are to be shown with parentheses, the parentheses count toward the field width. The field portion between the parentheses is called the inner field. The parentheses appear in the output when the number is negative, and are placed immediately outside of the inner field, so that if leading zeros are not requested, there may be one or more spaces between the opening parenthesis and the first digit. If the number is nonnegative, then each parenthesis is replaced by one space, flanking the inner field in the same manner as parentheses.
;; numeric formatting
(pic "######" 1234.1) -> " 1234"
(pic "######.#" 1234.1) -> " 1234.1"
(pic "#######.##" 1234.1) -> " 1234.10"
(pic "#######.##" -1234.1) -> " -1234.10"
(pic "0######.##" 1234.1) -> "0001234.10"
(pic "+######.##" 1234.1) -> " +1234.10"
(pic "-######.##" 1234.1) -> " 1234.10"
(pic "+0#####.##" 1234.1) -> "+001234.10"
(pic "-0#####.##" 1234.1) -> " 001234.10"
(pic "#######.##" -1234.1) -> " -1234.10"
;; digit separation
(pic "0,###,###.##" 1234.1) -> "0,000,123.10"
(pic "#,###,###.##" 1234.1) -> " 123.10"
;; overflow with !
(pic "#!#" 1234) -> "###"
(pic "#!#" 123) -> "###"
(pic "-##!#" -123) -> "#####"
(pic "+##!#" 123) -> "#####"
(pic "###!" 1234) -> "###"
;; negative parentheses
(pic "(#,###.##) 1234.56) -> " 1,234.56 "
(pic "(#,###.##) -234.56) -> "( 234.56)"
;; alignment, multiple arguments
(pic "<<<<<< 0#.# >>>>>>>" "foo" (+ 2 2) "bar")
--> "foo 04.0 bar"
;; quasiliteral
(let ((a 2) (b "###") (c 13.5))
(pic `abc@(+ a a)###.##@b>>>>` c "x"))
--> "abc4 13.50### x"
;; filename generation
(mapcar (do pic "foo~-0##.jpg") (rlist 0..5 8 12))
--> ("foo-000.jpg" "foo-001.jpg" "foo-002.jpg" "foo-003.jpg"
"foo-004.jpg" "foo-005.jpg" "foo-008.jpg" "foo-012.jpg")
(print obj [stream [pretty-p]])
(pprint obj [stream])
(prinl obj [stream])
(pprinl obj [stream])
(tostring obj)
(tostringp obj)
The print and pprint functions render a printed character representation of the obj argument into stream.
If the stream argument is not supplied, then the destination is the stream currently stored in the *stdout* variable.
If Boolean argument pretty-p is not supplied or is explicitly specified as nil, then the print function renders in a way which strives for read-print consistency: an object is printed in a notation which is recognized as a similar object of the same kind when it appears in TXR source code. Floating-point objects are printed as if using the format function, with formatting controlled by the *print-flo-format* variable.
If pretty-p is true, then print does not strive for read-print consistency. In TXR, the term pretty printing refers to rendering a printed representation of an object without the notational details required to unambiguously delimit the object, and represent its value and type without ambiguity. For instance, the four-character string "abcd", the two-byte buffer object #b'abcd' as well as the symbol abcd all pretty-print as abcd. To understand the meaning, the user has to refer to the documentation of the specific application which produces that representation.
When pretty-p is true, strings are printed by sending their characters to the output stream, as if by the put-string function, rather than being rendered in the string literal notation consisting of double quotes, and escape sequences for control characters. Likewise, character objects are printed via put-char rather than the #\ notation.
When pretty-p is true, buffer objects are printed as strings of hexadecimal digit pairs, without being embedded in the #b'...' notation, and without any line breaks. This behavior is new in TXR 275; see the COMPATIBILITY section.
The pretty-p flag causes symbols to be printed without their package prefix, except that symbols from the keyword package are still printed with the leading colon. Floating-point objects are printed as if using the format function, with formatting controlled by the *pprint-flo-format* variable.
When aggregate objects like conses, ranges and vectors are printed, the notations of these objects themselves are unaffected by the pretty-p flag; however, that flag is distributed to the elements.
The print function returns obj.
The pprint ("pretty print") function is equivalent to print, with the pretty-p argument hardcoded true.
The prinl function ("print and new line") behaves like a call to print with pretty-p defaulting to nil, followed by issuing a newline characters to the stream.
The pprinl function ("pretty print and new line") behaves like pprint followed by issuing a newline to the stream.
The tostring and tostringp functions are like print and pprint, but they do not accept a stream argument. Instead they print to a freshly instantiated string stream, and return the resulting string.
The following equivalences hold between calls to the format function and calls to the above functions:
(format stream "~s" obj) <--> (print obj stream)
(format t "~s" obj) <--> (print obj)
(format t "~s\n" obj) <--> (prinl obj)
(format nil "~s" obj) <--> (tostring obj)
For pprint, tostringp and pprinl, the equivalence is produced by using ~a in format rather than ~s.
For floating-point numbers, the above description of the behavior in terms of the format specifiers ~s and ~a only applies with respect to the default values of the variables *print-flo-format* and *pprint-flo-format*.
For characters, the print function behaves as follows: most control characters in the Unicode C0 and C1 range are rendered using the #\x notation, using two hex digits. Codes in the range D800 to DFFF, and the codes FFFE and FFFF are printed in the #\xNNNN with four hexadecimal digits, and character above this range are printed using the same notation, but with six hexadecimal digits. Certain characters in the C0 range are printed using their names such as #\nul and #\return, which are documented in the Character Literals section. The DC00 character is printed as #\pnul. All other characters are printed as #\char where char is the actual character.
Caution: read-print consistency is affected by trailing material. If additional digits are printed immediately after a number without intervening whitespace, they extend that number. If hex digits are printed after the character x, which is rendered as #\x, they look like a hex character code.
(tprint obj [stream])
The tprint function prints a representation of obj on stream.
If the stream argument is not supplied, then the destination is the stream currently stored in the *stdout* variable.
For all object types except lists and vectors, tprint behaves like pprinl.
If obj is a list or vector, then tprint recurses: the tprint function is applied to each element. An empty list or vector results in no output at all. This effectively means that an arbitrarily nested structure of lists and vectors is printed flattened, with one element on each line.
(display-width char)
(display-width string)
The display-width function calculates the number of places occupied by the printed representation of char or string on a monospace display which renders certain characters, such as the East Asian kanji and other characters, using two places.
For a string argument, this value is the sum of the individual display width of the string's constituent characters. The display width of an empty string is zero.
Control characters are assigned a display width of zero, regardless of their display control semantics, if any.
Characters marked by Unicode as being wide or full width, have a display width of two. Other characters have a display width of one.
(streamp obj)
The streamp function returns t if obj is any type of stream. Otherwise it returns nil.
(real-time-stream-p obj)
The real-time-streamp-p function returns t if obj is a stream marked as "real-time". If obj is not a stream, or not a stream marked as "real-time", then it returns nil.
Only certain kinds of streams accept the real-time attribute: file streams and tail streams. This attribute controls the semantics of the application of lazy-stream-cons to the stream. For a real-time stream, lazy-stream-cons returns a stream with "naive" semantics which returns data as soon as it is available, at the cost of generating spurious nil item when the stream terminates. The application has to recognize and discard that nil item. The ordinary lazy streams read ahead by one line and suppress this extra item, so their representation is more accurate.
When TXR starts up, it automatically marks the *stdin* stream as real-time, if it is connected to a TTY device (a device for which the POSIX function isatty reports true). This is only supported on platforms that have this function. The behavior is overridden by the -n command-line option.
(open-file path [mode-string])
The open-file function creates a stream connected to the file which is located at the given path, which is a string.
The mode-string argument is a string which uses the same conventions as the mode argument of the C language fopen function, with greater permissiveness, and some extensions.
The syntax of mode-string is described by the following grammar. Note that it permits no whitespace characters:
mode-string := [ mode ] [ options ]
mode := { selector [ + ] | + }
selector := { r | w | a | m | T }
options := { b | x | l | u | i | n | digit |
z[digit] | redirection | ?fdno }
digit := { 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 }
If the mode-string argument is omitted, the behavior is the same as an empty mode string.
The mode part of the mode string generates the following possibilities:
The T mode offers a way to create temporary files in a robust way in any file system which supports the mechanism. There is no concern about choosing a unique file name, since the file doesn't have one. The file is guaranteed to disappear if the process is terminated in any manner. In contrast, traditional temporary files which are initially named a name and then unlinked may remain if the process is abruptly terminated before it is able to call unlink.
On Linux, it is possible to link a file created with "T" into the filesystem, according to the following pattern:
;; atomically create file called "name" with content "hello"
(let* ((stream (open-file "." "T"))
(fd (fileno stream)))
(put-string "hello\n" stream)
(flush-stream stream)
(rlink `/proc/self/fd/@fd` "name"))
The atomic creation of a file can be simulated by the familiar pattern of writing to a visible temporary file and then renaming. However, the above pattern eliminates the risk that a temporary file will be left behind if the procedure is interrupted for any reason before reaching the rlink call. Any reason includes process termination that cannot be intercepted and handled, and operating system failure or power loss.
(open-tail path [mode-string [seek-to-end-p]])
The open-tail function creates a tail stream connected to the file which is located at the given path. The mode-string argument is a string which uses the same conventions as the mode argument of the C language fopen function. If this argument is omitted, then "r" is used. See the open-file function for a discussion of modes.
The seek-to-end-p argument is a Boolean which determines whether the initial read/write position is at the start of the file, or just past the end. It defaults to nil. This argument only makes a difference if the file exists at the time open-tail is called. If the file does not exist, and is later created, then the tail stream will follow that file from the beginning. In other words, seek-to-end-p controls whether the tail stream reads all the existing data in the file, if any, or whether it reads only newly added data from approximately the time the stream is created.
A tail stream has special semantics with regard to reading at the end of file. A tail stream never reports an end-of-file condition; instead it polls the file until more data is added. Furthermore, if the file is truncated, or replaced with a smaller file, the tail stream follows this change: it automatically opens the smaller file and starts reading from the beginning (the seek-to-end-p flag only applies to the initial open). In this manner, a tail stream can dynamically grow rotating log files.
Caveat: since a tail stream can reopen a new file which has the same name as the original file, it behave incorrectly if the program changes the current working directory, and the pathname is relative.
(open-directory path)
The open-directory function tries to create a stream which reads the directory given by the string argument path. If a filesystem object exists under the path, is accessible, and is a directory, then the function returns a stream. Otherwise, a file error exception is thrown.
The resulting stream supports the get-line operation. Each call to the get-line operation retrieves a string representing the next directory entry. The value nil is returned when there are no more directory entries. The . and .. entries in Unix filesystems are not skipped.
(tmpfile)
The tmpfile function creates a new temporary binary file which is different from any existing file. It opens a stream for that file and returns the stream. The stream is created with the open-file mode "w+b". When the stream is closed, or the TXR image terminates, the file is deleted.
Note: the tmpfile function is implemented using the same-named ISO C and POSIX library function. On POSIX systems of sufficient quality, tmpfile deletes the file before returning the open stream, such that the file object continues to exist while the stream is open, but is not known by any name in the file system. POSIX (IEEE Std 1003.1-2017) notes that in some implementations, "a permanent file may be left behind if the process calling tmpfile() is killed while it is processing a call to tmpfile".
Notes: if a unique file is required which exists in the file system under a known name until explicitly deleted, the mkstemp function may be used. If a unique directory needs to be created, the mkdtemp function may be used. These two functions are described in the Unix Filesystem Complex Operations section of the manual.
(make-string-input-stream string)
The make-string-input-stream function produces an input stream object. Character read operations on the stream object read successive characters from string. Output operations and byte operations are not supported.
(make-string-byte-input-stream string)
The make-string-byte-input-stream function produces an input stream object. Byte read operations on this stream object read successive byte values obtained by encoding string into UTF-8. Character read operations are not supported, and neither are output operations.
(make-strlist-input-stream list)
The make-strlist-input-stream function produces an input stream object based on a list of strings. Through the character read operations invoked on this stream, the list of strings appears as a list of newline-terminated lines. Output operations and byte operations are not supported.
(make-string-output-stream)
The make-string-output-stream function, which takes no arguments, creates a string output stream. Data sent to this stream is accumulated into a string object. String output streams support both character and byte output operations. Bytes are assumed to represent a UTF-8 encoding, and are decoded in order to form characters which are stored into the string.
If an incomplete UTF-8 code is output, and a character output operation then takes place, that code is assumed to be terminated and is decoded as invalid bytes. The UTF-8 decoding machine is reset and ready for the start of a new code.
The get-string-from-stream function is used to retrieve the accumulated string.
If the null character is written to a string output stream, the behavior is unspecified. TXR strings cannot contain null bytes. The pseudo-null character #\xDC00, also notated #\pnul, will produce a null byte when converted to UTF-8 and thus serves as an effective internal representation of the null character in external data.
(get-string-from-stream stream)
The stream argument must be a string output stream. This function finalizes the data sent to the stream and retrieves the accumulated character string.
If a partial UTF-8 code has been written to stream, and then this function is called, the byte stream is considered complete and the partial code is decoded as invalid bytes.
After this function is called, further output on the stream is not possible.
(make-strlist-output-stream)
The make-strlist-output-stream function is similar to make-string-output-stream. However, the stream object produced by this function does not produce a string, but a list of strings. The data is broken into multiple strings by newline characters written to the stream. Newline characters do not appear in the string list. Also, byte output operations are not supported.
(get-list-from-stream stream)
The get-list-from-stream function returns the string list which has accumulated inside a string output stream given by stream. The string output stream is finalized, so that further output is no longer possible.
(with-in-string-stream (stream-var string)
body-form*)
The with-in-string-stream macro binds the symbol stream-var as a variable, initializing it with a newly created string input stream. The string input stream is constructed from string as if by the (make-string-input-stream string) expression.
Then it evaluates the body-forms in the scope of the variable.
The value of the last body-form is returned, or else nil if no forms are present.
The stream-var argument must be a bindable symbol, as defined by the bindable function.
The string argument must be a form which evaluates to a character string value.
(with-in-string-byte-stream (stream-var string)
body-form*)
The with-in-string-byte-stream macro binds the symbol stream-var as a variable, initializing it with a newly created string byte input stream. The string input stream is constructed from string as if by the (make-string-byte-input-stream string) expression.
Then it evaluates the body-forms in the scope of the variable.
The value of the last body-form is returned, or else nil if no forms are present.
The string argument must be a form which evaluates to a character string value.
(with-out-string-stream (stream-var) body-form*)
The with-out-string-stream macro binds the symbol specified by the stream-var argument as a variable, initializing it with a newly created string output stream. The output stream is created as if by the make-string-output-stream function.
Then it evaluates body-forms in the scope of that variable.
After these forms are evaluated, the string is extracted from the string output stream, as if by the get-string-from-stream function, and returned as the result value of the form.
(with-out-strlist-stream (stream-var) body-form*)
The with-out-strlist-stream macro binds the symbol specified by the stream-var argument as a variable, initializing it with a newly created string list output stream. The output stream is created as if by the make-strlist-output-stream function.
Then it evaluates body-forms in the scope of that variable.
After these forms are evaluated, the string list is extracted from the string output stream, as if by the get-strlist-from-stream function, and returned as the result value of the form.
(make-byte-input-stream obj)
The make-byte-input-stream creates a stream which supports the get-byte operation for traversing a byte-wise representation of obj.
The function serves as a generic interface for calling one of several other stream constructing functions based on the type of the obj argument.
The obj argument must be either a buffer, in which case make-byte-input-stream behaves like make-buf-stream, or else a string, in which case the function behaves like make-string-byte-input-stream.
Note: the repertoire of types handled by make-byte-input-stream may expand in future language versions.
(close-stream stream [throw-on-error-p])
The close-stream function performs a close operation on stream, whose meaning is depends on the type of the stream. For some types of streams, such as string streams, it does nothing. For streams which are connected to operating system files or devices, will perform a close of the underlying file descriptor, and dissociate that descriptor from the stream. Any buffered data is flushed first.
close-stream returns a Boolean true value if the close has occurred without errors, otherwise nil.
For most streams, "without errors" means that any buffered output data is flushed successfully.
For command and process pipes (see open-command and open-process), success also means that the process terminates normally, with a successful error code, or an unsuccessful one. An abnormal termination is considered an error, as is the inability to retrieve the termination status, as well as the situation that the process continues running in spite of the close attempt. Detecting these situations is platform specific.
If the throw-on-error-p argument is specified, and isn't nil, then the function throws an exception if an error occurs during the close operation instead of returning nil.
If close-stream is called in such a way that it returns a value, without throwing an exception, and that value isn't nil, that value is retained. Additional calls to the function with the same stream object return that same value without having any effect on the stream. These additional calls ignore the throw-on-error-p argument.
The stream may be associated with a process, in one of several ways: implicitly, by the functions open-process and open-command and related functions, or explicitly by the open-fileno function, if a pid argument is specified. In this situation, close-stream waits for the termination of that process, after closing the underlying file descriptor. If the process terminates normally, then close-stream returns its termination status, which is zero if the termination is successful. If the status of the process cannot be obtained, or is an abnormal termination, then the return value is nil. In that situation, if throw-on-error-p is true, an exception is thrown instead.
(with-stream (stream-var init-form)
body-form*)
The with-stream macro binds the variable whose name is given by the stream-var argument, and macro arranges for the evaluation of body-forms in the scope of that variable.
The variable is initialized with the value produced by the evaluation of init-form which must be an expression which evaluates to a stream.
After each body-form is evaluated, the stream is closed, as if by the (close-stream stream-var) expression.
The value of the last body-form then becomes the result value of the form, or else nil if these forms are absent.
If the evaluation of the body-forms is abandoned, the stream is still closed. That is to say, the closure of the stream is a protected action, as if by the unwind-protect operator.
(get-error stream)
(get-error-str stream)
(clear-error stream)
When a stream operation fails, the get-error and get-error-str functions may be used to inquire about a more detailed cause of the error.
Not all streams support these functions to the same extent. For instance, string input streams have no persistent state. The only error which occurs is the condition when the string has no more data.
The get-error inquires stream about its error condition.
The function returns nil to indicate there is no error condition, t to indicate an end-of-data condition, or else a value which is specific to the stream type indicating the specific error type.
Note: for some streams, it is possible for the t value to be returned even though no operation has failed; that is to say, the streams "know" they are at the end of the data even though no read operation has failed. Code which depends on this will not work with streams which do not thus indicate the end-of-data a priori, but by means of a read operation which fails.
The get-error-str function returns a text representation of the error code. The nil error code is represented as the string no error; the t error code as eof and other codes have a stream-specific representation.
The clear-error function removes the error situation from a stream. On some streams, it does nothing. If an error has occurred on a stream, this function should be called prior to retrying any I/O or positioning operations. The return value is the previous error code, or nil if there was no error, or the operation is not supported on the stream.
(get-line [stream])
(get-char [stream])
(get-byte [stream])
These fundamental stream functions perform input. The stream argument is optional. If it is specified, it should be an input stream which supports the given operation. If it is not specified, then the *stdin* stream is used.
The get-char function pulls a character from a stream which supports character input. Streams which support character input also support the get-line function which extracts a line of text delimited by the end of the stream or a newline character and returns it as a string. (The newline character does not appear in the string which is returned).
Character input from streams based on bytes requires UTF-8 decoding, so that get-char may actually read several bytes from the underlying low-level operating system stream.
The get-byte function bypasses UTF-8 decoding and reads raw bytes from any stream which supports byte input. Bytes are represented as integer values in the range 0 to 255.
Note that if a stream supports both byte input and character input, then mixing the two operations will interfere with the UTF-8 decoding.
These functions return nil when the end of data is reached. Errors are represented as exceptions.
See also: get-lines
(get-string [stream [count [close-after-p]]])
The get-string function reads characters from a stream, and assembles them into a string, which is returned. If the stream argument is omitted, then the *stdin* stream is used.
The stream is closed after extracting the data, unless close-after-p is specified as nil. The default value of this argument is t.
If the count argument is missing, then all of the characters from the stream are read and assembled into a string.
If present, the count argument should be a positive integer indicating a limit on how many characters to read. The returned string will be no longer than count, but may be shorter.
(unget-char char [stream])
(unget-byte byte [stream])
These functions put back, into a stream, a character or byte which was previously read. The character or byte must match the one which was most recently read. If the stream argument is omitted, then the *stdin* stream is used.
If the operation succeeds, the byte or character value is returned. A nil return indicates that the operation is unsupported.
Some streams do not support these operations; some support only one of them. In general, if a stream supports get-char, it supports unget-char, and likewise for get-byte and unget-byte.
Streams may require a pushed back byte or character to match the character which was previously read from that stream position, and may not allow a byte or character to be pushed back beyond the beginning of the stream.
Space may be available for only one byte of pushback under the unget-byte operation.
The number of characters that may be pushed back by unget-char is not limited.
Pushing both a byte and a character, in either order, is also unsupported. Pushing a byte and then reading a character, or pushing a character and reading a byte, are unsupported mixtures of operations.
If the stream is binary, then pushing back a byte decrements its position, except if the position is already zero. At that point, the position becomes indeterminate.
The behavior of pushing back immediately after a seek-stream positioning operation is unspecified.
(put-string string [stream])
(put-line [string [stream]])
(put-char char [stream])
(put-byte byte [stream])
These functions perform output on an output stream. The stream argument must be an output stream which supports the given operation. If it is omitted, then *stdout* is used.
The put-char function writes a character given by char to a stream. If the stream is based on bytes, then the character is encoded into UTF-8 and multiple bytes are written. Streams which support put-char also support put-line and put-string.
The put-string function writes the characters of a string out to the stream as if by multiple calls to put-char. The string argument may be a symbol, in which case its name is used as the string.
The put-line function is like put-string, but also writes an additional newline character. The string is optional in put-line, and defaults to the empty string.
The put-byte function writes a raw byte given by the byte argument to stream, if stream supports a byte write operation. The byte value is specified as an integer value in the range 0 to 255.
All these functions return t. On failure, they do not return, but throw exceptions of type file-error.
(put-strings sequence [stream]])
(put-lines sequence [stream]])
These functions assume sequence to be a sequence of strings, or of symbols, or a mixture thereof. These strings are sent to the stream. The stream argument must be an output stream. If it is omitted, then *stdout* is used.
The put-strings function iterates over sequence and writes each element to the stream as if using the put-string function.
The put-lines function iterates over sequence and writes each element to the stream as if using the put-line function.
Both functions return t.
(flush-stream [stream])
The flush-stream function is meaningful for output streams which accumulate data which is passed on to the operating system in larger transfer units. Calling flush-stream causes all accumulated data inside stream to be passed to the operating system. If called on streams for which this function is not meaningful, it does nothing, and returns nil.
If stream is omitted, the current value of *stdout* is used.
(seek-stream stream offset whence)
The seek-stream function is meaningful for file streams. It changes the current read/write position within stream. It can also be used to determine the current position: see the notes about the return value below.
The offset argument is a positive or negative integer which gives a displacement that is measured from the point identified by the whence argument.
Note that for text files, there isn't necessarily a 1:1 correspondence between characters and positions due to line-ending conversions and conversions to and from UTF-8.
The whence argument is one of three keywords: :from-start, :from-current and :from-end. These denote the start of the file, the current position in the file and the end of the file.
If offset is zero, and whence is :from-current, then seek-stream returns the current absolute position within the stream, if it can successfully obtain it. Otherwise, it returns t if it is successful.
If a character has been successfully put back into a text stream with unget-char and is still pending, then the position value is unspecified. If a byte has been put back into a binary stream with unget-byte, and the previous position wasn't zero, then the position is decremented by one.
On failure, it throws an exception of type stream-error.
(truncate-stream stream [length])
The truncate-stream causes the length of the underlying file associated with stream to be set to length bytes.
The stream must be a file stream, and must be open for writing.
If length is omitted, then it defaults to the current position, retrieved as if by invoking the seek-stream with an offset argument of zero and whence argument of :from-current. Hence, after the truncate-stream operation, that position is one byte past the end of the file.
(stream-get-prop stream indicator)
(stream-set-prop stream indicator value)
These functions get and set properties on a stream. Only certain properties are meaningful with certain kinds of streams, and the meaning depends on the stream. If two or more stream types support a property of the same name, it is expected that the property has the same or similar meaning for both streams to the maximum extent that similarity is possible.
The stream-set-prop function sets a property on a stream. The indicator argument is a symbol, usually a keyword symbol, denoting the property, and value is the property value. If the stream understands and accepts the property, the function returns t. Otherwise it returns nil.
The stream-get-prop function inquires about the value of a property on a stream. If the stream understands the property, then it returns its current value. If the stream does not understand a property, nil is returned, which is also returned if the property exists, but its value happens to be nil.
The :name property is widely supported by streams of various types. It associates the stream with a name. This property is not always modifiable.
File, process and stream socket I/O streams have a :fd property which can be accessed, but not modified. It retrieves the same value as the fileno function.
The "real time" property supported by these streams, connected with the real-time-stream-p function, also appears as the :real-time property.
I/O streams also have a property called :byte-oriented which, if set, suppresses the decoding of UTF-8 on character input. Rather, each byte of the file corresponds directly to one character. Bytes in the range 1 to 255 correspond to the character code points U+0001 to U+00FF. Byte value 0 is mapped to the code point U+DC00.
The logging priority of the *stdlog* syslog stream is controlled by the :prio property.
If stream is a catenated stream (see the function make-catenated-stream) then these functions transparently operate on the current head stream of the catenation.
(make-catenated-stream stream*)
(cat-streams stream-list)
The make-catenated-stream function takes zero or more arguments which are input streams of the same type, and combines them into a single virtual stream called a catenated stream.
The cat-streams function takes a single list of input streams of the same type, and similarly combines them into a catenated stream.
A catenated stream does not support seeking operations or output, regardless of the capabilities of the streams in the list.
If the stream list is not empty, then the leftmost element of the list is called the head stream.
The get-char, get-byte, get-line, unget-char and unget-byte functions delegate to the corresponding operations on the head stream, if it exists. If the stream list is empty, they return nil to the caller.
If the get-char, get-byte or get-line operation on the head stream yields nil, and there are more lists in the stream, then the stream is closed, removed from the list, and the next stream, if any, becomes the head list. The operation is then tried again. If any of these operations fail on the last list, it is not removed from the list, so that a stream remains in place which can take the unget-char or unget-byte operations.
In this manner, the catenated streams appear to be a single stream.
Note that the operations can fail due to being unsupported. It is the caller's responsibility to make sure all of the streams in the list are compatible with the intended operations.
If the stream list is empty then an empty catenated stream is produced. Input operations on this stream yield nil, and the unget-char and unget-byte operations throw an exception.
(catenated-stream-p obj)
The catenated-stream-p function returns t if obj is a catenated stream. Otherwise it returns nil.
(catenated-stream-push new-stream cat-stream)
The catenated-stream-push function pushes new-stream to the front of the stream list inside cat-stream.
If an unget-byte or unget-char operation was successfully performed on cat-stream previously to a call to catenated-stream-push, those operations were forwarded to the front stream. If those bytes or characters are still pending, they are pending inside that stream, and thus are logically preceded by the contents of new-stream.
(open-files path-list [alternative-stream [mode-string]])
(open-files* path-list [alternative-stream [mode-string]])
The open-files and open-files* functions create a list of streams by invoking the open-file function on each element of path-list. By default, the mode string "r" is passed to open-file; if the mode-string argument specified, it overrides this default. In that situation, the specified mode should permit reading.
These streams are turned into a catenated stream as if they were the arguments of a call to make-catenated-stream.
The effect is that multiple files appear to be catenated together into a single input stream.
If the optional alternative-stream argument is supplied, then if path-list is empty, alternative-stream is returned instead of an empty catenated stream.
The difference between open-files and open-files* is that open-files creates all of the streams up-front. So if any of the paths cannot be opened, the operation throws. The open-files* variant is lazy: it creates a lazy list of streams out of the path list. The streams are opened as needed: before the second stream is opened, the program has to read the first stream to the end, and so on.
Collect lines from all files that are given as arguments on the command line. If there are no files, then read from standard input:
@(next (open-files *args* *stdin*))
@(collect)
@line
@(end)
(path-equal left-path right-path)
The path-equal function determines whether the two paths left-path and right-path are equal under a certain definition of equivalence, whose requirements are given below. The function returns t if the paths are equal, otherwise nil.
If left-path and right-path are strings which are identical under the equal function, then they are considered equal paths.
Otherwise, the two paths are equal if the relative path from left-path to right-path is "." (dot), as would be determined by the path-rel function, if it were applied to left-path and right-path as its arguments. If path-rel would return the dot path, then the two paths are equal. If path-rel would return any other value, or throw an exception, then the paths are unequal.
;; simple case
(path-equal "a" "a") -> t
(path-equal "a" "b") -> nil
;; trailing slashes don't matter
(path-equal "a" "a/") -> t
(path-equal "a/" "a/") -> t
;; .. components resolved:
(path-equal "a/b/../c" "a/c") -> t
;; . components resolved:
(path-equal "a" "a/././.") -> t
(path-equal "a/." "a/././.") -> t
;; (On Microsoft Windows)
;; different drive:
(path-equal "c:/a" "d:/b/../a") -> nil
;; same drive:
(path-equal "c:/a" "c:/b/../a") -> t
(abs-path-p path)
(portable-abs-path-p path)
The abs-path-p and portable-abs-path-p functions test whether the argument path is an absolute path, returning a t or nil indication.
The portable-abs-path-p function behaves in the same manner on all platforms, implementing a platform-agnostic definition of absolute path, as follows.
An absolute path is a string which either begins with a slash or backslash character, or which begins with an alphanumeric word, followed by a colon, followed by a slash or backslash.
The empty string isn't an absolute path.
Examples of absolute paths under portable-abs-path-p:
/etc
c:/tmp
ftp://user@server
disk0:/home
Z:\Users
Examples of strings which are not absolute paths:
.
abc
foo:bar/x
$:\abc
The abs-path-p is similar to portable-abs-path-p except that it reports false for paths which are not absolute paths according to the host platform. The following paths are not absolute on POSIX platforms:
c:/tmp
ftp://user@server
disk0:/home
Z:\Users
(pure-rel-path-p path)
The pure-rel-path-p function tests whether the string path represents a pure relative path, which is defined as a path which isn't absolute according to abs-path-p, which isn't the string "." (single period), which doesn't begin with a period followed by a slash or backslash, and which doesn't begin with an alphanumeric word terminated by a colon.
The empty string is a pure relative path.
Other examples of pure relative paths:
abc.d
.tmp/bar
1234
x
$:/xyz
Examples of strings which are not pure relative paths:
.
/
/etc
./abc
.\
foo:
$:\abc
(dir-name path)
(base-name path [suffix])
The dir-name and base-name functions calculate, respective, the directory part and base name part of a pathname.
The calculation is performed in a platform-dependent way, using the characters in the variable path-sep-chars as path component separators.
Both functions first remove from any further consideration all superfluous trailing occurrences of the directory separator characters from path. Thus input such as "a////" is reduced to just "a", and "///" is reduced to "/".
The resulting trimmed path is the effective path.
If the effective path is an empty string, then dir-name returns "." and base-name returns the empty string.
If the effective path is not empty, and contains no path separator characters, then dir-name returns "." and base-name returns the effective path.
Otherwise, the effective path is divided into two parts: the raw directory prefix and the remainder.
The raw directory prefix is the maximally long prefix of the effective path which ends in a separator character.
The dir-name function returns the raw directory prefix, if that prefix consists of nothing but a single directory separator character. Otherwise it returns the raw directory prefix, with the trailing path separator removed.
The base-name function returns the remaining part of the effective path, after the raw directory prefix.
If the suffix argument is given to base-name, it specifies a proper suffix to be removed from the returned base name. First, the base name is calculated according to the foregoing rules. Then, if suffix matches a trailing portion of the base name, but not the entire base name, it is removed from the base name.
The suffix parameter may be given a nil, argument, which is treated exactly as if it were absent. Note: this requirement allows for the following idiom to work correctly even in cases when p has no suffix:
;; calculate base name of p with short suffix removed
(base-name p (short-suffix p))
;; calculate base name of p with long suffix removed
(base-name p (long-suffix p))
(base-name "") -> ""
(base-name "/") -> "/"
(base-name ".") -> "."
(base-name "./") -> "."
(base-name "a") -> "a"
(base-name "/a") -> "a"
(base-name "/a/") -> "a"
(base-name "/a/b") -> "b"
(base-name "/a/b/") -> "b"
(base-name "/a/b///") -> "b"
;; with suffix
(base-name "" "") -> ""
(base-name "/" "/") -> "/"
(base-name "/" "") -> "/"
(base-name "." ".") -> "."
(base-name "." "") -> "."
(base-name "./" "/") -> "."
(base-name "a" "a") -> "a"
(base-name "a" "") -> "a"
(base-name "a.b" ".b") -> "a"
(base-name "a.b/" ".b") -> "a"
(base-name "a.b/" ".b/") -> "a.b"
(base-name "a.b/" "a.b") -> "a.b"
(long-suffix path [alt])
(short-suffix path [alt])
The long-suffix and short-suffix functions calculate the long suffix and short suffix of path, which must be a string.
If path does not contain any occurrences of the character . (period) in the role of a suffix delimiter, then path does not have a suffix. In this situation, both functions return the alt argument, which defaults to nil if it is omitted.
What it means for path to have a suffix delimiter is that the . character occurs somewhere in the last component of path, other than as the first character of that component. What constitutes the last component is specified in more detail below.
If a suffix delimiter is present, then the long or short suffix is the substring of path which includes the delimiting period and all characters which follow, except that if path ends in a sequence of one or more path separator characters, those characters are omitted from the returned suffix.
If multiple periods occur in the last component of the path, the delimiter for the long suffix is the leftmost period and the delimiter for the short suffix is the rightmost period.
If the delimiting period is the rightmost character of path, or occurs immediately before a trailing path separator, then the suffix delimited by that period is the period itself.
If path contains only one suffix delimiter, then its long and short suffix coincide.
For the purpose of identifying the last component of path, if path ends a sequence of one or more path-separator characters, then those characters are removed from consideration. If the remaining string contains path-separator characters, then the last component consists of that portion of it which follows the rightmost path-separator character. Otherwise, the last component is the entire string. The suffix, if present, is identified and extracted from this last component.
(short-suffix "") -> nil
(short-suffix ".") -> nil
(short-suffix "abc") -> nil
(short-suffix ".abc") -> nil
(short-suffix "/.abc") -> nil
(short-suffix "abc" "") -> ""
(short-suffix "abc.") -> "."
(short-suffix "abc.tar") -> ".tar"
(short-suffix "abc.tar///") -> ".tar"
(short-suffix "abc.tar.gz") -> ".gz"
(short-suffix "abc.tar.gz/") -> ".gz"
(short-suffix "x.y.z/abc.tar.gz/") -> ".gz"
(short-suffix "x.y.z/abc.tar.gz//") -> nil
(long-suffix "") -> nil
(long-suffix ".") -> nil
(long-suffix "abc") -> nil
(long-suffix ".abc") -> nil
(long-suffix "/.abc") -> nil
(long-suffix "abc.") -> "."
(long-suffix "abc.tar") -> ".tar"
(long-suffix "abc.tar///") -> ".tar"
(long-suffix "abc.tar.gz") -> ".tar.gz"
(long-suffix "abc.tar.gz/") -> ".tar.gz"
(long-suffix "x.y.z/abc.tar.gz/") -> ".tar.gz"
(trim-long-suffix path)
(trim-short-suffix path)
The trim-long-suffix and trim-short-suffix functions calculate the portion of path long suffix and short suffix of the string argument path, and return a path with the suffix removed.
Respectively, trim-long-suffix and trim-short-suffix calculate the suffix in exactly the same manner as long-suffix and short-suffix.
If path is found not to contain a suffix, then it is returned.
If path contains a suffix, then a new string is returned from which the suffix is deleted. If the suffix is followed by one or more path separator characters, these are preserved in the return value.
(trim-short-suffix "") -> ""
(trim-short-suffix "a") -> "a"
(trim-short-suffix ".") -> "."
(trim-short-suffix ".a") -> ".a"
(trim-short-suffix "a.") -> "a"
(trim-short-suffix "a.b") -> "a"
(trim-short-suffix "a.b.c") -> "a.b"
(trim-short-suffix "a./") -> "a/"
(trim-short-suffix "a.b/") -> "a/"
(trim-short-suffix "a.b.c/") -> "a.b/"
(trim-long-suffix "a.b.c") -> "a"
(trim-long-suffix "a.b.c/") -> "a/"
(trim-long-suffix "a.b.c///") -> "a///"
(add-suffix path suffix)
The add-suffix function combines the string arguments path and suffix in a way which harmonizes with the long-suffix and short-suffix functions.
If path does not end in a path separator character, that category being defined by the path-sep-chars variable, then add-suffix returns the trivial string catenation of path and suffix.
Otherwise, add-suffix returns a string formed by inserting suffix into path just prior to the sequence of trailing path separator characters. The returned string is a catenation of that portion of path which excludes the sequence of trailing path separators, followed by suffix, followed by the sequence of trailing path separators.
A path separator which occurs as a part of syntax that indicates an absolute pathname is not considered a trailing separator. A path which begins with a separator is absolute. Other platform-specific path patterns may constitute an absolute pathname.
Note: in cases when suffix does not begin with a period, or is inserted in such a way that it is the start of a path component, then the functions long-suffix and short-suffix will not recognize suffix in the resulting path.
(add-suffix "" "") -> ""
(add-suffix "" "a") -> "a"
(add-suffix "." "a") -> ".a"
(add-suffix "." ".a") -> "..a"
(add-suffix "/" ".b") -> "/.b"
(add-suffix "//" ".b") -> "/.b/"
(add-suffix "//" "b") -> "/b/"
(add-suffix "a" "") -> "a"
(add-suffix "a" ".b") -> "a.b"
(add-suffix "a/" ".b") -> "a.b/"
(add-suffix "a//" ".b") -> "a.b//"
;; On MS Windows
(add-suffix "c://" "x") -> "c:/x/"
(add-suffix "host://" "x") -> "host://x"
(add-suffix "host:///" "x") -> "host://x/"
(path-cat [dir-path {rel-path}*])
The path-cat function joins together zero or more paths, returning the combined path. All arguments are strings.
The following description defines the behavior when path-cat is given exactly two arguments, which are interpreted as dir-path and rel-path. A description of the variable-argument semantics follows.
Firstly, the two-argument path-cat is related to the functions dir-name and base-name in the following way: if p is some path denoting an object in the file system, then (path-cat (dir-name p) (base-name p)) produces a path p* which denotes the same object. The paths p and p* might not be equivalent strings.
The path-cat function ensures that paths are joined without superfluous path-separator characters, regardless of whether dir-path ends in a separator.
If a separator must be added, the character / (forward slash) is always used, even on platforms where \ (backslash) is also a pathname separator, and even if either argument includes backslashes.
The path-cat function eliminates trivial occurrences of the . (dot) path component. It preserves trailing separators in the following way: if rel-path ends in a path-separator character, then the returned string shall end in that character; and if rel-path vanishes entirely because it is equivalent to the dot, then the returned string is dir-name itself.
If dir-path is an empty string, then rel-path is returned, and vice versa.
The variadic semantics of path-cat are as follows.
If path-cat is called with no arguments at all, it returns the path "." (period) denoting the relative path of the current directory.
If path-cat is called with one argument, that argument is returned.
If path-cat is called with three or more arguments, a left-associative reduction takes place using the two-argument semantics. The first two arguments are catenated into a single path, which is then catenated with the third argument, and so on.
The above semantics imply that the following equivalence holds:
[reduce-left path-cat list] <--> [apply path-cat list]
(path-cat "" "") --> ""
(path-cat "" ".") --> "."
(path-cat "." "") --> "."
(path-cat "." ".") --> "."
(path-cat "abc" ".") --> "abc"
(path-cat "." "abc") --> "abc"
(path-cat "./" ".") --> "./"
(path-cat "." "./") --> "./"
(path-cat "abc/" ".") --> "abc/"
(path-cat "./" "abc") --> "abc"
(path-cat "/" ".") --> "/"
(path-cat "/" "abc") --> "/abc"
(path-cat "ab/cd" "ef") --> "ab/cd/ef"
(path-cat "a" "b" "c") --> "a/b/c"
(path-cat "a" "b" "" "c" "/") --> "a/b/c/"
(trim-path-seps path)
The trim-path-seps function removes a consecutive run of one or more trailing separators from the end of the input string path.
The function treats the
path
in a system-independent way: both the backslash and forward slash
are considered a trailing separator.
The function preserves any necessary trailing separators, such as that of the absolute path "/" or the trailing slashes in volume absolute paths such as "c:/".
(trim-path-seps "") -> ""
(trim-path-seps "/") -> "/"
(trim-path-seps "//") -> "/"
(trim-path-seps "a///") -> "a"
(trim-path-seps "/a///") -> "/a")
(trim-path-seps "\\") -> "\\"
(trim-path-seps "\\\\") -> "\\"
(trim-path-seps "\\a\\\\\\") -> "\\a")
(trim-path-seps "c:/") -> "c:/"
(trim-path-seps "c://") -> "c:/"
(trim-path-seps "c:///") -> "c:/"
(trim-path-seps "c:a///") -> "c:a"
;; not a volume prefix:
(trim-path-seps "/c:/a///") -> "/c:/a"
(trim-path-seps "/c://///") -> "/c:")
(trim-path-seps "c:\\") -> "c:\\"
(trim-path-seps "c:\\\\") -> "c:\\"
(trim-path-seps "c:a\\\\\\") -> "c:a"
;; mixtures
(trim-path-seps "c:/\\/\\/") -> "c:/"
(rel-path from-path to-path)
The rel-path function calculates the relative path between two file system locations indicated by string arguments from-path and to-path. The from-path is assumed to be a directory. The return value is a relative path which could be used to access an object named by to-path if from-path were the current working directory.
The calculation performed by rel-path is a pure calculation; it has no interaction with the host operating system. No component of either input path has to exist. Symbolic links are not resolved. This can lead to incorrect results, as noted below.
Either both the inputs must be absolute paths, or must both be relative, otherwise an error exception is thrown.
On the MS Windows platform, if one input specifies a drive letter prefix, the other input must specify the same prefix, or else an error exception is thrown; there is no relative path between locations on different drives. The behavior is unspecified if the arguments are two UNC paths indicating different hosts.
The rel-path function first splits both paths into components according to the platform-specific pathname separators indicated by the path-sep-chars variable.
Next, it eliminates all empty components, . (dot) components and .. (dotdot) components from both separated paths. All dot components are removed, and any component which is neither dot nor dotdot is removed if it is followed by dotdot.
Then, a common prefix is determined between the two component sequences, and a relative component sequence is calculated from them as follows:
If the component sequence corresponding to from-path is longer than the common prefix, then the excess part of that sequence after the common prefix must not contain any .. (dotdot) components, or else an error exception is thrown. Otherwise, every component in this excess part of the from-path component sequence is converted to .. in order to express the relative navigation from from-path up to the directory indicated by the common prefix.
Next, if the component sequence corresponding to to-path has any components in excess of the common prefix, those excess components are appended to this possibly empty sequence of dotdot components, in order to express navigation from the common prefix down to the to-path object. This excess sequence coming from to-path may include .. components.
Finally, if the resulting sequence is nonempty, it is joined together using the leftmost path separator character indicated in path-sep-chars and returned. If it is empty, then the string "." is returned.
Note: because the function doesn't access the file system and in particular does not resolve symbolic links or other indirection devices, the result may be incorrect. For example, suppose that the current working directory contains a symbolic link called up which expands to .. (dotdot). The expression (rel-path "up/a" "../a") is oblivious to this, and calculates "../../../a". The correct result in light of up being an alias for .. calls for a return value of ".". The exact problem is that any symbolic links in the excess part of from-path after the common prefix are assumed by rel-path to be simple subdirectory names, which can be navigated in reverse using a .. link. This reverse navigation assumption is false for any symbolic link which which does not act as an alias for a subdirectory in the same location.
In situations where this possibility exists, it is recommended to use realpath function to canonicalize the input paths.
The following is an example of the algorithm being applied to arguments "a/d/../b/x/y/" and "a/b/w", where the assumption is that this is on a POSIX platform where the leftmost character in path-sep-chars is /:
Firstly, both inputs are converted to component sequences, those respectively being:
("a" "d" ".." "b" "x" "y" "")
("a" "b" "w")
Next the .. and empty components are removed:
("a" "b" "x" "y")
("a" "b" "w")
At this point, the common prefix is identified:
("a" "b")
The from-path has two components in excess of the prefix:
("x" "y")
which are each replaced by "..".
The to-path has one component in excess of the common prefix, "w".
These two sequences are appended together:
(".." ".." "w")
The resulting path is then formed by joining these with the separator character, resulting in the relative path "../../w".
;; mixtures of relative and absolute
(rel-path "/abc" "abc") -> ;; error
(rel-path "abc" "/abc") -> ;; error
;; dotdot in excess part of from path:
(rel-path "../../x" "y") -> ;; error
(rel-path "." ".") -> "."
(rel-path "./abc" "abc") -> "."
(rel-path "abc" "./abc") -> "."
(rel-path "./abc" "./abc") -> "."
(rel-path "abc" "abc") -> "."
(rel-path "." "abc") -> "abc"
(rel-path "abc/def" "abc/ghi") -> "../ghi"
(rel-path "xyz/../abc/def" "abc/ghi") -> "../ghi"
(rel-path "abc" "d/e/f/g/h") -> "../d/e/f/g/h"
(rel-path "abc" "d/e/../g/h") -> "../d/g/h"
(rel-path "d/e/../g/h" ".") -> "../../.."
(rel-path "d/e/../g/h" "a/b") -> "../../../a/b"
(rel-path "x" "../../../y") -> "../../../../y"
(rel-path "x///" "x") -> "."
(rel-path "x" "x///") -> "."
(rel-path "///x" "/x") -> "."
The path-sep-chars variable holds a string consisting of the characters which the underlying operating system recognizes as pathname separators.
If a particular of these characters is considered preferred on the host platform, that character is placed in the first position of path-sep-chars.
Altering the value of this variable has no effect on any TXR Lisp library function.
(read [source
[err-stream [err-retval [name [lineno]]]]])
(iread [source
[err-stream [err-retval [name [lineno]]]]])
The read function converts text denoting TXR Lisp structure, into the corresponding data structure. The source argument may be either a character string, or a stream. If it is omitted, then *stdin* is used as the stream.
The source must provide the text representation of one complete TXR Lisp object. If source and the function being applied is read, then if the object is followed by any non-whitespace material, the situation is treated as a syntax error, even if that material is a syntactically valid additional object. The iread function ignores this situation. Other differences between read and iread are given below.
Multiple calls to read on the same stream will extract successive objects from the stream. To parse successive objects from a string, it is necessary to convert it to a string stream.
The optional err-stream argument can be used to specify a stream to which diagnostics of parse errors are sent. If absent, the diagnostics are suppressed.
The optional name argument can be used to specify the file name which is used for reporting errors. If this argument is missing, the name is taken from the name property of the source argument if it is a stream, or else the word string is used as the name if source is a string.
The optional lineno argument, defaulting to 1, specifies the starting line number. This, like the name argument, is used for reporting errors.
If there are no parse errors, the function returns the parsed data structure. If there are parse errors, and the err-retval parameter is present, its value is returned. If the err-retval parameter is not present, then an exception of type syntax-error is thrown.
The iread function ("interactive read") is similar to read except that it parses a modified version of the syntax. The modified syntax does not support the application of the dot and dotdot operators on a top-level expression. For instance, if the input is a.b or a .. b then iread will only read the a token whereas read will read the entire expression.
This modified syntax allows iread to return immediately when an expression is recognized, which is the expected behavior if the input is being read from an interactive terminal. By contrast, read waits for more input after seeing a complete expression, because of the possibility that the expression will be further extended by means of the dot or dotdot operators. An explicit end-of-input signal must be given from the terminal to terminate the expression.
The special variable *rec-source-loc* controls whether these functions record source location info similarly to load. Note: if these functions are used to scan data which is evaluated as Lisp code, it may be useful to set *rec-source-loc* true in order to obtain better diagnostics. However, source location recording incurs a performance and storage penalty.
(read-objects [source
[err-stream
[err-retval [name [lineno]]]]])
The read-objects function has the same argument syntax and semantics as the read function, except that rather than reading one object, it reads all the Lisp objects from the source, and returns a list of these objects.
If the stream is empty, then read-objects returns the empty list nil, whereas the read function treats the situation as an error.
(parse-errors stream)
The parse-errors function retrieves information, from a stream, pertaining to the status of the most recent parsing operation performed on that stream: namely, a previous call to read, iread or get-json.
If the stream object has not been used for parsing, or else the most recent parsing operation did not encounter errors, then parse-errors returns nil.
If the most recent parsing operation on stream encountered errors, then parse-errors function returns a positive integer value indicating the error count. Otherwise it returns nil.
If a parse error operation encounters a syntax error before obtaining any token from the stream, then the error count is zero and parse-errors returns nil. Consequently, parse-errors may be used after a failed parse operation to distinguish a true syntax error from an end-of-stream condition.
(record-adapter regex [stream [include-match]])
The record-adapter function returns a new stream object which acts as an adapter to the existing stream.
If an argument is not specified for stream, then the *std-input* stream is used.
With the exception of get-line, all operations on the returned adapter transparently delegate to the original stream object.
When the get-line function is used on the adapter, it behaves differently. A string is extracted from stream, and returned. However, the string isn't a line delimited by a newline character, but rather a record delimited by regex. This record is extracted as if by a call to the read-until-match function, invoked with the regex, stream and include-match arguments.
All behavior which is built on the get-lines function is affected by the record-delimiting semantics of a record adapter's get-line implementation. Notably, the get-lines and lazy-stream-cons functions return a lazy list of delimited records rather than of lines.
TXR Lisp streams provide support for establishing hanging indentations in text output. Each stream which supports output has a built-in state variable called indentation mode, and another variable indicating the current indentation amount. When indentation mode is enabled, then prior to the first character of every line, the stream prepends the indentation: space characters equal in number to the current indentation value. This logic is implemented by the put-char and put-string functions, and all functions based on these. The put-byte function does not interact with indentation. The column position tracking will be incorrect if byte and character output are mixed, affecting the placement of indentation.
Indentation mode takes on four numeric values, given by the four variables indent-off, indent-data, indent-code and indent-foff. As far as stream output is concerned, the code and data modes represented by indent-code and indent-data behave the same way: both represent the "indentation turned on" state. The difference between them influences the behavior of the width-check function. This function isn't used by any lower-level stream output routines. It is used by the object printing functions like print and pprint to break up long lines. The indent-off and indent-foff modes are also treated the same way by lower level stream output, indicating "indentation turned off". The modes are distinguished by print and pprint in the following way: indent-off is a "soft" disable which allows these object-printing routines to temporarily turn on indentation while traversing aggregate objects. Whereas the indent-foff ("force off") value is a "hard" disable: the object-printing routines will not enable indentation and will not break up long lines.
These variables hold integer values representing output stream indentation modes. The value of indent-off is zero.
(get-indent-mode stream)
(set-indent-mode stream new-mode)
(test-set-indent-mode stream compare-mode new-mode)
These functions retrieve and manipulate the stream indent mode. The get-indent-mode retrieves the current indent mode of stream. The set-indent-mode function sets the indent mode of stream to new-mode and returns the previous mode.
Note: it is encouraged to save and restore the indentation mode, and in a way that is exception safe. If a block of code sets up indentation on a stream such as *stdout* and is terminated by an exception, the indentation will remain in effect and affect subsequent output. The with-resources macro or unwind-protect operator may be used.
(test-set-indent-mode stream compare-mode new-mode)
(test-neq-set-indent-mode stream compare-mode new-mode)
The test-set-indent-mode function sets the indent mode of stream to new-mode if and only if its current mode is equal to compare-mode. Whether or not it changes the mode, it returns the previous mode.
The test-neq-set-indent-mode only differs in that it sets stream to new-mode if and only if the current mode is not equal to compare-mode.
(get-indent stream)
(set-indent stream new-indent)
(inc-indent stream indent-delta)
(inc-indent-abs stream indent-delta)
These functions manipulate the indentation value of the stream. The indentation takes effect the next time a character is output following a newline character.
The get-indent function retrieves the current indentation amount.
The set-indent function sets stream's indentation to the value new-indent and returns the previous value. Negative values are clamped to zero.
The inc-indent function sets stream's indentation relative to the current printing column position, and returns the old value. The indentation is calculated by adding indent-delta to the current column position. If a negative indentation results, it is clamped to zero.
The inc-indent-abs function sets stream's indentation relative to the current indentation value. The indentation is calculated by adding indent-delta to the current indentation amount. If a negative indentation results, it is clamped to zero.
(width-check stream alt-char)
The width-check function examines the state of the stream, taking into consideration the current printing column position, the indentation state, the indentation amount and an internal "force break" flag. It makes a decision either to introduce a line break by printing a newline character, or else to print the alt-char character.
If a decision is made not to emit a line break, but alt-char is nil, then the function has no effect at all.
The return value is t if the function has issued a line break, otherwise nil.
(force-break stream)
If the force-break function is called on a stream, it sets an internal "force break" flag which affects the future behavior of width-check. The width-check function examines this flag. If the flag is set, width-check clears it, and issues a line break without considering any other conditions.
The stream's force-break flag is also cleared whenever a newline character is output.
The force-break function returns stream.
Note: the force-break is involved in line breaking decisions. Whenever a list or list-like syntax is being printed, whenever an element of that syntax is broken into multiple lines, a break is forced after that element, in order to avoid output which resembles the following diagonally-creeping pattern:
(a b c (d e f
g h i) j (k l
m n) o)
but instead is rendered in a more horizontally compact pattern:
(a b c (d e f
g h i)
j (k l
m n)
o)
When the printer prints (d e f g h i) it uses the width-check function between the elements; that function issues the break between the f and g. The printer monitors the return value of width-check; it knows that since one of the calls returned t, the object had been broken into two or more lines. It then calls force-break after printing the last element i of that object. Then, due to the force flag, the outer recursion of the printer which is printing (a b c ...) will experience a break when it calls width-check before printing j.
Custom print methods defined on structure objects can take advantage of width-check and force-break in the same way so that user-defined output integrates with the formatting algorithm.
Streams have two properties which are used by the TXR Lisp object printer to optionally truncate the output generated by aggregate objects.
A stream can specify a maximum length for aggregate objects via the set-max-length function. Using the set-max-depth function, the maximum depth can also be specified.
This feature is useful when diagnostic output is being produced, and the objects involved are so large that the diagnostic output overwhelms the output device or the user, so as to become uninformative. Output limiting also prevents the printer's non-termination on infinite, lazy structures.
It is recommended that functions which operate on streams passed in as parameters save and restore these parameters, if they need to manipulate them, for instance using with-resources:
(defun output-function (arg stream)
;; temporarily impose maximum width and depth
(with-resources ((ml (set-max-length stream 42)
(set-max-length stream ml))
(mw (set-max-depth stream 12)
(set-max-depth stream mw)))
(prinl arg stream)
...))
(set-max-length stream value)
The set-max-length function establishes the maximum length for aggregate object printing. It affects the printing of lists, vectors, hash tables, strings as well as quasiliterals and quasiword list literals (QLLs).
The default value is 0 and this value means that no limit is imposed. Otherwise, the value must be a positive integer.
When the list, vector or hash-table object being printed has more elements than the maximum length, then elements are printed only up to the maximum count, and then the remaining elements are summarized by printing the ... (three dots) character sequence as if it were an additional element. This sequence is an invalid token; it cannot be read as input.
When a character string is printed, and the maximum length parameter is nonzero, a maximum character count is determined as follows. Firstly, if the maximum length value is less than 3, it is taken to be 3. Then it is multiplied by 8. Thus, a maximum length of 10 allows 80 characters, whereas a maximum length of 1 allows 24 characters.
If a string which exceeds the maximum number of characters is being printed with read-print consistency, as by the print function, then only a prefix of the string is printed, limited to the maximum number of characters. Then, the literal syntax is closed using the character sequence \..." (backslash, dot, dot, dot, double quote) whose leading invalid escape sequence \. (backslash, dot) ensures that the truncated object is not readable.
If a string which exceeds the maximum number of characters is being printed without read-print consistency, as by the pprint function, then only a prefix of the string is printed, limited to the maximum number of characters. Then the character sequence ... is emitted.
Quasiliterals are treated using a combination of behaviors. Elements of a quasiliteral are literal sequence of text, and embedded variables and expressions. The maximum length specifies both the maximum number of elements in the quasiliteral, and the maximum number of characters in any element which is a sequence of text. When either limit is exceeded, the quasiliteral is immediately terminated with the sequence \...` (escaped dot, dot, dot, backtick). The maximum character limit is applied to the units of text cumulatively, rather than individually. As in the case of string literals, the limit is determined by multiplying the length by 8, and clamping at a minimum value of 24.
When a QLL is printed, the space-separated elements of the literal are individually subject to the maximum character limit as if they were independent quasiliterals. Furthermore, the sequence of these elements is subject to the maximum length. If there are more elements in the QLL, then the sequence \...` (escaped dot, dot, dot, backtick) is emitted and thus the QLL ends.
The set-max-length function returns the previous value.
(set-max-depth stream value)
The set-max-length function establishes the maximum depth for the printing of nested objects. It affects the printing of lists, vectors, hash tables and structures. The default value is 0 and this value means that no limit is imposed. Otherwise, the value must be a positive integer.
The depth of an object not enclosed in any object is zero. The depth of the element of an aggregate is one greater than the depth of the aggregate itself. For instance, given the list (1 (2 3)) the list itself has depth 0, the atom 1 has depth 1, as does the sublist (2 3), and the 2 and 3 atoms have depth 2.
When an object is printed whose depth exceeds the maximum depth, then three dot character sequence ... is printed instead of that object. This notation is an invalid token; it cannot be read as input.
Additionally, when a vector, list, hash table or structure is printed which itself doesn't exceed the maximum depth, but whose elements do exceed, then that object is summarized, respectively, as (...), #(...), H#(...) and S#(...), rather than repeating the ... sequence for each of its elements.
The set-max-depth function returns the previous value.
(open-command system-command [mode-string])
(open-process program mode-string [argument-list])
(open-subprocess program mode-string
[argument-list [function]])
These functions spawn external programs which execute concurrently with the TXR program. They all return a unidirectional stream for communicating with these programs: either an output stream, or an input stream, depending on the contents of mode-string.
In open-command, the mode-string argument is optional, defaulting to the value "r" if it is missing. See the open-file function for a discussion of modes. The open-command function is implemented using POSIX popen. Those elements of mode-string which are applicable to popen are passed to it, and hence their semantics follows from their processing in that function.
The open-command function accepts, via the system-command string parameter, a system command, which is in a system-dependent syntax. On a POSIX system, this would be in the POSIX Shell Command Language.
The open-process function specifies a program to invoke via the command argument. This is subject to the operating system's search strategy. On POSIX systems, if it is an absolute or relative path, it is treated as such, but if it is a simple base name, then it is subject to searching via the components of the PATH environment variable. If open-process is not able to find program, or is otherwise unable to execute the program, the child process will exit, using the value of the C variable errno as its exit status. This value can be retrieved via close-stream.
The argument-list argument is a list of strings which specifies additional optional arguments to be passed to the program. The program argument becomes the first argument, and argument-list becomes the second and subsequent arguments. If argument-list is omitted, it defaults to empty.
If a coprocess is open for writing (mode-string is specified as "w"), then writing on the returned stream feeds input to that program's standard input file descriptor. Indicating the end of input is performed by closing the stream.
If a coprocess is open for reading (mode-string is specified as "r"), then the program's output can be gathered by reading from the returned stream. When the program finishes output, it will close the stream, which can be detected as normal end of data.
The standard input and error file descriptors of an input coprocess are obtained from the streams stored in the *stdin* and *stderr* special variables, respectively. Similarly, the standard output and error file descriptors of an output coprocess are obtained from the *stdout* and *stderr* special variables. These variables must contain streams on which the fileno function is meaningful, otherwise the operation will fail. What this functionality means is that rebinding the special variables for standard streams has the effect of redirection. For example, the following two expressions achieve the same effect of creating a stream which reads the output of the cat program, which reads and produces the contents of the file text-file.
;; redirect input by rebinding *stdin*
(let ((*stdin* (open-file "text-file")))
(open-command "cat"))
;; redirect input using POSIX shell redirection syntax
(open-command "cat < text-file")
The following is erroneous:
;; (let ((*stdin* (make-string-input-stream "abc")))
(open-command "cat"))
A string input or output stream doesn't have an operating system file descriptor; it cannot be passed to a coprocess.
The streams *stdin*, *stdout* and *stderr* are not synchronized with their underlying file descriptors prior to the execution of a coprocess. It is up to the program to ensure that previous output to *stdout* or *stderr* is flushed, so that the output of the coprocess isn't reordered with regard to output produced by the program. Similarly, input buffered in *stdin* is not available to the coprocess, even though it has not yet been read by the program. The program is responsible for preventing this situation also.
If a coprocess terminates abnormally or unsuccessfully, an exception is raised.
The mode-string argument of open-process supports a special redirection syntax. This syntax specifies I/O redirections which are done in the context of the child process, before the specified program is executed. Instances of the syntax are considered options; if mode-string specifies a mode such as r that mode must precede the redirections. Redirections may be mixed with other options.
Up to four redirections may be specified using one of two forms: a short form or the long form. If more than four redirections are specified, the mode-string is considered ill-formed.
The short form of the syntax consists of three characters: the prefix character >, a single decimal digit indicating the file descriptor to be redirected, and then a third character which is either another digit, or else one of the two characters n or x. If the third character is a digit, it indicates the target file descriptor of the redirection. For instance >21 indicates that file descriptor 2 is to be redirected to 1 (so that material written to standard error goes to the same destination as that written to standard output). If the third character is n, it means that the file descriptor will be redirected to the file /dev/null. For instance, >2n indicates that descriptor 2 (standard error) will be redirected to the null device. If the third character is x, it indicates that the file descriptor shall be closed. For instance >0x means to close descriptor 0 (standard input).
The long form of the syntax allows file descriptors that require more than one decimal digit. It consists of the same prefix character > which is immediately followed by an open parenthesis (. The parenthesis is immediately followed by one or more digits which give the to-be-redirected file descriptor. This is followed by one or more whitespace characters, and then either another multi-digit decimal file descriptor or one of the two letters n or x. This second element must be immediately followed by the closing parenthesis ). Thus >21 and >2n may be written in the long form, respectively, as >(2 1) and >(2 n), while >(32 47) has no short form equivalent. Multiple redirections may be specified, in any mixture of the long and short form. For instance r>21>0n>(27 31) specifies a process pipe that is open for reading, capturing the output of the process. In that process, standard error is redirected to standard output, standard input is connected to the null device, and descriptor 27 is redirected to descriptor 31.
The mode-string argument of open-process also supports a special ?fdno syntax. This syntax specifies an alternative file descriptor in the process to which the returned stream should be connected. By default, when the process is opened for writing, its standard output descriptor 1 is used, and when it is opened for reading, its standard input descriptor 0 is used. This option overrides the choice of descriptor. The fdno portion of the syntax must be a sequence of decimal digits, immediately following the ? character. For example, the mode string "?2" specifies that the process is to be open for input, such that the input stream captures the standard error output of that process. In this situation, the standard output will not be captured; it remains unredirected.
The open-subprocess function is a variant of open-process. This function has all the same argument conventions and semantics as open-process, adding the function argument. If this argument isn't nil, then it must specify a function which can be called with no arguments. This function is called in the child process after any redirections are established, just before the program specified by the program argument is executed. Moreover, the open-subprocess function allows program to be specified as nil in which case function must be specified. When function returns, the child process terminates as if by a call to exit* with an argument of zero.
(map-command-lines cmd lines [mode-opts])
(map-command-str cmd str [mode-opts])
(map-command-buf cmd buf [pos [bytes [skip]]]]])
The map-command-lines, map-command-str and map-command-buf functions filter data through an external command.
The cmd parameter has the same meaning as the corresponding parameter in the open-command function. The command open with the "w" mode, which is implied.
The mode-opts optional argument, if present, specifies extra mode options, which must be compatible with w.
The lines argument in map-command-lines must be a sequence of strings. These strings are transmitted to the command as newline-terminated lines, as if by the put-lines function. Simultaneously, the output of the command is read and divided into lines as if by the get-lines function. The entire output of the command is read before the function terminates, and the list of lines is returned.
Similarly, the str argument in map-command-str is transmitted to the executing command as its complete input, as if by put-string. Simultaneously, the output of the command is captured as a single string, as if using the get-string function. That string is returned.
The buf argument in map-command-buf must be a buffer. The bytes of the buffer are transmitted to the executing command, whose output bytes are gathered into a new buffer object which is returned. The optional pos argument, which defaults to zero, specifies the starting position within buf. Bytes from that position to the end of the buffer are transmitted to the command. The optional bytes argument specifies a limit on the number of bytes of the command's output that should be accumulated into a buffer. The default is unlimited. The optional skip argument, defaulting to zero, specifies how many initial bytes of the command's output must be discarded prior to reading the bytes that are to be accumulated.
(map-process-lines program args lines [mode-opts])
(map-process-str program args str [mode-opts])
(map-process-buf program args buf
>> [ pos >> [ bytes <> [ skip ]]]]])
The map-process-lines, map-process-str and map-process-buf are counterparts to map-command-lines, map-command-str and map-command-buf which specify the external process differently.
Instead of the cmd parameter, these functions feature a pair of parameters program and args which have the same semantics as the program and argument-list parameters of open-process.
Thus the relationship between these groups of three functions is like that between open-command and open-process.
In all other regards, these functions are identical to their counterparts.
The functions in this group create a stream, perform an I/O operation on it, and ensure that it is closed, in one convenient operation. They operate on files or command streams.
Several other functions in this category exist, which operate with buffers. They are documented in the Buffer Functions subsection under the FOREIGN FUNCTION INTERFACE section.
Many of the functions described in this section take an optional mode-opts argument. If this is specified, it must be a string which follows the options portion of the mode-string syntax described for the open-file function. This string must not specify the mode part. If specified, the mode-opts must be compatible with the implied mode. Functions that write a file have an implied mode of "w", those which append have an implied mode of "a", and those which read have an implied mode of "r". For instance, a mode-opts value of "x" is useful with file-put-string but not file-get-string,
(file-get name [mode-opts])
(file-get-string name [mode-opts])
(file-get-lines name [mode-opts])
The file-get function opens a text stream over the file indicated by the string argument name for reading, reads the printed representation of a TXR Lisp object from it, and returns that object, ensuring that the stream is closed.
The file-get-string is similar to file-get except that it reads the entire file as a text stream and returns its contents in a single character string.
The file-get-lines function opens a text stream over the file indicated by name and returns produces a lazy list of strings representing the lines of text of that file as if by a call to the get-lines function, and returns that list. The stream remains open until the list is consumed to the end, as indicated in the description of get-lines.
(file-put name obj [mode-opts])
(file-put-string name string [mode-opts])
(file-put-lines name list [mode-opts])
The file-put, file-put-string and file-put-lines functions open a text stream over the file indicated by the string argument name, write the argument object into the file in their specific manner, and then close the file.
If the file doesn't exist, it is created. If it exists, it is truncated to zero length and overwritten.
The file-put function writes a printed representation of obj using the prinl function. The return value is that of prinl.
The file-put-string function writes string to the stream using the put-string function. The return value is that of put-string.
The file-put-lines function writes list to the stream using the put-lines function. The return value is that of put-lines.
(file-append name obj [mode-opts])
(file-append-string name string [mode-opts])
(file-append-lines name list [mode-opts])
The file-append, file-append-string and file-append-lines functions open a text stream over the file indicated by the string argument name, write the argument object into the stream in their specific manner, and then close the stream.
These functions are close counterparts of, respectively, file-get, file-get-string and file-get-lines.
These functions behave differently when the indicated file already exists. Rather than being truncated and overwritten, the file is extended by appending the new data to its end.
(file-get-objects name [ mode-opts [error-stream]])
The file-get-objects function opens an input text stream over the file indicated by the name argument, which is a string.
All Lisp objects are read from the stream. Parse errors are reported to error-stream which defaults to *stdnull* (error output is discarded).
If there is a parse error, the function throws an exception, otherwise the list of parsed objects is returned.
(file-put-objects name seq [mode-opts])
The functions file-put-objects and file-append-objects open a text stream over the file indicated by the string argument name, and write each of the objects contained in sequence seq into the stream as if using the prinl function on each individual element of seq.
The file-put-objects function opens the file using the "w" mode, which overwrites the file if it exists, whereas file-append-objects uses "a", which appends to the file.
(command-get cmd [mode-opts])
(command-get-string cmd [mode-opts])
(command-get-lines cmd [mode-opts])
The command-get function opens text stream over an input command pipe created for the command string cmd, as if by the open-command function. It reads the printed representation of a TXR Lisp object from it, and returns that object, ensuring that the stream is closed.
The command-get-string is similar to command-get except that it reads the entire file as a text stream and returns its contents in a single character string.
The command-get-lines function opens a text stream over an input command pipe created for the command string cmd and returns produces a lazy list of strings representing the lines of text of that file as if by a call to the get-lines function, and returns that list. The stream remains open until the list is consumed to the end, as indicated in the description of get-lines.
(command-put cmd obj [mode-opts])
(command-put-string cmd string [mode-opts])
(command-put-lines cmd list [mode-opts])
The command-put, command-put-string and command-put-lines functions open an output text stream over an output command pipe created for the command specified in the string argument cmd, as if by the open-command function. They write the argument object into the stream in their specific manner, and then close the stream.
The command-put function writes a printed representation of obj using the prinl function. The return value is that of prinl.
The command-put-string function writes string to the stream using the put-string function. The return value is that of put-string.
The command-put-lines function writes list to the stream using the put-lines function. The return value is that of put-lines.
A stream type exists which allows buf objects to be manipulated through the stream interface. A buffer stream is created using the make-buf-stream function, which can either attach the stream to an existing buffer, or create a new buffer that can later be retrieved from the stream using get-buf-from-stream.
Operations on the buffer stream treat the underlying buffer much like if it were a memory-based file. Unless the underlying buffer is a "borrowed buffer" referencing the storage belonging to another object (such as the buffer object produced by the buf-d FFI type's get semantics) the stream operations can change the buffer's size. Seeking beyond the end of the buffer and then writing one or more bytes extends the buffer's length, filling the newly allocated area with zero bytes. The truncate-stream function is supported also. Buffer streams also support the :byte-oriented property.
Macros with-out-buf-stream and with-in-buf-stream are provided to simplify the steps involved in using buffer streams in some common scenarios. Note that in spite of the naming of these macros there is only one buffer stream type, which supports bidirectional I/O.
(make-buf-stream [buf])
The make-buf-stream function return a new buffer stream. If the buf argument is supplied, it must be a buf object. The stream is then associated with this object. If the argument is omitted, a buffer of length zero is created and associated with the stream.
(get-buf-from-stream buf-stream)
The get-buf-from-stream returns the buffer object associated with buf-stream which must be a buffer stream.
(with-out-buf-stream (var [buf-expr])
body-form*)
(with-in-buf-stream (var buf-expr)
body-form*)
The with-out-buf-stream and with-in-buf-stream macros both bind variable var to an implicitly created buffer stream, and evaluate zero or more body-forms in the environment where the variable is visible.
The buf-expr argument, which may be omitted in the use of the with-out-buf-stream macro, must be an expression which evaluates to a buf object.
The var argument must be a symbol suitable for naming a variable.
The implicitly allocated buffer stream is connected to the buffer specified by buf-expr or, when buf-expr is omitted, to a newly allocated buffer.
The code generated by the with-out-buf-stream macro, if it terminates normally, yields the buffer object as its result value.
The with-in-buf-stream returns the value of the last body-form, or else nil if no forms are specified.
(with-out-buf-stream (*stdout* (make-buf 24))
(put-string "Hello, world!"))
-> #b'48656c6c6f2c2077 6f726c6421000000 0000000000000000'
(with-out-buf-stream (*stdout*) (put-string "Hello, world!"))
-> #b'48656c6c6f2c2077 6f726c6421'
Objects of type cptr are Lisp values which contain a foreign pointer ("C pointer"). This data type is used by the dlopen function and is generally useful in conjunction with the Foreign Function Interface (FFI). An arbitrary pointer emanating from a foreign function can be captured as a cptr value, which can be passed back into foreign code. For this purpose, there exits also a matching FFI type called cptr.
The cptr type supports a symbolic type tag, which defaults to nil. The type tag plays a role in FFI. The FFI cptr type supports a tag attribute. When a cptr object is converted to a foreign pointer under the control of the FFI type, and that FFI type has a tag other than nil, the object's tag must exactly match that of the FFI type, or the conversion throws an error. In the reverse direction, when a foreign pointer is converted to a cptr object under control of the FFI cptr type, the object inherits the type tag from the FFI type.
Although cptr objects are conceptually non-aggregate values, corresponding to pointers, they are de facto aggregates due to their implementation as references to heap objects. When a cptr object is passed to a foreign function by pointer, for instance using a parameter of type (ptr cptr), its internal pointer is potentially updated to the new value coming from the function.
(cptr-int integer [type-symbol])
The cptr-int function converts integer into a pointer in a system-specific way which is consistent with the system's addressing structure. Then it returns that pointer contained in a cptr object.
The integer argument must be an integer which is in range for a pointer value. Note: this range is wider than the fixnum range; a portion of the range of bignum integers can denote pointers.
An extended range of values is accepted. The entire addressable space may be expressed by non-negative values. A range of negative values also expresses a portion of the address space, in accordance with the platform's concept of a signed integer.
For instance, on a system with 32-bit addresses, the values 0 to 4294967295 express all of the addresses as a pure binary value. Furthermore, the values -2147483648 to -1 also express the upper part of this range, corresponding, respectively, to the addresses 2147483648 to 4294967295. On that platform, values of integer outside of the range -2147483648 to 4294967295 are invalid.
The type-symbol argument should be a symbol. If omitted, it defaults to nil. This symbol becomes the cptr object's type tag.
(cptr-obj object [type-symbol])
The cptr-obj function converts object object directly to a cptr.
The object argument may be of any type.
The raw representation of object is simply stored in a new instance of cptr and returned.
The type-symbol argument should be a symbol. If omitted, it defaults to nil. This symbol becomes the cptr object's type tag.
The lifetime of the returned cptr object is independent from that of object. If the lifetime of object reaches its end before that of the cptr, the pointer stored inside the cptr becomes invalid.
(int-cptr cptr)
The int-cptr function retrieves the pointer value of the cptr object as an integer.
If an integer n is in a range convertible to cptr type, then the expression (int-cptr (cptr-int n)) reproduces n.
(cptr-buf buf [type-symbol])
The cptr-buf function returns a cptr object which holds a pointer to a buffer object's storage area. The buf argument must be of type buf.
The type-symbol argument should be a symbol. If omitted, it defaults to nil. This symbol becomes the cptr object's type tag.
The lifetime of the returned cptr object is independent from that of buf. If the lifetime of buf reaches its end before that of the cptr, the pointer stored inside the cptr becomes invalid.
(cptr-cast type-symbol cptr)
The cptr-cast function produces a new cptr object which has the same pointer as cptr but whose type is given by type-symbol.
Casting cptr objects with cptr-cast circumvents the safety mechanism which cptr type tagging provides.
(cptr-copy cptr)
The copy-cptr function creates a new cptr object similar to cptr, which has the same address and type symbol as cptr.
(cptr-zap cptr)
The cptr-zap function changes the pointer value of the cptr object to the null pointer.
The cptr argument must be of cptr type.
The return value is cptr itself.
Note: it is recommended to use cptr-zap when the program has taken some action which invalidates the pointer value stored in a cptr object, where a risk exists that the value may be subsequently misused.
(cptr-free cptr)
The cptr-free function passes the cptr object's pointer to the C library free function. After this action, it behaves exactly like cptr-zap.
The cptr argument must be of cptr type.
The return value is cptr itself.
Note: this function is unsafe. If the pointer didn't originate from the malloc family of memory allocation functions, or has already been freed, or copies of the pointer exist which are still in use, the consequences are likely catastrophic.
(cptrp value)
The cptrp function tests whether value is a cptr. It returns t if this is the case, nil otherwise.
(cptr-type cptr)
The cptr-type function retrieves the cptr object's type tag.
(cptr-get cptr [type])
The cptr-get function extracts a Lisp value by converting a C object at the memory location denoted by cptr, according to the FFI type type. The external representation at the specified memory location is is scanned according to the type and converted to a Lisp value which is returned.
If the type argument is specified, it must be a FFI type object. If omitted, then the cptr object's type tag is interpreted as a FFI type symbol and resolved to a type; the resulting type, if one is found is substituted for type. If the lookup fails an error exception is thrown.
The cptr object must be of type cptr and point to a memory area suitably aligned for, and large enough to hold a foreign representation of type, at the byte offset indicated by the offset argument.
If cptr is a null pointer, an exception is thrown.
The cptr-get operation is similar to the "get semantics" performed by FFI in order to extract the return value of foreign function calls, and by the FFI callback mechanism to extract the arguments coming into a callback.
The type argument may not be a variable length type, such as an array of unspecified size.
Note: the functions cptr-get and cptr-out are useful in simplifying the interaction with "semi-opaque" foreign objects: objects which serve as API handles that are treated as opaque pointers in API argument calls, but which expose some internal members that the application must access directly. The cptr objects pass through the foreign API without undergoing conversion, as usual. The application uses these two functions to perform conversion as necessary. Under this technique, the description of the foreign object need not be complete. Structure members which occur after the last member that the application is interested in need not be described in the FFI type.
(cptr-out cptr obj [type])
The cptr-out function converts a Lisp value into a C representation, which is stored at the memory location denoted by cptr, according to the FFI type type. The function's return value is obj.
If the type argument is specified, it must be a FFI type object. If omitted, then the cptr object's type tag is interpreted as a FFI type symbol and resolved to a type; the resulting type, if one is found is substituted for type. If the lookup fails an error exception is thrown.
The obj argument must be an object compatible with the conversions implied by type.
The cptr object must be of type cptr and point to a memory area suitably aligned for, and large enough to hold a foreign representation of type, at the byte offset indicated by the offset argument.
If cptr is a null pointer, an exception is thrown.
It is assumed that obj is an object which was returned by an earlier call to cptr-get, and that the cptr and type arguments are the same objects that were used in that call.
The cptr-out function performs the "out semantics" encoding action, similar to the treatment applied to the arguments of a callback prior to returning to foreign code.
The cptr-null variable holds a null pointer as a cptr instance.
Two cptr objects may be compared for equality using the equal function, which tests whether their pointers are equal.
The cptr-null variable compares equal to values which have been subject to cptr-zap or cptr-free.
A null cptr may be produced by the expression (cptr-obj nil); however, this creates a freshly allocated object on each evaluation.
The expression (cptr-int 0) also produces a null pointer on all platforms where TXR is found.
(cptr-size-hint cptr bytes)
The cptr-size-hint function indicates to the garbage collector that the given cptr object is associated with bytes of foreign memory that are otherwise invisible to the garbage collector.
Note: this function should be used if the foreign memory is indirectly managed by the cptr object in cooperation with the garbage collector. Specifically, cptr should have a finalizer registered against it which will liberate the foreign memory.
In TXR Lisp, stream objects aren't structure types, and therefore lie outside of the object-oriented programming system. However, TXR Lisp supports a delegation mechanism which allows a structure which provides certain methods to be used as a stream.
The function make-struct-delegate-stream takes as an argument the instance of a structure, which is referred to as the stream interface object. The function returns a stream object such that when stream operations are invoked on this stream, it delegates these operations to methods of the stream interface object.
A structure type called stream-wrap is provided, whose instances can serve as stream interface objects. This structure has a slot called stream which holds a stream, and it provides all of the methods required for the delegation mechanism used by make-struct-delegate-stream. This stream-wrap operations simply invoke the ordinary stream operations on the stream slot. The stream-wrap type can be used as a base class for a derived class which intercepts certain operations on a stream (by defining the corresponding methods) while allowing other operations to transparently pass to the stream (via the base methods inherited from stream-wrap).
(make-struct-delegate-stream object)
The make-struct-delegate-stream function returns a stream whose operations depend on the object, a stream interface object.
The object argument must be a structure which implements certain subsets of, or all of, the following methods: put-string, put-char, put-byte, get-line, get-char, get-byte, unget-char, unget-byte, put-buf, fill-buf, close, flush, seek, truncate, get-prop, set-prop, get-error, get-error-str, clear-error and get-fd.
Implementing get-prop is mandatory, and that method must support the :name property.
Failure to implement some of the other methods will impair the use of certain stream operations on the object.
stream.(put-string str)
The put-string method is implemented on a stream interface object. It should behave in a manner consistent with the description of the put-string stream I/O function.
stream.(put-char chr)
The put-char method is implemented on a stream interface object. It should behave in a manner consistent with the description of the put-char stream I/O function.
stream.(put-byte byte)
The put-byte method is implemented on a stream interface object. It should behave in a manner consistent with the description of the put-byte stream I/O function.
stream.(get-line)
The get-line method is implemented on a stream interface object. It should behave in a manner consistent with the description of the get-line stream I/O function.
stream.(get-char)
The get-char method is implemented on a stream interface object. It should behave in a manner consistent with the description of the get-char stream I/O function.
stream.(get-byte)
The get-byte method is implemented on a stream interface object. It should behave in a manner consistent with the description of the get-byte stream I/O function.
stream.(unget-char chr)
The unget-char method is implemented on a stream interface object. It should behave in a manner consistent with the description of the unget-char stream I/O function.
stream.(unget-byte byte)
The unget-byte method is implemented on a stream interface object. It should behave in a manner consistent with the description of the unget-byte stream I/O function.
stream.(put-buf buf pos)
The put-buf method is implemented on a stream interface object. It should behave in a manner consistent with the description of the put-buf stream I/O function.
Note: there is a severe restriction on the use of the buf argument. The buffer object denoted by the buf argument may be specially allocated and have a lifetime which is scoped to the method invocation. The put-buf method shall not permit the buf object to be used beyond the duration of the method invocation.
stream.(fill-buf buf pos)
The fill-buf method is implemented on a stream interface object. It should behave in a manner consistent with the description of the fill-buf stream I/O function.
Note: there is a severe restriction on the use of the buf argument. The buffer object denoted by the buf argument may be specially allocated and have a lifetime which is scoped to the method invocation. The fill-buf method shall not permit the buf object to be used beyond the duration of the method invocation.
stream.(close throw-on-error-p)
The close method is implemented on a stream interface object. It should behave in a manner consistent with the description of the close-stream stream I/O function.
With two exceptions, the value returned from close is retained by close-stream, such that repeated calls to close-stream then return that value without calling the close method. The exceptions are the values nil and : (the colon symbol). If either of these values is returned, and close-stream is invoked again on the same stream object, the close method will be called again.
Furthermore, if the : symbol is returned by the close method, this indicates a successful close, and the close-stream function returns the t symbol rather than the : symbol.
The rationale for this mechanism is that it supports reference-counted closing. A struct delegate stream may be written which is shared by several owners, which must each call close-stream before the underlying real stream is closed.
stream.(flush offs whence)
The flush method is implemented on a stream interface object. It should behave in a manner consistent with the description of the flush-stream stream I/O function.
stream.(seek offs whence)
The seek method is implemented on a stream interface object. It should behave in a manner consistent with the description of the seek-stream stream I/O function.
stream.(truncate len)
The truncate method is implemented on a stream interface object. It should behave in a manner consistent with the description of the truncate-stream stream I/O function.
stream.(get-prop sym)
The get-prop method is implemented on a stream interface object. It should behave in a manner consistent with the description of the get-prop stream I/O function.
stream.(set-prop sym nval)
The set-prop method is implemented on a stream interface object. It should behave in a manner consistent with the description of the set-prop stream I/O function.
stream.(get-error)
The get-error method is implemented on a stream interface object. It should behave in a manner consistent with the description of the get-error stream I/O function.
stream.(get-error-str)
The get-error-str method is implemented on a stream interface object. It should behave in a manner consistent with the description of the get-error-str stream I/O function.
stream.(clear-error)
The clear-error method is implemented on a stream interface object. It should behave in a manner consistent with the description of the clear-error stream I/O function.
stream.(get-fd)
The get-fd method is implemented on a stream interface object. It should behave in a manner consistent with the description of the fileno stream I/O function.
(defstruct stream-wrap nil
stream
(:method put-string (me str)
(put-string str me.stream))
(:method put-char (me chr)
(put-char chr me.stream))
(:method put-byte (me byte)
(put-byte byte me.stream))
(:method get-line (me)
(get-line me.stream))
(:method get-char (me)
(get-char me.stream))
(:method get-byte (me)
(get-byte me.stream))
(:method unget-char (me chr)
(unget-char chr me.stream))
(:method unget-byte (me byte)
(unget-byte byte me.stream))
(:method put-buf (me buf pos)
(put-buf buf pos me.stream))
(:method fill-buf (me buf pos)
(fill-buf buf pos me.stream))
(:method close (me throw-on-error)
(close-stream me.stream throw-on-error))
(:method flush (me)
(flush-stream me.stream))
(:method seek (me offs whence)
(seek-stream me.stream offs whence))
(:method truncate (me len)
(truncate-stream me.stream len))
(:method get-prop (me sym)
(stream-get-prop me.stream sym))
(:method set-prop (me sym nval)
(stream-set-prop me.stream sym nval))
(:method get-error (me)
(get-error me.stream))
(:method get-error-str (me)
(get-error-str me.stream))
(:method clear-error (me)
(clear-error me.stream))
(:method get-fd (me)
(fileno me.stream)))
The stream-wrap class provides a trivial implementation of a stream interface. It has a single slot, stream which should be initialized with a stream object. Each methods of stream-wrap, shown in its entirety in the above Syntax section, simply invoke the corresponding stream I/O library functions, passing the method arguments, and the value of the stream slot to that function, and consequently returning whatever that function returns.
Note: the stream-wrap function is intended to useful as an inheritance base. A user-defined structure can inherit from stream-wrap and provide its own versions of some of the methods, thereby intercepting those operations to customize the behavior.
For instance, a function equivalent to the record-adapter function could be implemented by constructing an object derived from stream-wrap which overrides the behavior of the get-line method, and then using the make-struct-delegate-stream to return a stream based on this object.
;;; Implementation of my-record-adapter,
;;; a function resembling
;;; the record-adapter implementation
(defstruct rec-input stream-wrap
regex
include-match-p
;; get-line overridden to use regex-based
;; extraction using read-until-match
(:method get-line (me)
(read-until-match me.regex me.stream
me.include-match-p))))
(defun my-record-adapter (regex stream include-match-p)
(let ((recin (new rec-input
stream stream
regex regex
include-match-p include-match-p)))
(make-struct-delegate-stream recin)))
TXR Lisp has a package system inspired by the salient features of ANSI Common Lisp, but substantially simpler.
Each symbol has a name, which is a string.
A package is an object which serves as a container of symbols; the package associates the name strings with symbols.
A symbol which exists inside a package is said to be interned in that package. A symbol can be interned in more than one package.
A symbol may also have a home package. A symbol which has a home packageis always interned in that package.
A symbol which has a home package is called an interned symbol.
A symbol which is interned in one or more packages, but has no home package, is a quasi-interned symbol. When a quasi-interned symbol is printed, if it is not interned in the package currently held in the *package* variable, it will appear in uninterned notation denoted by a #: prefix, even though it is interned in one or more packages. This is because in any situation when a symbol is printed with a package prefix, that prefix corresponds to the name of its home package. The reverse isn't true: when a symbol token is read bearing a package prefix, the token denotes any interned symbol in the indicated package, whether or not the package is the home package of that symbol.
Packages are held in a global list which can be used to search for a package by name. The find-package function performs this lookup. A package may be deleted from the list with the delete-package function, but it continues to exist until the program loses the last reference to that package. When a package is deleted with delete-package, its symbols are uninterned from all other packages.
A symbol existing in one package can be brought into another package via the use-sym function, causing it to be interned in the target package. A symbol which thus exists inside a package which is not its home package is called a foreign symbol, relative to that package. The contrasting term with foreign symbol is local symbol, which refers to a symbol, relative to a package, which is interned in that package and that package is also its home. Every symbol interned in a package is either foreign or local.
An existing symbol can also be brought into a package under a different name using the use-sym-as function, causing it to be interned under an alternative name. This has the effect of creating a local alias for a foreign symbol, and is intended as a renaming mechanism for resolving name clashes.
If a foreign symbol is introduced into a package, and has the same name as an existing local symbol, the local symbol continues to exist, but is hidden: it is not accessible via a name lookup on that package. While hidden, a symbol loses its home package and is thus degraded to either quasi-interned or uninterned status, depending on whether that symbol is interned in other packages.
When a foreign symbol is removed from a package via unuse-sym, then if a hidden symbol exists in that package of the same name, that hidden symbol is reinterned in that package and reacquires that package as its home package, becoming an interned symbol again.
Finally, packages have a fallback package list: a list of associated packages, which may be empty. The fallback package list is manipulated with the functions package-fallback-list and set-package-fallback-list, and with the :fallback clause of the defpackage macro. The fallback package list plays a role only in three situations: one in the TXR Lisp parser, one in the printer, and one in the interactive listener. Besides that, two library functions refer to it: intern-fb and find-symbol-fb.
The parser situation involving the fallback list occurs when the TXR Lisp parser resolves an unqualified symbol token: a symbol token not carrying a package prefix. Such a symbol name is resolved against the current package (the package currently stored in the *package* special variable). If a symbol matching the token is not found in the current package, then the packages in its fallback package list are searched for the symbol. The first matching symbol which is found in the fallback list is returned. If no matching symbol is found in the fallback list, then the token is interned as a new symbol is interned in the current package. The packages in the current package's fallback list may themselves have fallback lists. Those fallback lists are not involved; no such recursion takes place.
The printer situation involving the fallback list is as follows. If a symbol is being printed in a machine-readable way (not "pretty"), has a home package and is not a keyword symbol, then a search takes place through the current package first and then its fallback list. If the symbol is found anywhere in that sequence of locations, and is not occluded by a same-named symbol occurring earlier in that sequence, then the symbol is printed without a package prefix.
The listener situation involving the fallback list is a follows. When tab completion is used on a symbol without a package prefix, the listener searches for completions not only in the current package, but in the fallback list also.
The TXR Lisp package system doesn't support the ANSI Common Lisp concept of package use, replacing that concept with fallback packages.
Though the use-package and unuse-package functions exist and are similar to the ones in ANSI CL, they actually operate on individual foreign symbols, bringing them in or removing them, respectively. These functions effectively iterate over the local symbols of the used or unused package, and invoke use-sym or unuse-sym, respectively.
The TXR Lisp package system consequently doesn't support the concept of shadowing symbols, and conflicts do not exist. When a foreign symbol is introduced into a package which already has a symbol by that name, that symbol is silently removed from that package if it is itself foreign, or else hidden if it is local.
The TXR Lisp package system also doesn't feature the concept of internal and external symbols. The rationale is that this distinction divides symbols into subsets in a redundant way. Packages are already subsets of symbols. A module can use two packages to simulate private symbols. An example of this is given in the Package Examples section below.
The TXR Lisp fallback package list mechanism resembles ANSI CL package use, and satisfies similar use scenarios. However, this mechanism does not cause a symbol to be considered visible in a package. If a package foo contains no symbol bar, but one of the packages in foo's fallback list does contain bar, that symbol is nevertheless not considered visible in foo. The syntax foo:bar will not resolve. The fallback mechanism only comes into play when a package is installed as the current package in the *package* variable. It then allows unqualified symbol references to refer across the fallback list.
The TXR Lisp package system does not feature package nicknames, which have been found to be a source of clashes in large Common Lisp software collections, leading to the development of a feature called package local nicknames that is not part of ANSI CL, but supported by a number of implementations. In TXR Lisp, packages have only one name, accessible via package-name. TXR Lisp packages are held in an association list called *package-alist*, which is public, which associates string names with packages. The function find-package which is used by the parser when looking up the package prefix of a qualified symbol, only uses the names which appears as keys in this association list. Usually those names are the same as the names of the package objects. However, it's possible to manipulate this association list to create alias names for packages. Thus, it is possible for (find-package "foo") to return #<package: bar> if the name "foo" is associated, in *package-alist* with a package object named "bar".
The TXR Lisp package system doesn't feature package local nicknames. There are three reasons for this. One is that it doesn't have global package nicknames. The second is that the mechanism would be cumbersome, and add delay to the resolution of qualified symbols, requiring nicknames in the *package* to be searched for a package name, in addition to the dynamic *package-alist*. The third reason is that package local nicknames do not actually solve the problem of clashing symbols, when an application uses multiple packages that each define a symbol by the same name. Package nicknames only shorten the qualified names required to refer to the symbols, Instead, TXR Lisp allows a foreign symbol to be interned in a package under a name which is different from its symbol-name. Thus, rather than creating aliases for package names, TXR Lisp packages can locally rename the actual clashing symbols, which can then be referenced by unqualified names.
By manipulating *package-alist*, a TXR Lisp source file can nevertheless achieve the creation of a de facto package nickname, which is local to a loaded file, by following the following example:
;; make sure that when this file finishes loading,
;; or the loading is interrupted by an exception,
;; the "u" package alias is deleted from *package-alist*
(push-after-load
(set *package-alist* [remqual "u" *package-alist* car]))
;; push an alias named u for the usr package.
(push (cons "u" (find-package "usr")) *package-alist*)
;; u: can now be used, until the end of this file
(u:prinl (u:list 1 2 3))
;; Define three packages.
(defpackage mod-priv
(:fallback usr))
(defpackage mod)
(defpackage client
(:fallback mod usr)
(:use-from mod-priv other-priv))
;; Switch to mod-priv package
(in-package mod-priv)
(defun priv-fun (arg)
(list arg))
;; Another function with a name in the mod-priv package.
(defun other-priv (arg)
(cons arg arg))
;; Define a function in mod; a public function.
;; Note that we don't have to change to the mod package,
;; to define functions with names in that package.
;; We rely on interning being allowed for the qualified
;; mod:public-fun syntax.
(defun mod:public-fun (arg)
(priv-fun arg)) ;; priv-fun here is mod-priv:priv-fun
;; Switch to client package
(in-package client)
(priv-fun) ;; ERROR: refers to client:priv-fun, not defined
(mod:priv-fun) ;; ERROR: mod-priv:priv-fun not used in mod
(mod-priv:priv-fun 3) ;; OK: direct reference via qualifier
(public-fun 3) ;; OK: mod:public-fun symbol via fallback
(other-priv 3) ;; OK: foreign symbol mod-priv:other-priv
;; present in client due to :use-from
The following example shows how to create a package called custom in which the + symbol from the usr package is replaced with a local symbol. A function is then defined using the local symbol, which allows strings to be catenated with +:
(defpackage custom
(:fallback usr)
(:local + - * /))
(defmacro outside-macro (x) ^(+ ,x 42))
(in-package custom)
(defun binary-+ (: (left 0) (right 0))
(if (and (numberp left) (numberp right))
(usr:+ left right)
`@left@right`))
(defun + (. args)
[reduce-left binary-+ args])
(+) -> 0
(+ 1) -> 1
(+ 1 "a") -> "1a"
(+ 1 2) -> 3
(+ "a") -> "a"
(+ "a" "b" "c") -> "abc"
;; macro expansions using usr:+ are not affected
(outside-macro "a") -> ;; error: + invalid operands "a" 42
On the other hand, some directives are not this way. For instance the @(bind ...), syntax is processed as a true Lisp expression, in which the bind token is subject to the usual rules for interning a symbol, sensitive to *package* in the usual way.
The following notes describe the treatment of "special" directives that are involved in phrase structure syntax. It applies to all directives which head off a block that must be terminated by @(end), all "punctuation" directives like @(and) or @(end) and all subphrase indicators like @(last) or @(elif).
Firstly, each such directive may have a package prefix on its main symbol, yet is still recognized as the same token. That is to say, @(foo:collect) is still treated by the tokenizer and parser as the @(collect) token, regardless of the package prefix, and regardless of whether foo:end is the same symbol as the usr:end symbol.
However, this doesn't mean that any foo:collect is allowed to denote the collect directive.
A qualified symbol such as foo:collect must correspond to (be the same object as) precisely one of two symbols: either the same-named symbol in the usr package, or else the same-named symbol in the keyword package. If this condition isn't satisfied, the situation is a syntax error. Note that this check uses the original usr and keyword packages, not the packages which are currently named "usr" or "keyword" in the current *package-alist*.
A check is also performed for an unqualified symbol. An unqualified symbol like collect must also resolve, in the context of the current value of the *package* variable, to the same named-symbol in either the original usr or keyword package. Thus if the current package isn't usr, and @(collect) is being processed, the current package must be such that collect resolves to usr:collect. either because that symbol is present in the current pack via import, or else visible via the fallback list.
These rules are designed to approximate what the behavior would be if these directives were actually scanned as Lisp forms in the usual way and then recognized as phrase structure tokens according to the identity of their leading symbol. The additional restriction is added that that the directive symbol names are treated as reserved. If there exists a user-defined pattern function called mypackage:end it may not be invoked using the syntax @(mypackage:end), which is erroneous; though it is invocable indirectly via the @(call) directive.
If specified, the argument may be a character string, which is taken as the name of a package. It may also be a symbol, in which case the symbol's name, which is a character string, is used. Thus the objects :sys, usr:sys, abc:sys and "sys" all refer to the same package, the system package which is named "sys".
A package parameter may also simply be a package object.
Some functions, like use-package and unuse-package functions accept a list of packages as their first argument. This may be a list of objects which follow the above conventions: strings, symbols or package objects. Also, instead of a list, an atom may be passed: a string, symbol or package object. It is treated as a singleton list consisting of that object.
These variables hold predefined packages. The user-package contains all of the public symbols in the TXR Lisp library. The keyword-package holds keyword symbols, which are printed with a leading colon. The system-package is for internal symbols, helping the implementation avoid name clashes with user code in some situations.
These variables shouldn't be modified. If they are modified, the consequences are unspecified.
The names of these packages, respectively, are "usr", "sys", and "keyword".
This variable holds the current package. The global value of this variable is initialized to a package called "pub". The pub package has the usr package in its fallback list; thus when pub is current, all of the usr symbols, comprising the content of the TXR Lisp library, are visible.
All forms read and evaluated from the TXR command line, in the interactive listener, from files via load or compile-file or from the TXR pattern language are processed in this default pub package, unless arrangement are made to change to a different package.
The current package is used as the default package for interning symbol tokens which do not carry the colon-delimited package prefix.
The current package also affects printing. When a symbol is printed whose home package matches the current package, it is printed without a package prefix. (Keyword symbols are always printed with the colon prefix, even if the keyword package is current.)
(make-sym name)
The make-sym function creates and returns a new symbol object. The argument name, which must be a string, specifies the name of the symbol. The symbol does not belong to any package (it is said to be "uninterned").
Note: an uninterned symbol can be interned into a package with the rehome-sym function. Also see the intern function.
(gensym [prefix])
The gensym function is similar to make-sym. It creates and returns a new symbol object. If the prefix argument is omitted, it defaults to "g".
The difference between gensym and make-sym is that gensym creates the symbol's name by combining the prefix with a numeric suffix. The suffix is obtained by incrementing the *gensym-counter* and taking the new value. The name string then calculated from the prefix and the counter value as if by evaluating a form similar to (fmt "~a~,04d" prefix counter). From this it can be inferred that prefix can be an object of any kind.
Note: the generated symbol's name, though varying thanks to the incrementing counter, is not the basis of its uniqueness. The basis of the symbol's uniqueness is that it is a freshly created object, distinct from any other object. The related function make-sym still returns unique symbols even if repeatedly called with the same string argument.
This variable is initialized to 0. Each time the gensym function is called, it is incremented. The incremented value forms the basis of the numeric suffix which gensym uses to form the name of the new symbol.
(make-package name [weak])
The make-package function creates and returns a package named name, where name is a string. It is an error if a package by that name exists already. Note: ordinary creation of packages for everyday program modularization should be performed with the defpackage macro rather than by direct use of make-package.
If the weak parameter is given an argument which is a Boolean true, then the resulting package holds symbols weakly, from a garbage collection point of view. If the only reference to a symbol is that which occurs inside the weak package, then that symbol may be removed from the package and reclaimed by the garbage collector.
Note: weak packages address the following problem. The application creates a package for the purpose of reading Lisp data. Symbols occurring in that data therefore are interned into the package. Subsequently, the application retains references to some of the symbols, discarding the others. If the package isn't weak, then because the application is retaining some of the symbols, and those symbols hold a reference to the package, and the package holds a reference to all symbols that were interned in it, all of the symbols are retained. If a weak package is used, then the discarded symbols are eligible for garbage collection.
(delete-package package)
The delete-package breaks the association between a package and its name. After delete-package, the package object continues to exist, but cannot be found using find-package.
Furthermore, delete-package iterates over all remaining packages. For each remaining package p, it performs the semantic action of the (unuse-package package p) expression. That is to say, all of the remaining packages are scrubbed of any foreign symbols which are the local symbols of the deleted package.
(merge-delete-package dst-package [src-package])
The merge-delete-package iterates over all of the local symbols of src-package and rehomes each symbol into dst-package. Then, it deletes src-package.
Note: the local symbols are identified as if using package-local-symbols, rehoming is performed as if using rehome-sym, and deleting src-package is performed as if using delete-package.
(packagep obj)
The packagep function returns t if obj is a package, otherwise it returns nil.
(find-package name)
The argument name should be a string. If a package called name exists, then it is returned. Otherwise nil is returned.
The *package-alist* variable contains the master association list which contains an entry about each existing package.
Each element of the list is a cons cell whose car field is the name of a package and whose cdr is a package object.
Note: the TXR Lisp application can overwrite or rebind this variable to manipulate the active package list. This is useful for sandboxing: safely evaluating code that is obtained as an input from an untrusted source, or calculated from such an input.
The contents of *package-alist* have security implications because textual source code can refer to any symbol in any package by invoking a package prefix. For instance, even if the open function's name is not available in the current package (established by the *package* variable) that symbol can easily be obtained using the syntax usr:open.
However, the entire usr package itself can be removed from *package-alist*. In that situation, the syntax usr:open is no longer valid.
At the same time, selected symbols from the original usr can be nevertheless made available via some intermediate package, which is present in *package-alist* and which contains a subset of the usr symbols that has been curated for safety. That curated package may even be called usr, so that if for instance cons is present in that package, it may be referred to as usr:cons in the usual way.
(package-alist)
The package-alist function retrieves the value of *package-alist*.
Note: this function is obsolescent. There is no reason to use it in new code instead of just accessing *package-alist* directly.
(package-name package)
The package-name function retrieves the name of a package.
(package-symbols [package])
The package-symbols function returns a list of all the symbols which are interned in package.
(package-local-symbols [package])
(package-foreign-symbols [package])
The package-local-symbols function returns a list of all the symbols which are interned in package, and whose home package is that package.
The package-foreign-symbols function returns a list of all the symbols which are interned in package, which do not have that package as their home package, or do not have a home package at all.
The union of the local and foreign symbols contains exactly the same elements as the list returned by package-symbols: the symbols interned in a package are partitioned into local and foreign.
(package-fallback-list package)
(set-package-fallback-list package package-list)
The package-fallback-list returns the current fallback package list associated with package.
The set-package-fallback-list replaces the fallback package list of package with package-list.
The package-list argument must be a list which is a mixture of symbols, strings or package objects. Strings are taken to be package names, which must resolve to existing packages. Symbols are reduced to strings via symbol-name.
(intern name [package])
(intern-fb name [package])
The argument name must be a string. The optional argument package must be a package. If package is not supplied, then the value taken is that of *package*.
The intern function searches package for a symbol called name. If that symbol is found, it is returned. If that symbol is not found, then a new symbol called name is created and inserted into package, and that symbol is returned. In this case, the package becomes the symbol's home package.
The intern-fb function is very similar to intern except that if the symbol is not found in package then the packages listed in the fallback list of package are searched, in order. Only these packages themselves are searched, not their own fallback lists. If a symbol called name is found, the search terminates and that symbol is returned. Only if nothing is found in the fallback list will intern-fb create a new symbol and insert it into package, exactly like intern.
(unintern symbol [package])
The unintern function removes symbol from package.
The nil symbol may not be removed from the usr package; an error exception is thrown in this case.
If symbol isn't nil, then package is searched to determine whether it contains symbol as an interned symbol (either local or foreign), or a hidden symbol.
If symbol is a hidden symbol, then it is removed from the hidden symbol store. Thereafter, even if a same-named foreign symbol is removed from the package via unuse-sym or unuse-package, those operations will no longer restore the hidden symbol to interned status. In this case, unintern returns the hidden symbol that was removed from the hidden store.
If symbol is a foreign symbol, then it is removed from the package. If the package has a hidden symbol of the same name, that hidden symbol is reinterned in the package, and the package once again becomes its home package. In this case, symbol is returned.
If symbol is a local symbol, the symbol is removed from the package. In this case also, symbol is returned.
If symbol is not found in the package as either an interned or hidden symbol, then the function has no effect and returns nil.
(find-symbol name [package [notfound-val]])
(find-symbol-fb name [package [notfound-val]])
The find-symbol and find-symbol-fb functions search package for a symbol called name. That argument must be a character string.
If the package argument is omitted, the parameter defaults to the current value of *package*.
If the symbol is found in package then it is returned.
If the symbol is not found in package, then the function find-symbol-fb also searches the packages listed in the fallback list of package are searched, in order. Only these packages themselves are searched, not their own fallback lists. If a symbol called name is found, the search terminates and that symbol is returned.
The function find-symbol only searches package, ignoring its fallback list.
If a symbol called name isn't found, then these functions return notfound-val is returned, which defaults to nil.
Note: an ambiguous situation exists when notfound-val is a symbol, such as its default value nil, because if that symbol is successfully found, it is indistinguishable from notfound-val.
(rehome-sym symbol [package])
The arguments symbol and package must be a symbol and package object, respectively, and symbol must not be the symbol nil.
The rehome-sym function moves symbol into package. If symbol is already interned in a package, it is first removed from that package.
If a symbol of the same name exists in package, that symbol is first removed from package.
Also, if a symbol of the same name exists in the hidden symbol store of package, that hidden symbol is removed.
Then symbol is interned into package, and package becomes its home package, making it a local symbol of package.
Note: if symbol is currently the hidden symbol of some package, it is not removed from the hidden symbol store of that package. This is a degenerate case. The implication is that if that hidden symbol is ever restored in that package, it will once again have that package as its home package, and consequently it will turn into a foreign symbol of package.
(symbolp obj)
The symbolp function returns t if obj is a symbol, otherwise it returns nil.
(symbol-name symbol)
The symbol-name function returns the name of symbol.
(symbol-package symbol)
The symbol-package function returns the home package of symbol. If symbol has no home package, it returns nil.
(keywordp obj)
The keywordp function returns t if obj is a keyword symbol, otherwise it returns nil.
(bindable obj)
The bindable function returns t if obj is a bindable symbol, otherwise it returns nil.
All symbols are bindable, except for keyword symbols, and the special symbols t and nil.
(use-sym symbol [package])
(use-sym-as symbol name [package])
The use-sym function brings an existing symbol into package.
The use-sym-as is similar, but allows an alternative name to be specified. The symbol will be interned under that name, rather than under its symbol name.
In all cases, both function return symbol.
The following equivalence holds:
(use-sym s p) <--> (use-sym-as s (symbol-name s) p)
Thus, in the following descriptions, when the remarks are interpreted as applying to use-sym, the name argument is understood as referring to the symbol-name of the symbol argument.
If package is the home package of symbol, then the function has no effect.
Otherwise symbol is interned in package under name.
If a symbol is already interned in package under name, then that symbol is is replaced. If that replaced symbol is a local symbol of package, meaning that package is its home package, then that replaced symbol turns into a hidden symbol associated with the package. It is placed into a special hidden symbol store associated with package and is stripped of its home package, becoming quasi-interned or uninterned.
Note: use-sym and use-sym-as are the basis for the defpackage clauses :use-syms and :use-syms-as.
Note: if use-sym-as is used to introduce a foreign symbol into a package under a different name, that symbol cannot be removed with unintern. It can only be removed using unuse-sym.
(unuse-sym symbol [package])
The unuse-sym function removes symbol from package.
If symbol is not interned in package, the function does nothing and returns nil.
If symbol is a local symbol of package, an error is thrown: a package cannot "unuse" its own symbol. Removing a symbol from its own home package requires the unintern function.
Otherwise symbol is a foreign symbol interned in package and is removed.
If the package has a hidden symbol of the same name as symbol, that symbol is reinterned into package as a local symbol. In this case, that previously hidden symbol is returned.
If the package has no hidden symbol matching the removed symbol, then symbol itself is returned.
There are close similarities between the function unintern and unuse-sym, but the two are significantly different.
Firstly, unuse-sym cannot be used to remove a symbol from its home package. As noted above, this requires unintern.
Secondly, unuse-sym can be used to undo the effect of use-sym-as whereby a foreign symbol is introduced into a package under a different name. If symbol is not found under its name, unuse-sym will search the package for that symbol to discover whether it is present under a different name, and proceed with the removal using that name. The unintern function performs no such secondary check; if symbol is not found in the package under its own name, the operation fails, and so unintern cannot be used for undoing the effect of use-sym-as.
(use-package package-list [package])
(unuse-package package-list [package])
The use-package and unuse-package are convenience functions which perform a mass import of symbols from one package to another, or a mass removal, respectively.
The use-package function iterates over all of the local symbols of the packages in package-list. For each symbol s, it performs the semantic action implied by the (use-sym s package) expression.
Similarly unuse-package iterates package-list in the same way, performing, effectively, the semantic action of the (unuse-sym s package) expression.
The package-list argument must be a list which is a mixture of symbols, strings or package objects. Strings are taken to be package names, which must resolve to existing packages. Symbols are reduced to strings via symbol-name.
(defpackage name clause*)
The defpackage macro provides a convenient means to create a package and establish its properties in a single construct. It is intended for the ordinary situations in which packages support the organization of programs into modules.
The name argument, giving the package name, may be a symbol or a character string. If it is a symbol, then the symbol's name is taken to be name for the package.
If a package called name already exists, then defpackage selects that package for further operations. Otherwise, a new, empty package is created. In either case, this package is referred to as the present package in the following descriptions.
The name may be optionally followed by one or more clauses, which are processed in the order that they appear. Each clause is a compound form headed by a keyword. The supported clauses are as follows:
(in-package name)
The in-package macro causes the *package* special variable to take on the package denoted by name. The macro checks, at expansion time, that name is either a string or symbol. An error is thrown if this isn't the case.
The name argument expression isn't evaluated, and so must not be quoted.
The code generated by the macro performs a search for the package. If the package is not found at the time when the macro's expansion is evaluated, an error is thrown.
The *random-state* variable holds an object which encapsulates the state of a pseudorandom number generator. This variable is the default argument value for the random-fixnum and random functions, for the convenience of writing programs which are not concerned about the management of random state.
On the other hand, programs can create and manage random states, making it possible to obtain repeatable sequences of pseudorandom numbers which do not interfere with each other. For instance objects or modules in a program can have their own independent streams of random numbers which are repeatable, independently of other modules making calls to the random number functions.
When TXR starts up, the *random-state* variable is initialized with a newly created random state object, which is produced as if by the call (make-random-state 42).
The *random-warmup* special variable specifies the value which is used by make-random-state in place of a missing warmup-period argument.
To "warm up" a pseudorandom number generator (PRNG) means to obtain some values from it which are discarded, prior to use. The number of values discarded is the warm-up period.
The WELL512a PRNG used in TXR produces 32-bit values, natively. Thus each warm-up iteration retrieves and discards a 32-bit value. The PRNG has a state space consisting of a vector of sixteen 32-bit words, making the state space 4096 bits wide.
Warm up is required because PRNG-s, in particular PRNG-s with large state spaces and long periods, produce fairly predictable sequences of values in the beginning, before transitioning into chaotic behavior. This problem is worse for low complexity seeds, such as small integer values.
The sequences are predictable in two ways. Firstly, some initial values extracted from the PRNG may exhibit patterns ("problem 1"). Secondly, the initial values from sequences produced from similar seeds (for instance consecutive integers) may be similar or identical ("problem 2").
The default value of *random-warmup* is only 8. This is insufficient to ensure good initial PRNG behavior for seeds even as large as 64 bits or more. That is to say, even if as many as eight bytes' worth of true random bits are used as the seed, the PRNG will exhibit predictable behaviors, and a poor distribution of values.
Applications which critically depend on good PRNG behavior should choose large warm-up periods into the hundreds or thousands of iterations. If a small warm-up period is used, it is recommended to use larger seeds which initialize more of the 4096-bit state space.
TXR's PRNG implementation addresses "problem 1" by padding the unseeded portions of the state space with random values (from a static table that doesn't change). For instance, if the integer 1 is used to seed the space, then one 32-bit word of the space is set to the value 1. The remaining 15 are populated from the random table. This helps to ensure that a good PRNG sequence is obtained immediately. However, it doesn't address "problem 2": that similar seed values generate similar sequences, when the warm-up period is small. For instance, if 65536 different random state objects are created, from each of the 16-bit seeds in the range [0, 65536), and then a random 16-bit value is extracted from each state, only 1024 unique values result.
(make-random-state [seed [warmup-period])
The make-random-state function creates and returns a new random state, an object of the same kind as what is stored in the *random-state* variable.
The seed, if specified, must be an integer value, a buffer, an existing random state object, or else a vector returned from a call to the function random-state-get-vec.
Note that the sign of the seed is ignored, so that negative seed values are equivalent to their additive inverses.
If seed is not specified, then make-random-state produces a seed based on some information in the process environment, such as current time of day. It is not guaranteed that two calls to make-random-state that are separated by less than some minimum increment of real time produce different seeds. The minimum time increment depends on the platform.
On a platform with a millisecond-resolution real-time clock, the minimum time increment is a millisecond. Calls to make-random-state less than a millisecond apart may predictably produce the same seed.
If an integer or buffer seed is specified, then the integer value is mapped to a pseudorandom sequence, in a platform-independent way.
If an existing random state is specified as a seed, then it is duplicated. The returned random state object is a distinct object which is in the same state as the input object. It will produce the same remaining pseudorandom number sequence, as will the input object.
If a vector is specified as a seed, then a random state is constructed which duplicates the random state object which was captured in that vector representation by the random-state-get-vec function.
The warm-up-period argument specifies the number of values which are immediately obtained and discarded from the newly-seeded generator before it is returned. This procedure is referred to as PRNG warm-up.
Warm-up is not performed if seed is a vector or random state object. In this situation, if the warm-up-period is present, it may still be required to be an integer, even though it is ignored.
If warm-up is performed, but the warm-up-period argument is missing, then the value of the *random-warmup* special variable is used. Note: this variable has a default value which may be too small for some applications of pseudorandom numbers; see the Notes under *random-warmup*.
(random-state-p obj)
The random-state-p function returns t if obj is a random state, otherwise it returns nil.
(random-state-get-vec [random-state])
The random-state-get-vec function converts a random state into a vector of integer values. If the random-state argument, which must be a random state object, is omitted, then the value of the *random-state* is used.
(random-fixnum [random-state])
(random random-state modulus)
(rand modulus [random-state])
All three functions produce pseudorandom numbers, which are positive integers.
The numbers are obtained from a WELL512a PRNG, whose state is stored in the random state object.
The random-fixnum function produces a random fixnum integer: a reduced range integer which fits into a value that does not have to be heap-allocated.
The random and rand functions produce a value in the range [0, modulus). They differ only in the order of arguments. In the rand function, the random state object is the second argument and is optional. If it is omitted, the global *random-state* object is used.
The modulus argument must be a positive integer. If modulus is 1, then the function returns zero without altering the state of the pseudorandom number generator.
(random-float [random-state])
(random-float-incl [random-state])l
The random-float function produces a pseudorandom floating-point value in the range [0.0, 1.0).
The random-float-incl produces a pseudorandom floating-point value in the range [0.0, 1.0], thus differing from random-float by including the 1.0 limit value.
The numbers are obtained from a WELL512a PRNG, whose state is stored in the random state object given by the argument to the optional random-state parameter, which defaults to the value of *random-state*.
Because the floating-point type does not provide a representation of every real value in the range 0.0 to 1.0, it is not possible to impose the requirement that every value shall occur with equal likelihood.
Rather, these functions are intended to produce an a uniform distribution of values according to the following pragmatic requirements. A subset S of the real values in the specified range, [0.0, 1.0) or [0.0, 1.0] is identified whose elements are representable in the floating-point type and which are uniformly spaced along the interval. Then, a random element is chosen from S and returned, such that every element is equally likely to be selected.
Note that these requirements do not correspond to the more mathematically ideal concept of uniformly choosing actual real numbers in the [0, 1] interval of the real number line, and then finding the closest floating-point representation. Such a requirement would mean that the boundary values 0.0 and 1.0 appear in the output half as frequently as all the interior values, because each of these two floating-point values is a representations of a range of numbers, half of which lies outside of the [0, 1] interval.
(random-buf size [random-state])
The random-buf function creates a buf object of the specified size fills it with pseudorandom bytes, and returns it.
The bytes are obtained from the random state object given by the optional random-state parameter, which defaults to the value of *random-state*.
See the section Buffers for a description of buf objects.
(random-sample size seq [random-state])
The random-sample function returns a vector of size randomly selected elements from the sequence seq, using reservoir sampling.
If the number of elements in seq is equal to or smaller than size, then the function returns a vector of all the elements of seq in their original order.
In other cases, the selected elements are not required to appear in their original order.
No element of sequence seq is selected more than once; duplicate values can appear in the output only if seq itself contains duplicates.
(time)
(time-usec)
(time-nsec)
The time function returns the number of seconds that have elapsed since midnight, January 1, 1970, in the UTC timezone: a point in time called the epoch.
The time-usec function returns a cons cell whose car field holds the seconds measured in the same way, and whose cdr field extends the precision by giving number of microseconds as an integer value between 0 and 999999.
The time-nsec function is similar to time-usec except that the returned cons cell's cdr field gives a number of nanoseconds as an integer value between 0 and 999999999.
Note: on hosts where obtaining nanosecond precision is not available, the time-nsec function obtains a microseconds value instead, and multiplies it by 1000.
(time-string-local time format)
(time-string-utc time format)
These functions take the numeric time returned by the time function, and convert it to a textual representation in a flexible way, according to the contents of the format string.
The time-string-local function converts the time to the local timezone of the host system. The time-string-utc function produces time in UTC.
The format argument is a string, and follows exactly the same conventions as the format string of the C library function strftime.
The time argument is an integer representing seconds obtained from the time function or from the car field of the cons returned by the time-usec function.
(time-str-local format [time])
(time-str-utc format [time])
The functions time-str-local and time-str-utc are equivalent, respectively, to time-string-local and time-string-utc with the arguments reversed. Thus the following equivalences hold:
(time-str-local F T) <--> (time-string-local T F)
(time-str-utc F T) <--> (time-string-utc T F)
Additionally, if no argument is supplied to the time parameter, its value is obtained by invoking the time function.
(time-fields-local [time])
(time-fields-utc [time])
These functions take numeric time in the format returned by the time function and convert it to a list of seven fields.
The time-fields-local function converts the time to the local timezone of the host system, whereas the time-fields-utc function produces time in UTC.
The fields returned as a list consist of six integers, and a Boolean value. The six integers represent the year, month, day, hour, minute and second. The Boolean value indicates whether daylight savings time is in effect (always nil in the case of time-fields-utc).
The time parameter is an integer representing seconds obtained from the time function. If the argument is absent, the value is obtained by calling time.
(defstruct time nil
year month day hour min sec
wday yday
dst gmtoff zone)
The time structure represents a time broken down into individual fields. The structure almost directly corresponds to the struct tm type in the ISO C language. There are differences. Whereas the struct tm member tm_year represents a year since 1900, the year slot of the time structure represents the absolute year, not relative to 1900. Furthermore, the month slot represents a one-based numeric month, such that 1 represents January, whereas the C member tm_mon uses a zero-based month. The dst slot is a TXR Lisp Boolean value. The slots hour, min, sec, wday and yday correspond directly to tm_hour, tm_min, tm_sec, tm_wday and tm_yday.
The slot gmtoff represents the number of seconds east of UTC, and zone holds a string giving the abbreviated time zone name. On platforms where the C type struct tm has fields corresponding to these slots, values for these slots are calculated and stored into them by the time-struct-local and time-struct-utc functions, and also the related time-local and time-utc methods. On platforms where the corresponding fields are not present in the C language struct tm, these slots are unaffected by those functions, retaining the default initial value nil or a previously stored value, if applicable. Lastly, the values of gmtoff and zone are not ignored by functions which accept a time structure as a source of input values.
(time-struct-local [time])
(time-struct-utc [time])
These functions take numeric time in the format returned by the time function and convert it to an instance of the time structure.
The time-struct-local function converts the time to the local timezone of the host system, whereas time-struct-utc function produces time in UTC.
The time parameter is an integer representing seconds obtained from the time function. If the argument is absent, the value is obtained by calling time.
(time-parse format string)
(time-parse-local format string)
(time-parse-utc format string)
The time-parse function scans a time description in string according to the specification given in the format string. If the scan is successful, a structure of type time is returned, otherwise nil.
The format argument follows the same conventions as the POSIX C library function strptime.
Prior to obtaining the time from format and string the returned structure is created and initialized with a time which represents time 0 ("the epoch") if interpreted in the UTC timezone as by the time-utc method.
The time-parse-local and time-parse-utc functions return an integer time value: the same value that would be returned by the time-local and time-utc methods, respectively, when applied to the structure object returned by time-parse. Thus, these equivalences hold:
(time-parse-local f s) <--> (time-parse f s).(time-local)
(time-parse-utc f s) <--> (time-parse f s).(time-utc)
Note: the availability of these three functions depends on the availability of strptime.
Note: on some platforms, like the GNU C Library, the strptime function supports the parsing of numeric and symbolic time zones. The gmtoff slot of the structure ends up being set accordingly. The time-local and time-utc functions take the gmtoff field into account, adjusting the returned time accordingly.
If these are specified.
time-struct.(time-local)
time-struct.(time-utc)
The time structure has two methods called time-local and time-utc.
The time-local function considers the slots of the time structure instance time-struct to be local time, and returns its integer representation as the number of seconds since the epoch.
The time-utc function is similar, except it considers the slots of time-struct to be in the UTC time zone.
Note: these functions work by converting the slots into arguments to which make-time or make-time-utc is applied.
Note: if the gmtoff slot is not nil, its value is subtracted from the returned result.
time-struct.(time-string format)
The time structure has a method called time-string.
This method accepts a format string argument, which it uses to convert the fields to a character string representation which is returned.
The format argument is a string, and follows exactly the same conventions as the format string of the C library function strftime.
time-struct.(time-parse format string)
The time-parse method scans a time description in string according to the specification given in the format string.
If the scan is successful, the structure is updated with the parsed information, and the remaining unmatched portion of string is returned. If all of string is matched, then an empty string is returned. Slots of time-struct which are originally nil are replaced with zero, even if these zero values are not actually parsed from string.
If the scan is unsuccessful, then nil is returned and the structure is not altered.
The format argument follows the same conventions as the POSIX C library function strptime.
Note: the time-parse method may be unavailable if the host system does not provide the strptime function. In this case, the time-parse static slot of the time struct is nil.
(make-time year month day
hour minute second dst-advice)
(make-time-utc year month day
hour minute second dst-advice)
The make-time function returns a time value, similar to the one returned by the time function. The time value is constructed not from the system clock, but from a date and time specified as arguments. The year argument is a calendar year, like 2014. The month argument ranges from 1 to 12. The hour argument is a 24-hour time, ranging from 0 to 23. These arguments represent a local time, in the current time zone.
The dst-advice argument specifies whether the time is expressed in daylight savings time (DST). It takes on three possible values: nil, the keyword :auto, or else the symbol t. Any other value has the same interpretation as t.
If dst-advice is t, then the time is assumed to be expressed in DST. If the argument is nil, then the time is assumed not to be in DST. If dst-advice is :auto, then the function tries to determine whether DST is in effect in the current time zone for the specified date and time.
The make-time-utc function is similar to make-time, except that it treats the time as UTC rather than in the local time zone. The dst-advice argument is supported by make-time-utc for function call compatibility with make-time. It may or may not have any effect on the output (since the UTC zone by definition doesn't have daylight savings time).
(crc32-stream stream [nbytes [crc-prev]])
The crc32-stream calculates the CRC-32 sum over the bytes read from stream, starting at the stream's current position.
If the nbytes argument is specified, it should be a nonnegative integer. It gives the number of bytes which should be read and included in the sum. If the argument is omitted, then bytes are read until the end of the stream.
The optional crc-prev argument defaults to zero. It is fully documented under the crc32 function.
The crc32-stream functions returns the calculated CRC-32 as a nonnegative integer.
(crc32 obj [crc-prev])
The crc32 function calculates the CRC-32 sum over obj, which may be a character string or a buffer.
If obj is a buffer, then the sum is calculated over all of the bytes contained in that buffer, according to its current length.
If obj is a character string, then the sum is calculated over the bytes which constitute its UTF-8 representation.
The optional crc-prev argument defaults to zero. If specified, it should be a nonnegative integer in the 32-bit range. This argument is useful when a single CRC-32 must be calculated in multiple operations over several objects. The first call should specify a value of zero, or omit the argument. To continue the checksum, each subsequent call to the function should pass as the crc-prev argument the CRC-32 obtained from the previous call.
The crc32 function returns the calculated CRC-32 as a nonnegative integer.
The parameters of the algorithm are as follows. The polynomial is #x04C11DB7; the input and result are reflected; the initial value is #xFFFFFFFF; and the final value is bitwise xor-ed with #xFFFFFFFF.
;; Single operation
(crc32 "ABCD") --> 3675725989
;; In two steps, demonstrating crc-prev argument:
(crc32 "CD" (crc32 "AB")) -> 3675725989
(sha1-stream stream [nbytes [buf]])
(sha256-stream stream [nbytes [buf]])
(md5-stream stream [nbytes [buf]])
The sha1-stream and sha256-stream functions calculate, respectively, the NIST SHA-1 and SHA-256 digests over the bytes read from stream, starting at the stream's current position.
The md5-stream function calculates the MD5 digest, using the RSA Data Security, Inc. MD5 Message-Digest Algorithm.
If the nbytes argument is specified, it should be a nonnegative integer. It gives the number of bytes which should be read and included in the digest. If the argument is omitted, then bytes are read until the end of the stream.
If the buf argument is omitted, the digest value is returned as a new, buffer object. This buffer is 32 bytes long in the case of SHA-256, holding a 256-bit digest, and 16 bytes long in the case of MD5, holding a 128-bit digest. If the buf argument is specified, it must be a buffer that is at least 16 bytes long in the case of MD5, and at least 32 bytes long in the case of SHA-256. The hash is placed into that buffer, which is then returned.
(sha1 obj [buf])
(sha256 obj [buf])
(md5 obj [buf])
The sha1 and sha256 function calculate, respectively, the NIST SHA-1 and SHA-256 digests over obj, which may be a character string or a buffer.
Similarly, the md5 functions calculates the MD5 digest over obj, using the RSA Data Security, Inc. MD5 Message-Digest Algorithm.
If obj is a buffer, then the digest is calculated over all of the bytes contained in that buffer, according to its current length.
If obj is a character string, then the digest is calculated over the bytes which constitute its UTF-8 representation.
If the buf argument is omitted, the digest value is returned as a new, buffer object. This buffer is 32 bytes long in the case of SHA-256, holding a 256-bit digest, and 16 bytes long in the case of MD5, holding a 128-bit digest. If the buf argument is specified, it must be a buffer that is at least 16 bytes long in the case of MD5, and at least 32 bytes long in the case of SHA-256. The hash is placed into that buffer, which is then returned.
(sha1-begin)
(sha1-hash ctx obj)
(sha1-end ctx [buf])
The three functions sha1-begin, sha1-hash and sha1-end implement a stateful computation of SHA-1 digest which allows multiple input sources to contribute to the result. Furthermore, the context object may be serially reused for calculating multiple digests.
The sha1-begin function, which takes no arguments, returns a new SHA-1 digest-producing context object.
The sha1-hash updates the state of the SHA-1 digest object ctx by including obj into the digest calculation. The obj argument may be: a character or character string, whose UTF-8 representation is digested; a buffer object, whose contents are digested; or an integer, representing a byte value in the range 0 to 255 included in the digest. The sha1-hash function may be called multiple times to include any mixture of strings and buffers into the digest calculation.
The sha1-end function finalizes the digest calculation and returns the digest in a buffer. If the buf argument is omitted, then a new 20-byte buffer is created for this purpose. Otherwise, buf must specify a buf object that is at least 20 bytes long. The digest is stored into this buffer and that the buffer is returned.
The sha1-end function additionally resets the ctx object into the initial state of a newly created context object, so that it may be used for another digest session.
(sha256-begin)
(sha256-hash ctx obj)
(sha256-end ctx [buf])
The three functions sha256-begin, sha256-hash and sha256-end implement a stateful computation of SHA-256 digest which allows multiple input sources to contribute to the result. Furthermore, the context object may be serially reused for calculating multiple digests.
The sha256-begin function, which takes no arguments, returns a new SHA-256 digest-producing context object.
The sha256-hash updates the state of the SHA-256 digest object ctx by including obj into the digest calculation. The obj argument may be: a character or character string, whose UTF-8 representation is digested; a buffer object, whose contents are digested; or an integer, representing a byte value in the range 0 to 255 included in the digest. The sha256-hash function may be called multiple times to include any mixture of strings and buffers into the digest calculation.
The sha256-end function finalizes the digest calculation and returns the digest in a buffer. If the buf argument is omitted, then a new 32-byte buffer is created for this purpose. Otherwise, buf must specify a buf object that is at least 32 bytes long. The digest is stored into this buffer and that the buffer is returned.
The sha256-end function additionally resets the ctx object into the initial state of a newly created context object, so that it may be used for another digest session.
(md5-begin)
(md5-hash ctx obj)
(md5-end ctx [buf])
The three functions md5-begin, md5-hash and md5-end implement a stateful computation of MD5 digest which allows multiple input sources to contribute to the result. Furthermore, the context object may be serially reused for calculating multiple digests.
The md5-begin function, which takes no arguments, returns a new MD5 digest-producing context object.
The md5-hash updates the state of the MD5 digest object ctx by including obj into the digest calculation. The obj argument may be: a character or character string, whose UTF-8 representation is digested; a buffer object, whose contents are digested; or an integer, representing a byte value in the range 0 to 255 included in the digest. The md5-hash function may be called multiple times to include any mixture of strings and buffers into the digest calculation.
The md5-end function finalizes the digest calculation and returns the digest in a buffer. If the buf argument is omitted, then a new 16-byte buffer is created for this purpose. Otherwise, buf must specify a buf object that is at least 16 bytes long. The digest is stored into this buffer and that the buffer is returned.
The md5-end function additionally resets the ctx object into the initial state of a newly created context object, so that it may be used for another digest session.
The TXR Lisp library provides a macro called awk which is inspired by the Unix utility Awk. The macro implements a processing paradigm similar to that of the utility: it scans one or more input streams, which are divided into records and fields, under the control of user-settable regular-expression-based delimiters. The records and fields are matched against a sequence of programmer-defined conditions (called "patterns" in the original Awk), which have associated actions. Like in Awk, the default action is to print the current record.
Unlike Awk, the awk macro is a robust, self-contained language feature which can be used anywhere where a TXR Lisp expression is called for, cleanly nests with itself and can produce a return value when done. By contrast, a function in the Awk language, or an action body, cannot instantiate a local Awk processing machine.
The awk macro implements some of the most important Awk conventions and semantics, in Lisp syntax, while eschewing others. It does not implement the Awk convention that variables become defined upon first mention; variables must be defined to be used. It doesn't implement Awk's weak type system. A character string which looks like a number isn't a number, and an empty string or undefined variable doesn't serve as zero in arithmetic expressions enclosed in the macro. All expression evaluation within awk is the usual TXR Lisp evaluation.
The awk macro also does not provide a library of functions corresponding to those in the Awk library, nor does it provide counterparts various global variables in Awk such as the ENVIRON and PROCINFO arrays, or RSTART and RLENGTH. Such features of Awk are extraneous to its central paradigm.
(awk {(condition action*)}*)
The awk macro processes one or more input sources, which may be streams or files. Each input source is scanned into records, and each record is broken into fields. For each record, the sequence of condition-action clauses (except for certain special clauses) is processed. Every condition is evaluated, and if it yields true, the corresponding actions are evaluated.
The condition and action forms are understood to be in a scope in which certain local identifiers exist in the variable namespace as well as in the function namespace. These are called awk functions and awk macros.
If condition is one of the following keyword symbols, then it is a special clause, with special semantics: :name, :let, :inputs, :output, :begin, :set, :end, :begin-file, :set-file and :end-file. These clause types are explained below. In such a clause, the action expressions are not necessarily forms to be evaluated; the treatment of these expressions depends on the clause. Otherwise, if condition is not one of the above keyword symbols, the clause is an ordinary condition-action clause, and condition is a TXR Lisp expression, evaluated to determine a Boolean value which controls whether the action forms are evaluated. In every ordinary condition-action clause which contains no action forms, the awk macro substitutes the single action equivalent to the form (prn): a call to the local awk function prn. The behavior of this macro, when called with no arguments, as above, is to print the current record (contents of the variable rec) followed by the output record terminator from the variable ors.
While the processing loop in awk scans an input source, it also binds the special variable *stdin* to the open stream associated with that source. This binding is in effect across all ordinary clauses, as well as across the special clauses :begin-file and :end-file.
The following is a description of the special clauses:
If the :name form is omitted, the implicit block is named awk.
It is an error for two or more :name forms to appear.
Note: in TXR 255 and older, the :name clause must have an argument which is a symbol. The symbol nil is not permitted.
If multiple :let clauses are present, they are effectively consolidated into a single clause, in the order they appear.
Note that the lexical variables, functions and macros established by the awk macro (called, respectively, awk macros, awk functions and awk variables) are in an inner scope relative to :let bindings. For instance if :let creates a binding for a variable called fs, that variable will be visible only to subsequent forms appearing in the same :let clause or later :let clauses, and also visible in :inputs and :output clauses. In :begin, :set, :end, and ordinary clauses, it will be shadowed by the awk variable fs, which holds the field-separator regular expression or string.
Each input source must be one of three kinds of objects. It may be a stream object, which must be capable of character input. It may be a list of strings, which awk will convert to an input stream as if by the make-strlist-input-stream function. Or else it must be a character string, which denotes a filesystem pathname which awk will open for reading.
If the :inputs clause is omitted, then a defaulting behavior occurs for obtaining the list of input sources. If the special variable *args* isn't the empty list, then *args* is taken as the input sources. Otherwise, the *stdin* stream is taken as the one and only input source.
If the awk macro uses *args* via the above defaulting behavior, it copies *args* and sets that variable to nil. This is done in order that if awk is used from the TXR command line, for example using the -e command-line option, after awk terminates, TXR will not try to open the next argument as a script file or treat it as an option. Note: programs which want awk not to modify *args* can explicitly specify *args* as the argument to the :inputs keyword, rather than allow *args* to be used through the defaulting behavior. Only the defaulting behavior consumes the arguments by overwriting *args* with nil.
It is an error to specify more than one :inputs clause.
The :output clause, if present, has the effect of creating a local binding for the *stdout* special variable. This new value of *stdout* is visible to all forms within the macro. If a :let clause is present, it establishes bindings in a scope which is nested within the scope established by :output. Therefore, init-forms in the :let may refer to the new value of *stdout* established by :output. Furthermore, :let can rebind *stdout*, causing the definition provided by :output to be shadowed.
In the case when the :output argument is a string such that a new stream is opened on the file, the awk macro will close that stream when it finishes executing. Moreover, that stream is treated uniformly as a member of the set of streams that are implicitly managed by the redirection macros in the same awk macro invocation. In brief, the implication is that if :output creates a stream for the file pathname "out.txt" and somewhere in the same awk macro, there is a redirection of the form, or equivalent to (-> "out.txt") then this redirection shall refer to the same stream that was established by :output. Note also that in this example situation, the expression (-> "out.txt" :close) has the effect of closing the :output stream.
Upon termination, the end clauses are processed in the order they appear. Each form is evaluated, left to right.
In the normal termination case, the value of the last form of the last end clause appears as the return value of the awk macro.
Note that only termination of the awk macro initiated from condition-action clauses, :begin-file clauses, or :end-file clauses triggers :end clause processing. If termination of the awk macro is initiated from within a :let, :inputs, :output or :begin clause, then end clauses are not processed. If an :end clause performs a nonlocal transfer, the remaining :end forms in that clause and :end clauses which follow are not evaluated.
If both :begin and :begin-file forms are specified, then before the first input is processed, :begin clauses are processed first, then the :begin-file clauses.
If both :end and :end-file forms are specified, then before after the last input is processed, :end-file clauses are processed first, then the :end clauses.
The :end-file clauses are processed unconditionally, no matter how the processing of an input source terminates, whether terminated naturally by running out of records, prematurely by invocation of the next-file macro, or via a dynamic nonlocal control transfer such as a block return or exception throw.
If a :begin-file clause performs a nonlocal transfer, :end-file processing is not triggered, because the processing of the input source is deemed not to have taken place.
The awk variable rec holds the current record. It is automatically updated prior to the processing of the condition-action clauses. Prior to the extraction of the first record, its value is nil.
It is possible to assign to rec. The value assigned to rec must be a character string. Immediately upon the assignment, the character string is delimited into fields according to the field separator awk variable fs, and these fields are assigned to the field list f. At the same time, the nf variable is updated to reflect the new number of fields. Likewise, modification of these variables causes rec to be reconstructed by a catenation of the textual representation of the fields in f separated by copies of the output field separator ofs.
The orec variable ("original record") also holds the current record. It is automatically updated prior to the processing of the condition-action clauses at the same time as rec with the same contents. Like rec, it is initially nil before the first record is read. The orec variable is unaffected by modification of the variables rec, f and nf. It may be assigned. Doing so has no effect on any other variable.
The awk variable f holds the list of fields. Prior to the first record being read, its value is nil. Whenever a new record is read, it is divided into fields according to the field separator variable fs, and these fields are stored in f as a list of character strings.
If the variable f is assigned, the new value must be a sequence. The variable nf is automatically updated to reflect the length of this sequence. Furthermore, the rec variable is updated by catenating a string representation of the elements of this sequence, separated by the contents of the ofs (output field separator) awk variable.
Note that assigning to a DWIM bracket form which indexes f, such as for instance [f 0] constitutes an implicit modification of f, and triggers the recalculation of rec. Modifications of the f list which do not involve an implicit or explicit assignment to the variable f itself do not have this recalculating effect.
Unlike in Awk, assigning to the nonexistent field [f m] where m >= nf is erroneous.
The awk variable nf holds the current number of fields in the sequence f. Prior to the first record being read, it is initially zero.
If nf is assigned, then f is modified to reflect the new number of fields. Fields are deleted from f if the new value of nf is smaller. If the new value of nf is larger, then fields are added. The added fields are empty strings, which means that f must be a sequence of a type capable of holding elements which are strings.
If nf is assigned, then rec is also recalculated, in the same way as described in the documentation for the f variable.
The awk variable nr holds the current absolute record number. Record numbers start at 1. Absolute means that this value does not reset to 1 when awk switches to a new input source; it keeps incrementing for each record. See the fnr variable.
Prior to the first record being read, the value of nr is zero.
The awk variable fnr holds the current record number within the file. The first record is 1.
Prior to the first record being read from the first input source, the value of fnr is zero. Thereafter, it resets to 1 for the first record of each input source and increments for the remaining records of the same input source.
The awk variable arg is an integer which indicates what input source is being processed. Prior to input processing, it holds the value zero. When the first record is extracted from the first input source, it is set to 1. Thereafter, it is incremented whenever awk switches to a new input source.
The awk variable fname provides access to a character string which, if the current input is a file stream, is the name of the underlying file. Assigning to this variable changes its value, but has no effect on the input stream. Whenever a new input source is used by awk, this variable is set from the file name on which it is opening a stream. When using an existing stream rather than opening a file, awk sets this variable from the :name property of the stream.
Note that the redirection macros <- and <! have no effect on this variable. Within their scope, fname retains its value.
The awk variable rs specifies a string or regular expression which is used for delimiting characters read from the inputs into pieces called records.
Note: the record extraction is internally implemented using record streams instantiated by the record-adapter function.
The regular-expression pattern stored in rs is used to matches substrings in the input which separate or terminate records. Unless the krs variable is set true, the substrings which match rs are discarded and the records consist of the nonmatching extents between them.
The initial value of rs is "\n": the newline character. This means that, by default, records are lines.
If rs is changed to the value nil, then record separation operates in paragraph mode, which is described below.
If a match for the record separator occurs at the end of the stream, it is not considered to delimit an empty record, but acts as the terminator for the previous record.
When a new value is assigned to rs, it has no effect on the most recently scanned and delimited record which is still current, or previous records. The new value applies to the next, not yet read record.
In paragraph mode, records are separated by a newline character followed by one or more blank lines (empty lines or lines containing only a mixture of tabs and spaces). This means that, effectively, the record-separating sequences match the regular expression /\n[ \n\t]*\n/.
There are two differences between paragraph mode and simply using the above regular expression as rs. The first difference is that if the first record which is read upon entering paragraph mode is empty (because the input begins with a match for the separator regex), then that record is thrown away, and the next record is read. The second difference is that, if field separation based on the fs variable is in effect, then regardless of the value of fs, newline characters separate fields. Therefore, the programmer-defined fs doesn't have to include a match for newline. Moreover, if it is a simple fixed string, it need not be converted to a regular expression which also matches a newline.
The awk variable krs stands for "keep record separator". It is a Boolean variable, initialized to nil.
If it is set to a true value, then the separating text matched by the pattern in the rs variable is retained as part of the preceding record rather than removed.
When a new value is assigned to krs, it has no effect on the most recently scanned and delimited record which is still current, or previous records. The new value applies to the next, not yet read record.
The awk variable fs and ft each specify a string or regular expression which is used for each record that is stored in the rec variable into fields.
Both variables are initialized to nil, in which case a default behavior is in effect, described below.
Use of these variable is mutually exclusive; it is an error for both of these variables to simultaneously have a value other than nil. The value stored in either variable must be nil, a character string or a regular expression. If it contains a string or regex, it is said to contain a pattern. A string value effectively behaves as a fixed regular expression which matches the sequence of characters in the string verbatim, without treating any of them as regex operators.
The splitting of rec into fields is influenced by the Boolean kfs ("keep field separators") variable, whose effect is discussed in its description. If kfs is false, the splitting is carried out as follows.
If fs contains a pattern, then rec is treated specially when it is the empty string: in that case, the pattern in fs is ignored, and no fields are produced: the field list f is the empty list, and nf is zero. A nonempty record is split by searching it for matches for the fs pattern. If a match does not occur, then the entire record is a field. If one match occurs, then the record is split into two fields, either of which, or both, might be empty. If two matches occur, the record is split into three fields, and so on. If fs finds only an empty string match in the record, then it is considered to match each of the empty strings between two consecutive characters of the record. Consequently, the record is split into its individual characters, each one becoming a field. Note: all of these behaviors, except for the special treatment of the empty record, are accomplished by a call to the split-str function.
If the variable ft ("field tokenize") contains a pattern, that pattern is used to positively recognize tokens within the input record, rather than to match separating material between them. Those matching tokens then constitute the fields. The tokenizing is performed using the tok-str function.
If fs and ft are both nil, as is initially the case, then the splitting into fields is performed as if the ft variable held the regular expression /[^\n\t ]+/. This means that, by default, fields are sequences of consecutive characters which are not spaces, tabs or newlines. Newlines are excluded from fields (and thus separate them) because they can occur in a record when the value of the record separator rs is customized.
The awk variable kfs is a Boolean flag which is initialized to nil.
If it is set to any other value, it indicates a request to retain the pieces of the record which separate the fields (even when they are empty strings). The retained pieces appear as fields, interspersed among the regular fields so that all of the fields appear in the order in which they were extracted from the record.
When kfs is set, and tokenization-style delimiting is in effect due to ft being set, there is always at least one field, even if the record is empty. If the record doesn't match the tokenizing regular expression in ft then a single field is generated, then the entire record is taken as one field, denoting the nonmatching space, even if the record is the empty string.
If the record matches one or more tokens, then the first and last field will always contain the nonmatching material before the first and last token, respectively. This is true even if the material is empty. Thus [f 0] always has the material before the first token, whether or not the first token is matched immediately at the first character position in the record. This behavior follows from the semantics of the keep-sep parameter of the tok-str function.
Similarly, when splitting based on fs is in effect and kfs is set, there is always at least one field, even if the record is empty. If fs finds no match in the record, then the entire record, even if empty, is taken as one field. In that case, there are no separator to retain. When fs finds one or more matches, then these are included as fields. Separators are always between the fields. If a separator finds a nonempty match at the beginning of a record, that causes an empty field to be split off: the separator is understood as intervening between an empty string before the first character of the record, and subsequent material which follows the text matched by the separator. Thus the first field is an empty field, and the second is the matched text which isincluded due to kfs being set. An analogous situation occurs at the end of the record: if fs matches a nonempty string at the tail of the record, it splits off an empty last field, preceded by a field holding the matched separator portion. Empty matches are only permitted to occur between the characters of the record, not before the first character of after the last. If fs matches the entire record, then there will be three fields: the first and last of these three will be empty strings, and the middle field, the separator, will be a copy of the record. Under kfs, empty matches cause empty string to be included among the fields. All of this follows from the semantics of the keep-sep parameter of the split-str function.
The awk variable fw controls the fixed-width-based delimiting of records into fields.
The variable is initialized to nil. In that state, it has no effect. When this variable holds a non-nil value, it is expected to be a list of integers. The use of the fs or ft variables is suppressed, and fields are extracted according to the widths indicated by the list. The fields are consecutive, such that if the list is (5 3) then the first five characters of the record are identified as field [f 0] and the next three characters after that as [f 1].
Only complete fields are extracted from the record. If, after the extraction of the maximum possible complete fields, more characters remain, those characters are assigned to an extra field.
An empty record produces an empty list of fields regardless of the integers stored in fw.
A zero width extracts a zero length field, except when no more characters remain in the record.
If nil is stored into fw then control over field separation is relinquished to the fs or ft variables, according to their current values.
If fw holds a value other than nil or else a list of nonnegative integers, the behavior is unspecified.
The following table shows how various combinations of the value the input record rec and field widths in the variable fw give rise to field values f:
rec fw f
---------------------------------
"abc" (0) ("" "abc")
"abc" (2) ("ab" "c")
"abc" (1 2) ("a" "bc")
"abc" (1 3) ("a" "bc")
"abc" (1 1) ("a" "b" "c")
"abc" (3) ("abc")
"abc" (4) ("abc")
"" (4) nil
"" (0) nil
The awk variable ofs hold the output field separator. Its initial value is a string consisting of a single space character.
When the prn function prints two or more arguments, or fields, the value of ofs is used to separate them.
Whenever rec is implicitly updated due to a change in the variable f or nf, ofs is used to separate the fields, as they appear in rec.
The awk variable ors, though it stands for "output record separator" holds what is in fact the output record terminator. It is named after the ORS variable in Awk.
Each call to the prn function terminates its output by emitting the value of ors.
The initial value of ors is a character string consisting of a single newline, and so the prn function prints lines.
The awk variable res is implicitly bound over the scope of the action forms of every condition-action clause. It holds the result of the condition form.
Because the action forms execute only if the condition yields true, it follows that res is never observed with a value of nil unless the program explicitly assigns that value.
Note: this is an original feature in the TXR Lisp awk macro, which has no counterpart in POSIX or GNU Awk.
(awk
(:inputs '("carpet"))
(#/a.*p/ (prn res)))
Output:
arp
In this example, the result of the #/a.*p/ regular expression being applied to the input carpet is the string "arp" and so over that clause, prn takes on that string as its value. Thus, thanks to prn, the action has access to the matching part of the record.
(prn form*)
The awk function prn performs output into the *stdout* stream. The :output clause affects the destination by rebinding *stdout*.
If called with no arguments, prn prints rec followed by ors.
Otherwise, it prints the values of the arguments, separated by ofs, followed by ors.
When a condition-action clause specifies no action forms, then a call to prn with no arguments is the default action.
Each argument form is printed by conversion to a string, as if by the expression `@val` where val is some variable which holds the value produced by the evaluation of form. Thus if the value is nil, the output for that argument is an empty string, rather than the text "nil".
(next)
The awk macro next may be invoked in a condition-pattern clause. It terminates the processing of that clause, and all subsequent clauses, causing awk to process the next record, if there is one. If there is no next record, awk terminates.
(again)
The awk macro again may be invoked in a condition-pattern clause. It terminates the processing of that clause, and all subsequent clauses. Then, the current value of the record, namely the datum stored in the awk variable rec, is delimited into fields, and all of the condition-pattern clauses are processed again.
No other state is modified. In particular, the record number nr and the orec variable holding the original record both retain their current values.
Note: this is an original feature in the TXR Lisp awk macro, which has no counterpart in POSIX or GNU Awk.
(next-file)
The awk macro next-file may be invoked in a condition-pattern clause. It terminates the processing of that clause and all subsequent clauses. Then awk abandons the current input source and moves to the next one. If there is no next input source, awk terminates.
(rng from-condition to-condition)
(-rng from-condition to-condition)
(rng- from-condition to-condition)
(-rng- from-condition to-condition)
(--rng from-condition to-condition)
(--rng- from-condition to-condition)
(rng+ from-condition to-condition)
(-rng+ from-condition to-condition)
(--rng+ from-condition to-condition)
The nine awk macros in the rng family may be used anywhere within an ordinary condition-pattern awk clause.
Each provides a Boolean test which is true if the current record lands within a range of records delimited by conditions. Each provides its own distinct, useful nuance, which is identified by the mnemonic characters prefixed or suffixed to the name.
The basic rng macro inclusively matches ranges of records. Each such range begins with a record for which from-condition yields true, and ends on the record for which to-condition is true. What it means to match is that the rng expression yields a Boolean true value when it is evaluated in the context of processing any of the records which are included in the range.
The table below summarizes the semantic variations of these nine range macro operators. The leftmost column labeled DATA represents the stream of records being processed. Each entry in this column gives the literal piece of text which comprises the content of one record in the stream. The remaining nine columns, labeled with the nine range operators, inform about the behavior of these operators with respect to these records. In each of these columns the letter X marks those records for which the column's range operator yields true, if it is invoked with the arguments #/H/ and #/T/ as its from-condition and to-condition, respectively. For example, the rng column shows the values of the (rng #/H/ #/T/) expression, indicating that the expression starts being true when the H1 record is seen, stays true for the T1 record, and then reverts to false:
DATA rng -rng rng- -rng- --rng --rng- rng+ -rng+ --rng+
----------------------------------------------------------
PROLOG
H1 X X X
H2 X X X X X X
H3 X X X X X X
B1 X X X X X X X X X
B2 X X X X X X X X
T1 X X X X X X
T2 X X X
T3 X X X
EPILOG
The prefix and suffix characters of the operator names are intended to be mnemonic. A single - (dash) indicates the exclusion of one record. A double -- (dash dash) indicates the exclusion of all leading records which match from-condition; this appears on the left side only. The + character, appearing on the right only, indicates that all consecutive records which match to-condition are included in the range, not only the first one.
Ranges are oblivious to the division between successive sources of input; a range can start in one file of records and terminate in another. To prevent a range from spanning input transitions, additional complexity is required in the expression.
Ranges expressed using the rng family macros may combine with other expressions, including other ranges, and allow arbitrary nesting: the from-condition or to-condition can be a range, or an expression containing ranges.
The expressions from-condition and to-condition are ordinary expressions which are evaluated. However, their evaluation is unusual in two ways.
Firstly, if either expression produces, as its result, a function or regular-expression object, then that function or regular-expression object is applied to the current record (value of the rec variable), and the result of that application is then taken as the result of the condition. This allows for expressions like (rng (f^ #/start/) #/end/) which denotes a range which begins with a record which begins with the prefix "start" and ends with a record which contains "end" as a substring.
Secondly, the conditions are evaluated out of order with respect to the surrounding expression in which they occur. Ranges and their constituent from-condition and to-condition are evaluated just prior to the processing of the condition-action clauses. Each rng expression is reduced to a Boolean value. Then, when the condition-action clauses are processed and their condition and action forms are evaluated, each occurrence of a rng expression simply denotes its previously evaluated Boolean value.
Therefore, it is not possible for expressions to short circuit the evaluation of ranges. Ranges cannot "miss" their starting or terminating conditions; every range occurring anywhere in the condition-action clauses is tested against every record that is processed.
Because of this perturbed evaluation order, code which happens to place side effects into ranges may produce surprising results.
For instance, the expression (if nil (rng (prinl 'hello) (prinl 'world))) will produce output even though the if condition is nil, and, moreover, this output will happen before the clauses are processed in which this if expression appears. At the time when the if itself is evaluated, the rng expression merely fetches a previously computed Boolean value which indicates whether the range is active for this record.
Also, the behavior is unspecified if range expressions attempt to modify any of the special awk variables rec, f, fs, ft and kfs. It is not recommended to place any side effects into range expressions.
A more detailed description of the range operators follows.
(ff opip-arg *)
The awk macro ff (filter fields) provides a shorthand for filtering the field list f trough a pipeline of chained functions expressed using opip argument syntax.
The following equivalence holds, except that f refers to the awk variable even if the ff invocation occurs in code which establishes a binding which shadows f.
(ff a b c ...) <--> (set f [(opip a b c ...) f])
;; convert all fields from string to floating-point
(ff (mapcar flo-str))
(mf opip-arg *)
The awk macro mf (map fields) provides a shorthand for mapping each field individually trough a pipeline of chained functions expressed using opip argument syntax.
The following equivalence holds, except that f refers to the awk variable even if the mf invocation occurs in code which establishes a binding which shadows f.
(mf a b c ...) <--> (set f (mapcar (opip a b c ...) f))
;; convert all fields from string to floating-point
(mf flo-str)
(fconv {clause | : | - }*)
The awk macro fconv provides a succinct way to request conversions of the textual fields. Conversions are expressed by clauses which correspond with fields.
Each clause is an expression which must evaluate to a function. The clause is evaluated in the same manner as the argument a dwim operator, using Lisp-1-style name lookup. Thus, functions may be specified simply by using their name as a clause.
Furthermore, several local functions exist in the scope of each clause, providing a shorthand notation. These are described below.
Conversion proceeds by applying the function produced by a clause to the field to which that clause corresponds, positionally. The return value of the function applied to the field replaces the field.
When a clause is specified as the symbol - (minus) it has a special meaning: this minus clause occupies a field position and corresponds to a field, but performs no conversion on its field.
The : (colon) keyword symbol isn't a clause and does not correspond to a field position. Rather, it acts as a separator among clauses. It need not appear at all. If it appears, it may appear at most twice. Thus, the clauses may be separated into up to three sequences.
If the colon does not appear, then all the clauses are prefix clauses. Prefix clauses line up with fields from left to right. If there are fewer fields than prefix clauses, the values of the excess clauses are evaluated, but ignored. Vice versa, if there are fewer prefix clauses than fields, then the excess fields are not subject to conversions.
If the colon appears once, then the clauses before the colon, if any, are prefix clauses, as described in the previous paragraph. Clauses after the colon, if any, are interior clauses. Interior clauses apply to any fields which are left unconverted by the prefix clauses. All interior clauses are evaluated. If there are fewer fields than interior clauses, then the values of the excess interior clauses are ignored. If there are more fields than clauses, then the clause values are cycled: reused from the beginning against the excess fields, enough times to convert all the fields.
If the colon appears twice, then the clauses before the first colon, if any, are prefix clauses, the clauses between the two colons are interior clauses, and those after the second colon are suffix clauses. The presence of suffix clauses change the behavior relative to the one-colon case as follows. After the conversions are performed according to the prefix clauses, the remaining fields are counted. If there are are only as many fields as there are suffix clauses, or fewer, then the interior clauses are evaluated, but ignored. The remaining fields are processed against the suffix clauses. If after processing the prefix clauses there are more fields remaining than suffix clauses, then a number of rightmost fields equal to the number of suffix clauses is reserved for those clauses. The interior fields are applied only to the unreserved middle fields which precede these reserved rightmost fields, using the same repeating behavior as in the one-colon case. Finally, the previously reserved rightmost fields are processed using the suffix clauses.
The following special convenience functions are in scope of the clauses, effectively providing a shorthand for commonly-needed conversions:
The return value of fconv is f.
Note: because f is nil when no fields have been extracted, a fconv expression can be used as the condition in an awk clause which triggers the action if one or more fields have been extracted, and performs conversions on them.
Note: although fconv is intended for converting textual fields, and the semantic descriptions below consequently make references to string inputs, the behavior of fconv with respect to non-string fields can be inferred. For instance if a field actually holds the floating-point value 3.14, and the i conversion is applied to it, it will produce 3, because it works by means of the toint function.
Note: a somewhat less flexible mechanism for converting fields, related to fconv, is present in the :fields clause of the awk macro, which can specify names for the positional fields, along with conversion functions. The :fields clause has different syntax, and doesn't support the : (colon) separator, instead assuming a fixed number of fields enumerated from the left.
;; convert up to first three fields to integer:
(awk ((fconv i i i)))
;; convert all fields to floating-point
(awk ((fconv : r :)))
;; convert first and second fields to integer
;; from hexadecimal;
;; convert last field to integer from octal;
;; process pairs of fields in between
;; these by leaving the first element of
;; each pair unconverted and converting second
;; to floating-point;
(awk ((fconv x x : - r : o)))
;; convert all fields, except the first,
;; from integer, turning empty strings
;; and non-integer junk as zero;
;; leave first field unconverted:
(awk ((fconv - : iz)))
(-> path form*)
(->> path form*)
(<- path form*)
(!> command form*)
(<! command form*)
These awk macros provide convenient redirection of output and input to and from files and commands.
When at least one form argument is present, the functions ->, ->> and !> evaluate each form in a dynamic environment in which the *stdout* variable is bound to a file output stream, for the first two functions, or output command pipe in the case of the last one.
Similarly, when at least form argument is present, the remaining functions <- and <! evaluate each form in a dynamic environment in which *stdin* is bound to a file input stream or input command pipe, respectively.
The path and command arguments are treated as forms, and evaluated. They should evaluate to strings.
The first evaluation of one of these macros for a given path or command being used in a particular direction (input or output) and type (file or command) creates a stream. That stream is then associated with the given path or command string, together with the direction and type. Upon a subsequent evaluation of one of these macros for the same path or command string, direction and type, a new stream is not opened; rather, the previously associated stream is used.
The scope of these macros is the entire containing awk form; they may be used in the :let and :fun clauses.
The -> macro indicates that the file named path is to be opened for writing and overwritten, or created if it doesn't exist. The ->> macro indicates that the file named by path is to be opened in append mode, created if necessary. The <- macro indicates that the file given by path is to be opened for reading.
The !> macro indicates that command is to be opened as an output command pipe. The <! macro indicates that command is to be opened as an input command pipe.
If any of these macros is invoked without any form arguments, then it yields the stream object associated with path or command argument, direction and type. If the association doesn't exist, the stream is first created.
If form arguments are present, then the value of the last one is yielded as a value, except in the case when the last form yields the :close keyword symbol.
If the last form yields the :close keyword symbol, the association between the path or command, direction and type and the stream is removed, and the stream is closed. In this case, the result value of the macro isn't the :close symbol, but rather the return value of the close-stream call that is implicitly applied to the stream.
Even if there is only one form which yields :close, the stream is created, if it doesn't exist prior to the macro invocation.
In each invocation of these macros, after every form is evaluated, the stream is implicitly flushed, if it is an output stream.
The association between the pipe or command strings, direction and type is scoped to the innermost enclosing awk macro. An inner awk macro cannot refer to the associations established in an outer awk macro. An outer awk macro can obtain an association's stream object and communicate that stream to the nested macro where it can be used.
When the surrounding awk macro terminates, all of the streams opened by these redirection macros are closed, without breaking those associations. If lexical closures are captured inside the macro, and then invoked after the macro has terminated, and inside those closures the redirection macros are used, those macro instances will with closed stream objects, and so attempts to perform I/O will fail.
;; print lines with fields separated by ofs,
;; and [f 2] converted to integer:
(awk ((and [f 2] (fconv - - iz) (> [f 2] 5))))
;; print strictly original lines from orec
(awk ((and [f 2] (fconv - - iz) (> [f 2] 5))
(prn orec)))
(awk (#/(G|D)(2\d[\w]*)/))
Note the subtle flaw here: the
[\w]*
portion of the regular expression contributes nothing
to what lines are matched. The following example
has a similar flaw.
(awk (#/(G|D)([\d\w]*)/))
(awk (:let (r #/xyz/))
((and [f 3] [r [f 1]] (not [r [f 3]]))))
(awk (:let (r (regex-compile "\\\\")))
((and [f 1] [r [f 1]])))
;; original: {OFS=":";print $(NF-1), $NF}
;;
(awk (t (set ofs ":") (prn [f -2] [f -1])))
Note that the above behaves
more correctly than the original Awk example because in the
when there is only one field,
$(NF-1)
reduces to
$0
which refers to the entire record, not to the field.
This sort of bug is why the TXR Lisp
awk
does not imitate the design decision to make the record
the first numbered field.
;; original:
;; {s += $1}
;; END {print "sum is ", s, " average is", s/NR}
;;
(awk (:let (s 0) (n 0))
([f 0] (fconv r) (inc s [f 0]) (inc n))
(:end (prn `sum is @s average is @(/ s n)`)))
Note that the original is not robust against blank lines in the input. Blank lines are treated as if they had a first column field of zero, and are counted toward the denominator in the calculation of the average.
(awk ((rng #/start/ #/stop/)))
(awk (:let prev)
((nequal [f 0] prev) (prn) (set prev [f 0])))
(awk (:begin (prn `@{*args* " "}`)))
Note: if this is evaluated in the command line, for instance with the -e option, an explicit exit is required to prevent the arguments from being processed by TXR after awk completes:
;; Process variable as if it were a file:
(awk (:inputs (make-string-input-stream
(getenv "PATH")))
(:set fs ":")
(t (tprint f)))
;; Just get, split and print; awk macro is irrelevant
(awk (:begin (tprint (split-str (getenv "PATH") ":"))))
(awk (:let (n (toint n)))
(#/Page/ (set [f 1] (pinc n)))
(t))
the command line:
txr -Dn=5 prog.tl input
prints the file, filling in page numbers starting at 5.
Note that environment variable names, their values, and command-line arguments are all regarded as being externally encoded in UTF-8. TXR performs the encoding and decoding automatically.
The *args-full* variable holds the original, complete list of arguments passed from the operating system, including the program executable name.
During command-line-option processing, TXR may transform the argument list. The hash-bang mechanism, and the --args and --eargs options can inject new command-line arguments, as can code which is executed during argument processing via the -e options and others.
The *args-eff* variable holds the list of effective arguments, which is the argument list after these transformations are applied. This variable is established and set to the same value as *args-full* prior to command-line processing, but is not updated with its final value until after command-line processing.
The *args* variable holds a list of strings representing the remaining arguments which follow any options processed by the TXR executable, and the script name. This list is a suffix of *args-eff*. Thus, the arguments before *args* can be calculated using the expression (ldiff *args-eff* *args*).
The *args* variable is available to TXR Lisp expressions invoked from the command line via the -p, -e and other such options. During these evaluations, *args* holds all the remaining options, after the invoking option and its argument expression. In other words, code executed from the command line has access to the remaining arguments which follow it. Furthermore, this code may modify the value of *args*. Such a modification is visible to the option processing code. That is to say code executed from the command line can rewrite the remaining list of arguments, and that list takes effect.
(env)
The env function retrieves the list of environment variables. Each variable is represented by a single entry in the list: a string which contains an = (equal) character somewhere, separating the variable name from its value.
Multiple calls to env may return the same list, or lists which share structure.
If a list returned by env is modified, the behavior is unspecified.
See also: the env-hash function.
(env-hash)
The env-hash function returns an :equal-based hash whose keys and values are strings. The hash table is populated with the environment variables, represented as key-value character string pairs.
The env-hash function allocates the hash table when it is first invoked; thereafter, it returns the same hash table.
The hash table is updated by the functions setenv, unsetenv and getenv.
Note: calls to the underlying C library functions setenv and getenv, and other direct manipulations of the environment, will not update the hash table.
(getenv name)
(setenv name value [overwrite-p])
(unsetenv name)
These functions provide access to, as well as manipulation of, environment variables. Of these three, setenv and unsetenv might not be available on some platforms, or unsetenv might be be present in a simulated form which sets the variable name to the empty string rather than deleting it.
The getenv function searches the environment for the environment variable whose name is name. If the variable is found, its value is returned. Otherwise nil is returned.
The setenv function creates or modifies the environment variable indicated by name. The value string argument specifies the new value for the variable. If value is nil, then setenv behaves like unsetenv, except that it observes the overwrite-p argument. That is to say, the meaning of a null value is that the variable is to be removed.
If the overwrite-p argument is specified, and is true, then the variable is overwritten if it already exists. If the argument is false, then the variable is not modified if it already exists. If the argument is not specified, it defaults to the value t, effectively giving rise to a two-argument form of setenv which creates or overwrites environment variables.
A variable removal is deemed to be an overwrite. Thus if both value and overwrite-p are nil, then setenv does nothing.
The setenv function unconditionally returns value regardless of whether or not it overwrites or removes an existing variable.
The unsetenv function removes the environment variable specified by name, if it exists. On some platforms, it instead sets the environment variable to the empty string.
Note: supporting removal semantics in setenv allows for the following simple save/modify/restore pattern:
(let* ((old-val (getenv "SOME-VAR")))
(unwind-protect
(progn (setenv "SOME-VAR" new-val)
...)
(setenv "SOME-VAR" old-val)))
This works in the case when SOME-VAR exists, as well as in the case that it doesn't exist. In both cases, its previous value or, respectively, non-existence, is restored by the unwind-protect cleanup form.
These functions interact with the list returned by the env function and with the hash table returned by the env-hash function as follows.
A previously returned list returned by env is not modified. The setenv and unsetenv functions may cause a subsequent call to env to return a different list. The getenv function has no effect on the list.
The hash table previously returned by env-hash is modified by setenv in the manner consistent with its semantics. A new entry is created in the table, if required, and an existing entry is overwritten only if the overwrite-p flag is specified. Likewise, if setenv is invoked in a way that causes the environment variable to be deleted, it is removed from the hash also. The unsetenv function causes the variable to be removed from the hash table also. The getenv function accesses the underlying environment and updates the hash table with the name-value pair which is retrieved.
(replace-env env-list)
The replace-env function replaces the environment with the environment variables specified in env-list. The argument is a list of character strings, in the same format as the list returned by the env function: each element of the list describes an environment variable as a single character string in which the name is separated by the value by the = character. As a special concession, if this character is missing, the replace-env function treats that entry as being a name with an empty value.
The replace-env first empties the existing environment, rendering it devoid of environment variables. Then it installs the entries specified in env-list.
The return value is env-list.
Note: replace-env may be used to specify an exact environment to child programs executed by functions like open-process, sh or run.
Note: the previous environment may be saved by calling env and retaining the returned list. Then after modifying the environment, the original environment can be restored by passing that retained list to replace-env.
The *child-env* variable specifies the list of command-line variables established for programs executed via the functions exec, run, sh, open-command and open-process.
The initial top-level value of this variable is the symbol t which indicates that *child-env* is to be ignored, such that the executed program inherits the current set of environment variables.
If *child-env* has any other value, it must be a possibly empty list of environment variables, in the same format as what is returned by env function and accepted by replace-env. That value completely specifies the environment that executed programs shall receive.
(let ((*child-env* '("a=b")))
;; /usr/bin/env sees only "a" environment variable
(get-lines (open-process "/usr/bin/env" "r")))
-> ("a=b")
TXR Lisp provides a support for recognizing, extracting and validating the POSIX-style options from a list of command-line arguments.
The supported options can be defined as a list of option descriptor objects each of which is constructed by a call to the opt function. Each option can have a long name, a short name, a type, and a description.
The getopts function takes a list of option descriptors, and a list of arguments, producing a parse, or else throwing an exception of type opt-error if an error is detected. The returned object, an instance of struct type opts, can then be queried for specific option values, or for the remaining non-option arguments.
The opthelp function takes a list of option descriptors and an output stream, and generates help text on that stream. A program supporting a --help option can use this to generate that portion of its help text which describes the available options. Also provided are functions opthelp-conventions and opthelp-types, which have the same interface as opthelp and print additional information. These may be used together with opthelp to provide more detailed help under a single --help option, or under separate options like --extra-help.
The define-option-struct macro provides a more streamlined, declarative mechanism built on the same facility. The options are declared in a more condensed way, and using symbols instead of strings. Furthermore, the parsed option values become slot values of an object, named by the same symbols.
A command-line option can have a short or long name. A short name is always one character long, and treated specially in the command-line syntax. Long options have names two or more characters long. An option can have both a long and short name. Options may not begin with the - (ASCII dash) character. A long option may not contain the = character.
Short options are invoked by specifying an argument with a single leading - followed by the option character. Multiple short options which take no argument can be "clumped": combined into a single argument consisting of a single - followed by multiple short option characters.
An option can take an argument, in which case the argument is required. An option which takes no argument is Boolean, and a Boolean option never takes an argument: "takes no argument" and "Boolean" effectively mean the same thing.
Long options are invoked as an argument which begins with a -- (double dash) immediately followed by the name. When a long option takes an argument, it is mandatory. It must be specified in the same argument, separated from the name by the = character. If that is omitted, then the next command-line argument is taken as the argument. That argument is removed, and not recognized as an option, even if it looks like one.
A Boolean long option can be explicitly specified as false using the --no- prefix rather than the -- prefix.
Short options may be invoked using long name syntax; if a is a short option, then it may be referenced on the command line as --a and treated as a long option in all other ways, including the use of --no- to explicitly specify false for a Boolean option.
If a short option takes an argument, it may not clump with other short option. The following command-line argument is taken as the options argument. That argument is removed and is not recognized as an option even if it looks like one.
If the command-line argument -- occurs in the command line where an option would otherwise be recognized, it signifies the end of the options. The subsequent arguments are the non-option arguments, even if they resemble options.
The following example illustrates a complete TXR Lisp program which parses command-line options:
(defvarl options
(list (opt "v" "verbose" :dec
"Verbosity level. Higher values produce more chatter.")
(opt nil "help" :bool
"List this help text.")
(opt "x" nil :hex
"The X factor: a number with a mysterious\ \
interpretation, affecting the program\ \
behavior in strange ways.")
(opt "z" nil) ;; undocumented option
(opt nil "cee" :cint
"C style integer.")
(opt "g" "gravity" :float
"Gravitational constant. This gives\ \
the gravitational field\ \
strength at the Earth's surface.")
(opt "l" "lit" :str
"A character string given in TXR Lisp notation.")
(opt "c" nil 'upcase-str
"Custom treatment: ARG is converted to uppercase.")
(opt "b" "bool" :bool
"A flag you can flip true.")))
(defvarl prog-name *load-path*)
(let ((o (getopts options *args*)))
(when [o "help"]
(put-line "Usage:\n")
(put-line ` @{prog-name} [options] arg*`)
(opthelp options)
(exit 0))
(put-line `args after opts are: @{o.out-args ", "}`))
The next example is equivalent to the previous, but using the define-option-struct macro:
(define-option-struct prog-opts nil
(v verbose :dec
"Verbosity level. Higher values produce more chatter.")
(nil help :bool
"List this help text.")
(nil extra-help
:bool
"List help text with more detailed information.")
(x nil :hex
"The X factor: a number with a mysterious\ \
interpretation, affecting the program\ \
behavior in strange ways.")
;; undocumented Boolean:
(z nil)
(nil cee :cint
"C style integer.")
(g gravity :float
"Gravitational constant. This gives\ \
the gravitational field\ \
strength at the Earth's surface.")
(l lit :str
"A character string given in TXR Lisp notation.")
(c nil upcase-str
"Custom treatment: ARG is converted to uppercase.")
(b bool :bool
"A flag you can flip true."))
(defvarl prog-name *load-path*)
(let ((o (new prog-opts)))
o.(getopts *args*)
(when (or o.help o.extra-help)
(put-line "Usage:\n")
(put-line ` @{prog-name} [options] arg*`)
o.(opthelp)
(when o.extra-help
o.(opthelp-types)
o.(opthelp-conventions))
(exit -1))
(put-line `args after opts are: @{o.out-args ", "}`))
(defstruct opt-desc
short long helptext type
... unspecified slots)
The opt-desc structure describes a single command-line option.
The short and long slots are either nil or else hold strings. The short slot gives the option's short name: a one-character-long string which may not be the ASCII dash character -. The long slot gives the option's long name: a string two or more characters long which doesn't begin with a dash. An option must have at least one of these names.
The helptext slot provides a descriptive string. This string may be long. The opthelp function displays this text, formatting into multiple lines as necessary. If helptext is nil, the option is considered undocumented.
The type slot may be a symbol naming a global function which takes one argument, or it may be such a function object. Otherwise it must be one of the following keyword symbols:
If type is a function, then the option requires an argument. The argument string is passed to the function, and the value is whatever the function returns.
The opt-desc structure may have additional slots which are not specified.
The opt convenience function is provided for constructing opt-desc objects.
(opt short long [type [helptext]])
The opt function provides a slightly condensed syntax for constructing an object of type opt-desc.
The required arguments short and long are strings, corresponding to opt-desc slots of the same name.
The optional parameter type corresponds to the same-named slot and defaults to :bool.
The optional parameter helptext corresponds to the same-named slot and defaults to nil (no help text provided for the option).
The opt function follows this equivalence:
(opt a b c d) <--> (new opt-desc short a long b
type c helptext d)
(defstruct opts nil
in-args out-args
... unspecified slots)
The opts structure represents a parsed command line, containing decoded information obtained from the options and an indication of where the non-option arguments start.
The opts structure supports direct indexing for option retrieval. That is the only documented interface for accessing the parsed options; the implementation of the information set describing the parsed options is unspecified.
The in-args slot holds the original argument list.
The out-args slot holds the tail of the argument list consisting of the non-option arguments.
The mechanism by means of which out-args is calculated, and by means of which the information about the options is populated, is unspecified. The only interface to that mechanism is the getopts function.
The opts object supports indexing, including indexed assignment.
If o is an instance of opts returned by getopts, then the expression [o "v"] tests whether the option "v" is available in o; that is, whether it has been specified in the command line. If so, then its associated value is returned, otherwise nil is returned. This nil is ambiguous: for a Boolean option it indicates that either the option was not specified, or that it was explicitly specified as false. For a Boolean option that was specified (positively), the value t is returned.
The expression [o "v" dfl] yields the value of option "v" if that option has been specified. If the option hasn't been specified, then the expression yields the value dfl.
Assigning to [o "v"] is possible. This replaces the value associated with option "v". The assignment is erroneous if no such option was parsed from the command line, even if it is a valid option.
If an option is defined with both a long form and a short form, and either form of that option occurs in the command line being processed, then the option appears under both names in the index.
For instance if option "--verbose" has the short form "-v", and either option occurs, then both the keys "v" and "verbose" will exist in the opts structure returned by getopts. Note that this behavior is different from that of the structure produced define-option-struct macro. Under that approach, if an option is defined with a long and short name, the structure will have only a single slot for that option, named after the long name.
(getopts option-desc-list arg-list)
The getopts function takes a list of opt-desc structures and a list of strings arg-list representing command-line arguments.
The arg-list is parsed. If the parse is unsuccessful, an exception of type opt-error is thrown, derived from error.
If there are problems in option-desc-list itself, then an exception of type error is thrown.
If the parse is successful, getopts returns an instance of the opts structure describing the parsed options and listing the non-option arguments.
(opthelp opt-desc-list [stream])
(opthelp-types opt-desc-list [stream])
(opthelp-conventions opt-desc-list [stream])
The opthelp function processes the list of opt-desc structures opt-desc-list and compiles a customized body of help describing all of the options which have help text. These are presented in alphabetical order. Options which do not have help text, if any, are simply listed together under a heading which indicates their undocumented status.
The text is formatted to fit within 79 columns, and begins and ends with a blank line. Its format consists of headings which begin in the first column, and paragraphs and tables which feature a two space left margin. A blank line follows each section heading. The heading begins with a capital letter. Its remaining words are uncapitalized, and it ends with a colon.
The text is sent to stream, if specified. This argument defaults to *stdout*.
If there are problems in option-desc-list itself, then an exception of type error is thrown.
The opthelp-types supplementary help function processes the opt-desc-list, considering only those options which are documented. If any of them have typed arguments, then a legend is printed explaining the types. The legend includes only information about those option argument types which appear in opt-desc-list.
The opthelp-conventions supplementary help function processes opt-desc-list, considering only those options which are documented. It prints a guide to the use of options, which includes information only about the kinds of options actually present in opt-desc-list.
(define-option-struct name super opt-specifier*)
The define-option-struct macro defines a struct type, instances of which provide command-line option parsing.
The name and super parameters are subject to the same requirements and have the same semantics as the same-named parameters of defstruct.
The opt-specifier arguments are lists of between two and four elements: (short-symbol long-symbol [type [help-text]]). The short-symbol and long-symbol must be symbols suitable for use as slot names. One of them may be specified as nil indicating that the option has no long form, or no short form.
If a opt-specifier specifies both a short-symbol and a long-symbol then only a slot named by long-symbol shall exist in the structure.
The struct type defined by define-option-struct has four methods: getopts, opthelp, opthelp-types and opthelp-conventions. It also has two slots: in-args and out-args, which function in a manner identical to their same-named counterparts in the opts class.
The getopts method takes a single argument: the argument list to be processed. When the argument list is successfully processed.
The opthelp, opthelp-types and opthelp-conventions methods take an optional stream argument.
Note: to encode the option names "t" or "nil", or option names which clash with the slot names in-args and out-args or the method names such as getopts or opthelp, symbols with these names from a package other than usr must be used.
(errno [new-errno])
(set (errno) new-value)
The errno function retrieves the current value of the C library error variable errno. If the argument new-errno is present and is not nil, then it specifies a value which is stored into errno. The value returned is the prior value.
The place form of errno does not take an argument.
(strerror errno-value)
The strerror returns a character string which provides the host platform's description of the integer errno-value obtained from the errno function.
If the host platform fails to provide a description, the function returns nil.
(exit [status])
The exit function terminates the entire process (running TXR image), specifying the termination status to the operating system. Values of the optional status parameter may be nil, t, or an integer value. The value nil indicates an unsuccessful termination status, whereas t indicates a successful termination status. An absence of the status argument also specifies a successful termination status. If status is an integer value, it specifies a successful termination if it is 0, otherwise the interpretation of the value is platform-specific.
These variables correspond to the POSIX "errno constants", namely E2BIG, EACCES, EADDRINUSE and so forth. Variables corresponding to all of the <errno.h> constants from the Issue 6 2004 edition of POSIX are included. The variables eownerdead and enotrecoverable from Issue 7 2018 are subject to the availability of the corresponding constants in the host platform.
(abort)
The abort function terminates the entire process (running TXR image), specifying an abnormal termination status to the process.
Note: abort calls the C library function abort which works by raising the SIG_ABRT signal, known in TXR as the sig-abrt variable. Abnormal termination of the process is this signal's default action.
(at-exit-call function)
(at-exit-do-not-call function)
The at-exit-call function registers function to be called when the process terminates normally. Multiple functions can be registered, and the same function can be registered more than once. The registered functions are called in reverse order of their registrations.
The at-exit-do-not-call function removes all previous at-exit-call registrations of function.
The at-exit-call function returns function.
The at-exit-do-not-call function returns t if it removed anything, nil if no registrations of function were found.
(usleep usec)
The usleep function suspends the execution of the program for at least usec microseconds.
The return value is t if the sleep was successfully executed. A nil value indicates premature wakeup or complete failure.
Note: the actual sleep resolution is not guaranteed, and depends on granularity of the system timer. Actual sleep times may be rounded up to the nearest 10 millisecond multiple on a system where timed suspensions are triggered by a 100 Hz tick.
(mkdir path [mode])
(ensure-dir path [mode])
mkdir tries to create the directory named path using the POSIX mkdir function. An exception of type file-error is thrown if the function fails. Returns t on success.
The mode argument specifies the request numeric permissions for the newly created directory. If omitted, the requested permissions are #o777 (511): readable and writable to everyone. The requested permissions are subject to the system umask.
The function ensure-dir also creates a directory named path. Unlike mkdir, it also attempt to create all the necessary parent directories, and does not throw an error if path refers to an existing object, if that object is a directory or a symbolic link to a directory. Rather, in that case it returns nil instead of t.
(chdir {path | stream | fd})
chdir changes the current working directory to the object specified by the argument, and returns t, or else throws an exception of type file-error.
If the argument is a string, it is interpreted as a path, in which case the POSIX chdir function is used. If the argument is a stream then an integer file descriptor is retrieved from that stream using the fileno function. That descriptor can be specified directly as a fd argument. In the case of these these two argument types, the fchdir function is used.
(pwd)
The pwd function retrieves the current working directory. If the underlying getcwd C library function fails with an errno other than ERANGE, an exception will be thrown.
(rmdir path)
The rmdir function removes the directory named by path. If successful, it returns t, otherwise it throws an exception of type file-error.
Note: rmdir calls the same-named POSIX function, which requires path to be the name of an empty directory.
(remove-path path [throw-on-error-p])
The remove-path function tries to remove the filesystem object named by path, which may be a file, directory or something else.
If successful, it returns t.
The optional Boolean parameter throw-on-error-p, which defaults to nil.
A failure to remove the object results in an exception of type file-error being thrown, unless the failure reason is that the object indicated by path doesn't exist. In this non-existence case, the behavior is controlled by the throw-on-error argument. If that argument is true, the exception is thrown. Otherwise, the function returns normally, producing the value nil to indicate that it didn't perform a removal.
(rename-path from-path to-path)
The rename-path function tries to rename filesystem path from-path, which may refer to a file, directory or something else, to the path to-path.
If successful, it returns t.
A failure results in an exception of type file-error.
(sh system-command)
(run program [argument-list])
The sh function executes system-command using the system command interpreter. The run function spawns a program, searching for it using the system PATH. Using either method, the executed process receives environment variables from the parent.
TXR blocks until the process finishes executing. If the program terminates normally, then its integer exit status is returned. The value zero indicates successful termination.
The return value nil indicates an abnormal termination, or the inability to run the process at all.
In the case of the run function, if the child process is created successfully but the program cannot be executed, then the exit status will be an errno value from the failed exec attempt.
The standard input, output and error file descriptors of an executed command are obtained from the streams stored in the *stdin*, *stdout* and *stderr* special variables, respectively. For a detailed description of the behavior and restrictions, see the open-command function, whose description of this mechanism applies to the run and sh function also.
Note: as of TXR 120, the sh function is implemented using run and not by means of the system C library function, as previously. The run function is used to invoke the system interpreter by name. On Unix-like systems, the string /bin/sh is assumed to denote the system interpreter, which is expected to support a pair of arguments -c command to specify the command to be executed. On MS Windows, the interpreter is assumed to be the relative pathname cmd.exe and expected to support /C command as a way of specifying a command to execute.
(sh-esc str)
(sh-esc-all str)
(sh-esc-dq str)
(sh-esc-sq str)
The functions sh-esc, sh-esc-all, sh-esc-dq and sh-esc-sq transform the argument string str for safe insertion into commands. These functions are intended for use on POSIX systems, where the command interpreter used by the functions sh and open-command and related functions is the POSIX Shell Command Language.
The sh-esc function adds quoting and escaping into its argument in such a way that the resulting string may be inserted as an argument into a command.
The sh-esc-all function performs a stricter escaping and quoting, such that the transformed string may be inserted into any syntactic context where a textual operand is required for any reason, such as the pattern in the ${var%pattern} construct.
The sh-esc-dq function escapes its argument for insertion into a double-quoted field in a shell command line. It does not add the double quotes themselves.
The sh-esc-dq function escapes its argument for insertion into a single-quoted field in a shell command line. It does not add the single quotes themselves.
The precise set of characters which, according to the sh-esc function, require escaping or quoting, is the following:
| & ; < > ( ) $ ` \ " ' tab newline space * ? [ # ~
If none of these characters occur in str, then sh-esc returns str.
The sh-esc-all function considers all the above characters, and also these:
= %
The sh-esc-dq function escapes the following characters by preceding them with the \ (backslash) character:
$ ` \ "
The sh-esc-sq function replaces every occurrence of the ' character (single quote, apostrophe) with the sequence '\'' (single quote, backslash, single quotes, single quote). This sequence has the effect of terminating the enclosing single-quoted field, then producing a single quote via a backslash escape, and then opening a single-quoted field.
(defstruct stat nil
dev ino mod nlink uid gid
rdev size blksize blocks
atime atime-nsec mtime mtime-nsec
ctime ctime-nsec path)
The stat structure defines the type of object which is returned by the stat and lstat functions. Except for path, atime-nsec, ctime-nsec and mtime-nsec, the slots are the direct counterparts of the members of POSIX C structure struct stat. For instance the slot dev corresponds to st_dev.
The path slot is set by the functions stat and lstat. Its value is nil when the path is not available.
The atime-nsec, ctime-nsec and mtime-nsec fields give the fractional parts of atime, ctime and mtime, respectively. They are derived from the newer style information in which the POSIX function provides the timestamps in struct timespec format. If that is not available from the platform, these fields take on values of zero.
(stat {path | stream | fd} [struct])
(lstat path)
(fstat {path | stream | fd} [struct])
The stat function retrieves information about a filesystem object whose pathname is given by the string argument path, or else about a system object associated with the open stream stream, or one associated with the integer file descriptor fd.
If a stream is specified, that stream must be of a kind from which the fileno function can retrieve a file descriptor, otherwise an exception of type file-error is thrown.
If the object is not found or cannot be accessed, an exception is thrown.
Otherwise, if the struct argument is missing, information is retrieved and returned, in the form of a new structure of type stat. If the struct argument is present, it must be either: an instance of the struct structure type, or of a type derived from that type by inheritance, or else structure type which has all the same slots as the struct type. The retrieved information is stored into struct and that object is returned rather than a new object.
If path refers to a symbolic link, the stat function retrieves information about the target of the link, if it exists, or else throws an exception of type file-error.
The lstat function behaves the same as stat on objects which are not symbolic links. For a symbolic link, it retrieves information about the link itself, rather than its target.
The path slot of the returned structure holds a copy of their path argument value. When information is retrieved using a stream or fd argument, this slot is nil.
The fstat function is an alias for stat.
Note: until TXR 231, stat and fstat were distinct functions: stat accepted only path arguments, whereas fstat function accepted only stream or fd arguments.
The following variables exist, having integer values. These are bitmasks which can be applied against the value given by the mode slot of the stat structure returned by the function stat: s-ifmt, s-ifsock, s-iflnk, s-ifreg, s-ifblk, s-ifdir, s-ifchr, s-ififo, s-isuid, s-isgid, s-isvtx, s-irwxu, s-irusr, s-iwusr, s-ixusr, s-irwxg, s-irgrp, s-iwgrp, s-ixgrp, s-irwxo, s-iroth, s-iwoth and s-ixoth.
These variables correspond to the C language constants from POSIX: S_IFMT, S_IFLNK, S_IFREG and so forth.
The logtest function can be used to test these against values of mode. For example (logtest mode s-irgrp) tests for the group read permission.
(umask [mask])
The umask function provides access to the Unix C library function of the same name, which controls which permissions are denied when files are newly created.
If umask is called with no argument, it returns the current value of the mask.
If the mask argument is present, it must be an integer specifying the new mask to be installed. The previous mask is returned.
If mask is absent, then umask returns the previous mask.
Note: the value of the mask argument may be calculated as a bitwise or of the following constants: s-irwxu, s-irusr, s-iwusr, s-ixusr, s-irwxg, s-irgrp, s-iwgrp, s-ixgrp, s-irwxo, s-iroth, s-iwoth and s-ixoth, which correspond to the POSIX C constants S_IRWXU, S_IRUSR, S_IWUSR, S_IXUSR, S_IRWXG, S_IRGRP, S_IWGRP, S_IXGRP, S_IRWXO, S_IROTH, S_IWOTH and S_IXOTH.
Implementation note: since the umask C library function provides no way to retrieve the current mask without overwriting with a new one, the TXR umask function, when given no argument, simulates the pure retrieval of the mask by calling the C function with an argument of #o777 to temporarily install the maximally safe mask. The value returned is then reinstated as the mask by another call to umask, and that value is also returned.
(makedev minor major)
(minor dev)
(major dev)
The parameters minor, major and dev are all integers. The makedev function constructs a combined device number from a minor and major pair (by calling the Unix makedev function). This device number is suitable as an argument to the mknod function (see below). Device numbers also appear as values of the dev slot of the stat structure.
The minor and major functions extract the minor and major device number from a combined device number.
(chmod target mode)
The chmod function changes the permissions of the filesystem object specified by target. It is implemented in terms of the POSIX functions chmod and fchmod. If mode is a character string representing a symbolic mode, then the function also makes use of stat or fstat and umask.
The permissions are specified by mode, which must be an integer or a string.
An integer mode is a bitwise combination of permission mode bits. The value is passed directly to the POSIX chmod or fchmod function. Note: to construct a mode value, applications may use logior to combine the values of the variables like s-irusr or s-ixoth or take advantage of the well-known numeric structure of POSIX permissions to express them octal in octal notation. For instance the mode #o750 denotes that the owner has read, write and execute permissions, the group owner has read and execute, others have no permission. This value may also be calculated using (logior s-irwxu s-irgrp s-ixgrp).
If the argument to mode is a string, it is interpreted according to the symbolic syntax of the POSIX chmod utility. For instance, a mode value of "a+w,-s" means to give all users (owner, group and others) write permission, and remove the setuid and setgid bits.
The full syntax and semantics of symbolic mode strings is given in the POSIX standard IEEE 1003.1.
The function throws a file-error exception if an error occurs, otherwise it returns t.
The target argument may be a character string, in which case it specifies a pathname in the filesystem. In this case, the POSIX function chmod is invoked.
The target argument may also be an integer file descriptor, or a stream. In these two cases, the POSIX fchmod function is invoked. For a stream target, the integer file descriptor is retrieved from the stream using fileno function.
;; Set permissions of foo.txt to "rw-r--r--"
;; (owner can read and write; group owner
;; and other users can only read).
;; numerically:
(chmod "foo.txt" #o644)
;; symbolically:
(chmod "foo.txt" (logior s-irusr s-iwusr
s-irgrp
s-iroth))
Implementation note: The implementation of the symbolic mode processing is based on the descriptions given in IEEE 1003.1-2018, Issue 7 and also on the chmod program from from GNU Coreutils 8.28: and experiments with its behavior, and its documentation.
(chown target id gid)
(lchown target id gid)
The chown and lchown functions change the user and group ownership of the filesystem object specified by target.
They implemented in terms of the POSIX functions chown, fchown and lchown.
The ownership attributes are specified by uid and gid, both integer arguments.
The existing ownership attributes may be obtained using the stat function.
These functions throw a file-error exception if an error occurs, otherwise they returns t.
The target argument may be a character string, in which case it specifies a pathname in the filesystem. In this case, the same-named POSIX function chown is invoked by chown, whereas lchown likewise invokes its respective same-named POSIX counterpart. The difference is that if target is a pathname denoting a symbolic link, then lchown operates on the symbolic link, whereas chown dereferences the symbolic link.
The target argument may also be an integer file descriptor, or a stream. In these two cases, the POSIX fchown function is invoked by either function. For a stream target, the integer file descriptor is retrieved from the stream using fileno function.
Note: in most POSIX systems, unprivileged processes may not change the user ownership denoted by uid. They may change the group ownership indicated in gid, if that value corresponds to the effective group ID of the calling process or one of its ancillary group IDs.
To avoid trying to change the user ownership (and therefore failing), the caller should specify a uid value which matches the object's existing owner.
(utimes target atime-s atime-ns mtime-s mtime-ns)
(lutimes target atime-s atime-ns mtime-s mtime-ns)
The functions utimes and lutimes change the access and modification timestamps of a file indicated by the target argument.
The difference between the two functions is that if target is the pathname of a symbolic link, then lutimes operates on the symbolic link itself, whereas utimes resolves the symbolic link.
Note: the full, complete functionality of these functions requires the platform to provide the POSIX functions futimens and utimensat functions. If these functions are not available, then other functions are relied on, with some reductions in functionality, that are documented below.
The target argument specifies the file to operate on. It may be an integer file descriptor, an open stream, or a character string representing a pathname.
The atime-s and mtime-s parameters specify the whole seconds part of the new access and modification times, expressed as seconds since the epoch.
The atime-ns and mtime-ns parameters specify the fractional part of the access and modification times, expressed in nanoseconds. If an integer argument is given to these parameters, it must lie in the range 0 to 999999999, or else the symbols nil or t may be passed as arguments.
If the symbol nil is passed as the nanoseconds part of the access or modification time, then the access or modification time, respectively, shall not be modified by the operation. The corresponding seconds argument is ignored.
If the symbol t is passed as the nanoseconds part of the access or modification time, then the access or modification time, respectively, shall be obtained from the current system time. The corresponding seconds argument is ignored.
If the utimensat and futimens functions are not available from the host system, then the above nil and t convention in the nanoseconds arguments is not supported; the function will fail by throwing an exception if an attempt is made to pass these arguments.
If the utimensat and futimens functions are not available from the host system, then operating on a symbolic link with lutimes is only possible if the system provides the lutimes C library function, otherwise the operation fails by throwing an exception (if given a path argument for target, even if that path isn't a symbolic link).
If the implementation falls back on the utimes, futimes, and lutimes functions, then the nanoseconds arguments are truncated to microsecond precision.
If the implementation falls back on utime, then the nanoseconds arguments are ignored; the times are effectively truncated to whole seconds.
(mknod path mode [dev])
The mknod function tries to create an entry in the filesystem: a file, FIFO, or a device special file, under the name path. If it is successful, it returns t, otherwise it throws an exception of type file-error.
The mode argument is a bitwise or combination of the requested permissions, and the type of object to create: one of the constants s-ifreg, s-ififo, s-ifchr, s-ifblk or s-ifsock. The permissions are subject to the system umask.
If a block or character special device (s-ifchr or s-ifblk) is being created, then the dev argument specifies the major and minor numbers of the device. A suitable value can be constructed from a major and minor pair using the makedev function.
;; make a character device (8, 3) called /dev/foo
;; requesting rwx------ permissions
(mknod "dev/foo" (logior #o700 s-ifchr) (makedev 8 3))
(mkfifo path mode)
The mkfifo function creates a POSIX FIFO object. If it is successful, it returns t, otherwise it throws an exception of type file-error.
The mode argument is a bitwise or combination of the requested permissions, and is subject to the system umask.
Note: the mknod function can also create FIFOs, specified via the bitwise combination of the s-ififo type and the permission mode bits.
(symlink target path)
(link target path)
(rlink target path)
The symlink function creates a symbolic link called path whose contents are the absolute or relative path target. target does not actually have to exist.
The link function creates a hard link. The object at target is installed into the filesystem at path also.
The rlink function is like link except that if target is a symbolic link, it is resolved, and the link is made to the resulting object. On Linux, and some other platforms link will create a hard link to the symbolic link. The behavior is not specified by POSIX.
If these functions succeed, they return t. Otherwise they throw an exception of type file-error.
(readlink path)
If path names a filesystem object which is a symbolic link, the readlink function reads the contents of that symbolic link and returns it as a string. Otherwise, it fails by throwing an exception of type file-error.
(realpath path)
The realpath function provides access to the same-named POSIX function. It processes the input string path by expanding all symbolic links, removes all superfluous ".." and "." path components, and extra component-separating slash characters, to produce a canonical absolute pathname.
If the underlying POSIX function indicates failure, then nil is returned. In that situation the errno value is available using the errno function.
Functions in this category are complex functionality implemented using a combination of multiple calls into the host system's POSIX API.
(copy-file from-path to-path [perms-p [times-p]])
(copy-file from-list to-dir [perms-p [times-p]])
The copy-file function creates a replica of the file from-path at the destination path to-path.
Both paths are opened using open-file in binary mode, as if using (open-file from-path "b") and (open-file to-path "wb") respectively. Then bytes are read from one stream and written to the other, in blocks which whose size is a power of two at least as large as 16834.
If the optional Boolean parameter perms-p is specified, and is true, then the permissions of from-path are propagated to to-path.
If the optional Boolean parameter times-p is specified, and is true, then the access and modification timestamps of from-path are propagated to to-path.
The copy-file function returns nil if it is successful, and throws an exception derived from file-error on failure.
The copy-files function copies multiple files, whose pathnames are given by the list argument from-list into the target directory whose path is given by to-dir.
The target directory must exist.
For each source path in from-list, the copy-files function forms a target path by combining the base name of the source path with target-dir. (See the base-name and path-cat functions). Then, the source path is copied to the resulting target path, as if by the copy-file function.
The copy-files function returns nil if it is successful, and throws an exception derived from file-error on failure.
Additionally, copy-files provides an internal catch for the retry and skip restart exceptions. If the caller, using a handler frame established by handle, catches an error emanating from the copy-files function, it can retry the failed operation by throwing the retry exception, or continue copying with the next file by throwing the skip exception.
;; Copy all "/mnt/cdrom/*.jpg" files into "images" directory,
;; preserving their time stamps,
;; continuing the operation in the face of
;; file-error exceptions.
(handle
(copy-files (glob "/mnt/cdrom/*.jpg") "images" nil t)
(file-error (throw 'skip)))
(cat-files to-path from-path*)
The cat-files function catenates the contents of zero or more files into one file. The destination path is specified by to-path. Regardless of whether there are any from-path arguments, the file named by to-path is created, if necessary or else truncated to zero length. Then, the files named by each from-path are traversed in left-to-right order; the contents of each file is appended to the destination file.
(copy-path-rec from-path to-path option*)
The copy-path-rec function replicates a file system object identified by the pathname from-path, creating a similar object named to-path.
If from-path is a directory, it is recursively traversed and its structure and content is replicated under to-path.
The option arguments are keywords, which may be the following:
The copy-path-rec function creates all necessary pathname components required for to-path to come into existence, as if by using the ensure-dir function.
Whenever an object under from-path has a counterpart in to-path which already exists, the situation is handled as follows:
Special objects such as FIFOs, character devices, block devices and sockets are copied by creating a new, similar objects at the destination path. In the case of devices, the major and minor numbers of the copy are derived from the original, so that the copy refers to the same device. However, the copy of a socket or a FIFO is effectively a new, different endpoint because these objects are identified by their pathname. Processes using the copy of a socket or a FIFO will not connect to processes which are working with the original.
The copy-path-rec function returns nil if it is successful. It throws an exception derived from file-error when encountering failures.
Additionally copy-path-rec provides an internal catch for the retry and skip restart exceptions. If the caller, using a handler frame established by handle, catches an error emanating from the copy-files function, it can retry the failed operation by throwing the retry exception, or continue copying with the next object by throwing the skip exception.
(remove-path-rec path)
The remove-path-rec function attempts to remove the filesystem object named by path. If path refers to a directory, that directory is recursively traversed to remove all of its contents, and is then removed.
The remove-path-rec function returns nil if it is successful. It throws an exception derived from file-error when encountering failures.
Additionally remove-path-rec provides an internal catch for the retry and skip restart exceptions. If the caller, using a handler frame established by handle, catches an error emanating from the copy-files function, it can retry the failed operation by throwing the retry exception, or continue removing other objects by throwing the skip exception. Skipping a failed remove operation may cause subsequent operations to fail. Notably, the failure to remove an item inside a directory means that removal of that directory itself will fail, and ultimately, path will still exist when remove-path-rec completes and returns.
(chmod-rec path mode)
(chown-rec path uid gid)
The chmod-rec and chown-rec functions are recursive counterparts of chmod and lchown.
The filesystem object given by path is recursively traversed, and each of its constituent objects is subject to a permission change in the case of chown-rec, or an ownership change in the case of chown-rec.
The chmod-rec function alters the permission of each object that is not a symbolic link using the chmod function, and mode is interpreted accordingly: it may be an integer or string. Each object which is a symbolic link is ignored.
The chown-rec function alters the permission of each object encountered, including symbolic links, using the lchown function.
These functions establish restart catches, similarly to remove-path-rec and copy-path-rec, allowing the caller to retry individual failed operations or skip the objects on which operations have failed.
(touch path [ref-path])
The touch function updates the modification timestamp of the filesystem object named by path. If the object doesn't exist, it is created as a regular file.
If ref-path is specified, then the modification timestamp of the object denoted by path is updated to be equivalent to the modification timestamp of the object denoted by ref-path. Otherwise ref-path being absent, the modification timestamp of path is set to the current time.
If path is a symbolic link, it is dereferenced; touch operates on the target of the link.
(mkdtemp prefix)
The mkdtemp function combines the prefix, which is a string, with a generated suffix to create a unique directory name. The directory is created, and the name is returned.
If the prefix argument ends in with a sequence of one or more X characters, the behavior is unspecified.
Note: this function is implemented using the same-named POSIX function. Whereas the POSIX function requires the template to end in a sequence of at least six X characters, which are replaced by the generated suffix, the TXR Lisp function handles this detail internally, requiring only the prefix part without those characters.
(mkstemp prefix [suffix])
The mkstemp function create a unique file name by adding a generated infix between the prefix and suffix strings. The file is created, and a stream open in "w+b" mode for the file is returned.
If either the prefix or suffix contain X characters, the behavior is unspecified.
If suffix is omitted, it defaults to the empty string.
The name of the file is available by interrogating the returned stream's :name property using the function stream-get-prop.
Notes: this function is implemented using the POSIX function mkstemp or, if available, using the mkstemps function which is not standardized, but appears in the GNU C Library and some other systems. If mkstemps is unavailable, then the suffix functionality is not available: the suffix argument must either be omitted, or must be an empty string.
Whereas the C library functions require the template to contain a sequence at least six X characters, which are replaced by the generated portion, the TXR Lisp function handles this detail internally, requiring no such characters in any of its inputs.
Functions in this category perform various tests on the attributes of filesystem objects.
The functions all have a path parameter, which accepts three types of arguments. If a character string is specified, it denotes a filesystem path to be probed for properties such as ownership and permissions. The object is probed using the stat function except in the case of path-symlink-p which uses lstat. If instead a stream is specified as path, then the associated filesystem descriptor is probed for these properties. If an integer value is specified, it is treated as a POSIX open file descriptor that is to be probed. Otherwise, a stat structure, for example one returned by the stat or lstat function may be specified, in which case no system object is probed. The properties to be tested are those given in the stat object.
Note: in a situation when it is necessary to use any of these functions to probe the properties of a symbolic link itself (other than the function path-symlink-p which does so implicitly) it is necessary to first invoke lstat on the symlink's path, and then pass the resulting stat structure to that function instead of the path.
Some of the accessibility tests (functions which determine whether the calling process has certain access rights) may not be perfectly accurate, since they are based strictly on portable information available via stat, together with the basic, portable POSIX APIs for inquiring about security credentials, such as getuid. They ignoring any special permissions which may exist such as operating system and file system specific extended attributes (for example, file immutability connected to a "secure level" and such) and special process capabilities not reflected in the basic credentials.
With the exception of two functions, the accessibility tests use the real credentials of the caller, rather than the effective credentials. Thus, in a setuid process, where the real and effective privileges are different, the access tests inquire about whether the real user has the given access, not the effective user. In this aspect, the functions are similar to the POSIX access function which also uses real credentials. The functions path-private-to-me-p and path-strictly-private-to-me-p use effective credentials, because they answer a different question: can the given filesystem object be trusted? The trust has to be determined from the point of view of the effective user, because security-sensitive actions are being performed in their context; and the effective user does not trust the real user.
(path-exists-p path)
The path-exists-p function returns t if path is a string which resolves to a filesystem object. Otherwise it returns nil. If the path names a dangling symbolic link, it is considered nonexistent.
If path is an object returned by stat or lstat, path-exists-p unconditionally returns t.
(path-file-p path)
(path-dir-p path)
(path-symlink-p path)
(path-blkdev-p path)
(path-chrdev-p path)
(path-sock-p path)
(path-pipe-p path)
path-file-p tests whether path exists and is a regular file.
path-dir-p tests whether path exists and is a directory.
path-symlink-p tests whether path exists and is a symbolic link.
Similarly, path-blkdev-p tests for a block device, path-chrdev-p for a character device, path-sock-p for a socket and path-pipe-p for a named pipe.
(path-dir-empty path)
The path-dir-empty function returns t if path is an empty directory.
Implementation note: this function performs a test similar to path-dir-p; then, if it is confirmed that path is a directory, a directory stream is opened and entries are read. If an entry is seen which has a name other than "." or ".." then it is concluded that the directory is not empty and nil is returned. If no such entry is seen, then the directory is deemed empty and t is returned.
(path-setgid-p path)
(path-setuid-p path)
(path-sticky-p path)
path-setgid-p tests whether path exists and has the set-group-ID permission set.
path-setuid-p tests whether path exists and has the set-user-ID permission set.
path-sticky-p tests whether path exists and has the "sticky" permission bit set.
(path-mine-p path)
(path-my-group-p path)
path-mine-p tests whether path exists, and is effectively owned by the calling process; that is, it has a user ID equal to the real user ID of the process.
path-my-group-p tests whether path exists, and is effectively owned by a group to which the calling process belongs. This means that the group owner is either the same as the real group ID of the calling process, or else is among the supplementary group IDs of the calling process.
(path-readable-to-me-p path)
path-readable-to-me-p tests whether the calling process can read the object named by path. If necessary, this test examines the real user ID of the calling process, the real group ID, and the list of supplementary groups.
(path-writable-to-me-p path)
path-writable-to-me-p tests whether the calling process can write the object named by path. If necessary, this test examines the real user ID of the calling process, the real group ID, and the list of supplementary groups.
(path-read-writable-to-me-p path)
path-readable-to-me-p tests whether the calling process can both read and write the object named by path. If necessary, this test examines the real user ID of the calling process, the real group ID, and the list of supplementary groups.
(path-executable-to-me-p path)
path-executable-to-me-p tests whether the calling process can execute the object named by path, or perform a search (name lookup, not implying sequential readability) on it, if it is a directory. If necessary, this test examines the real user ID of the calling process, the real group ID, and the list of supplementary groups.
(path-private-to-me-p path)
(path-strictly-private-to-me-p path)
The path-private-to-me-p and path-strictly-private-to-me-p functions report whether the calling process can rely on the object indicated by path to be, respectively, private or strictly private to the security context implied by its effective user ID.
"Private" means that beside the effective user ID of the calling process and the superuser, no other user ID has write access to the object, and thus its contents may be trusted to be be free from tampering by any other user.
"Strictly private" means that not only is the object private, as above, but users other than the effective user ID of the calling process and superuser also not not have read access.
The rules which the function applies are as follows:
A file to be examined is initially assumed to be strictly private.
If the file is not owned by the effective user ID of the caller, or else by the superuser, then it is not private.
If the file grants write permission to "others", then it is not private.
If the file grants read permission to "others", then it is not strictly private.
If the file grants write permission to the group owner, then it is not private if the group contains names other than that of the file owner or the superuser.
If the file grants read permission to the group owner, then it is not strictly private if the group contains names other than that of the file owner or the superuser.
Note that this interpretation of "private" and "strictly private" is vulnerable to the following time-of-check to time-of-use race condition with regard to the group check. At the time of the check, the group might be empty or contain only the caller as a member. But by the time the file is subsequently accessed, the group might have been innocently extended by the system administrator to include additional users, who can maliciously modify the file.
Another issue is that if any components of path can be subverted by another user, test may not be trusted. It becomes vulnerable to a time-of-check to time-of-use race condition.
The function path-components-safe function is provided to perform a security check on an entire path.
(path-components-safe path)
On Unix platforms, the path-components-safe performs a security check on an entire relative or absolute path, returning t if the entire path is examined without encountering an error, and the check passes, otherwise nil. On native Microsoft Windows, the function unconditionally returns true.
An exception may be thrown if an an inaccessible or nonexistent path component is encountered, too many symbolic links have to be resolved or there is some other problem preventing the traversal of path.
The objective of this function is to determine that every portion of path is writable only to the effective user: that if the path is used for filesystem access, its meaning cannot be altered by an adversarial user who is able to control a symbolic link or a directory component.
The function expands symbolic links on its own, one level at a time, and walks the components coming from a link target.
Note: directories which are owned by root, and have the sticky bit, as is the usual configuration of tmp are considered safe, even though multiple users have write permissions.
(path-newer left-path right-path)
(path-older left-path right-path)
The path-newer function compares two paths or stat results by modification time. It returns t if left-path exists, and either right-path does not exist, or has a modification time stamp in the past relative to left-path.
The path-older function is equivalent to path-newer with the arguments reversed.
Note: path-newer takes advantage of subsecond timestamp resolution information, if available. The implementation is based on using the mtime-nsec field of the stat structure, if it isn't nil.
(path-same-object left-path right-path)
The path-same-object function returns t if left-path and right-path resolve to the same filesystem object: the same inode number on the same device.
(path-search name [search-path])
The path-search function searches for the existence of a filesystem object named by name in the directories specified search-path.
If name is the empty string or one of the two strings "." (dot) or ".." (dotdot), then nil is returned. If name contains any path separator characters (any of the set of characters found in the path-sep-chars string) then the function returns name without performing any search. In all these trivial cases, the search-path argument is ignored.
The search-path argument, if present, may be a string or a list of strings. If omitted, then it takes on the value of the PATH environment variable if that variable exists, or else takes on the value nil indicating an empty search path.
If search-path is a string, it is converted to a list of directories by splitting on the separator character, which may be : (colon) or ; (semicolon) depending on the system. Then, for each directory in the list, path-search affixes the name to that component, as if using the path-cat function, and tests whether the resulting path refers to an existing filesystem object. If so, then the search terminates and that resulting path is returned. If the entire list is traversed without finding a filesystem object, then nil is returned. If any error whatsoever occurs while determining whether the resulting path exists, the situation is treated as nonexistence, and the search continues.
Note: subtle discrepancies may exist between path-search and the host platform's mechanisms for searching for an executable program. For instance, since path-search is interested in existence only, it may return a path which exists, but is not executable. Whereas a path searching implementation which tests for executability will in that case continue searching, and not return that path.
(getuid)
(geteuid)
(getgid)
(getegid)
These functions directly correspond to the POSIX C library functions of the same name. They retrieve the real user ID, effective user ID, real group ID and effective group ID, respectively, of the calling process.
(setuid uid)
(seteuid uid)
(setgid gid)
(setegid gid)
These functions directly correspond to the POSIX C library functions of the same name. They set the real user ID, effective user ID, real group ID and effective group ID, respectively, of the calling process. On success, they return t. On failure, they throw an exception of type system-error.
(getgroups)
The getgroups function retrieves the list of supplementary group IDs of the calling process by calling the same-named POSIX C library function.
Whether or not the effective group ID retrieved by getegid is included in this list is system-dependent. Programs should not depend on its presence or absence.
(setgroups gid-list)
The setgroups function corresponds to a C library function found in some Unix operating systems, complementary to the getgroups function. The argument to gid-list must be a list of numeric group IDs. If the function is successful, this list is installed as the list of supplementary group IDs of the calling process, and the value t is returned. On failure, it throws an exception of type system-error.
(getresuid)
(getresgid)
These functions directly correspond to the POSIX C library functions of the same names available in some Unix operating systems. Each function retrieves a three element list of numeric IDs. The getresuid function retrieves the real, effective and saved user ID of the calling process. The getresgid function retrieves the real, effective and saved group ID of the calling process.
(setresuid real-uid effective-uid saved-uid)
(setresgid real-gid effective-gid saved-gid)
These functions directly correspond to the POSIX C library functions of the same names available in some Unix operating systems. They change the real, effective and saved user ID or group ID, respectively, of the calling process.
A value of -1 for any of the IDs specifies that the ID is not to be changed.
Only privileged processes may arbitrarily change IDs to different values.
Unprivileged processes are restricted in the following way: each of the new IDs that is replaced must have a new value which is equal to one of the existing three IDs.
(defstruct passwd nil
name passwd uid gid
gecos dir shell)
The passwd structure corresponds to the C type struct passwd. Objects of this struct are produced by the password database query functions getpwent, getpwuid, and getpwnam.
(getpwent)
(setpwent)
(endpwent)
The first time getpwent function is called, it returns the first password database entry. On subsequent calls it returns successive entries. Entries are returned as instances of the passwd structure. If the function cannot retrieve an entry for any reason, it returns nil.
The setpwent function rewinds the database scan.
The endpwent function releases the resources associated with the scan.
(getpwuid uid)
The getpwuid searches the password database for an entry whose user ID field is equal to the numeric uid. If the search is successful, then a passwd structure representing the database entry is returned. If the search fails, nil is returned.
(getpwnam name)
The getpwnam searches the password database for an entry whose user name is equal to name. If the search is successful, then a passwd structure representing the database entry is returned. If the search fails, nil is returned.
(defstruct group nil
name passwd gid mem)
The group structure corresponds to the C type struct group. Objects of this struct are produced by the password database query functions getgrent, getgrgid, and getgrnam.
(getgrent)
(setgrent)
(endgrent)
The first time getgrent function is called, it returns the first group database entry. On subsequent calls it returns successive entries. Entries are returned as instances of the passwd structure. If the function cannot retrieve an entry for any reason, it returns nil.
The setgrent function rewinds the database scan.
The endgrent function releases the resources associated with the scan.
(getgrgid gid)
The getgrgid searches the group database for an entry whose group ID field is equal to the numeric gid. If the search is successful, then a group structure representing the database entry is returned. If the search fails, nil is returned.
(getgrnam name)
The getgrnam searches the group database for an entry whose group name is equal to name. If the search is successful, then a group structure representing the database entry is returned. If the search fails, nil is returned.
(crypt key salt)
The crypt function is a wrapper for the Unix C library function of the same name. It calculates a hash over the key and salt arguments, which are strings. The hash is returned as a string.
The key and salt arguments are converted into UTF-8 prior to being passed into the underlying platform function. The hash value is assumed to be UTF-8 and converted to Unicode characters, though it is not expected to contain anything but 7 bit ASCII characters.
Note: if C library function crypt uses a static buffer for its return value. If that function is used, the Lisp string returned by the TXR Lisp function carries its own copy of that buffer. Where available, the crypt_r function is used which avoids static storage.
Implementations of the C function vary in their error reporting. Some implementations return a null pointer for invalid salts, whereas others return valid "error token" strings which vary between implementations. To work consistently across numerous implementations, the TXR Lisp crypt function throws an error exception if the C library function returns either a null pointer, or a valid pointer to a string that is less than 13 characters long, regardless of its content.
On platforms where certain advanced features of POSIX signal handling are available at the C API level, TXR exposes signal-handling functionality.
A TXR program can install a TXR Lisp function (such as an anonymous lambda, or the function object associated with a named function) as the handler for a signal.
When that signal is delivered, TXR will intercept it with its own safe, internal handler, mark the signal as deferred (in a TXR sense) and then dispatch the registered function at a convenient time.
Handlers currently are not permitted to interrupt the execution of most TXR internal code. Immediate, asynchronous execution of handlers is currently enabled only while TXR is blocked on I/O operations or sleeping. Additionally, the sig-check function can be used to dispatch and clear deferred signals. These handlers are then safely called if they were subroutines of sig-check, and not asynchronous interrupts.
These variables correspond to the C signal constants SIGHUP, SIGINT and so forth. The variables sig-winch, sig-iot, sig-stkflt, sig-io, sig-lost and sig-pwr may not be available since a system may lack the corresponding signal constants. See notes for the function log-authpriv.
The highest signal number is 31.
(set-sig-handler signal-number handling-spec)
(get-sig-handler signal-number)
The set-sig-handler function is used to specify the handling for a signal, such as the installation of a handler function. It updates the signal handling for a signal whose number is signal-number (usually one of the constants like sig-hup, sig-int and so forth), and returns the previous value. The get-sig-handler function returns the current value.
The signal-number must be an integer the range 1 to 31.
Initially, all 31 signal handling specifications are set to the value t.
The handling-spec parameter may be a function. If a function is specified, then the signal is enabled and connected to that function until another call to set-sig-handler changes the handling for that signal.
If handling-spec is the symbol nil, then the function previously associated with the signal, if any, is removed, and the signal is disabled. For a signal to be disabled means that the signal is set to the SIG_IGN disposition (refer to the C API).
If handling-spec is the symbol t, then the function previously associated with the signal, if any, is removed, and the signal is set to its default disposition. This means that it is set to SIG_DFL (refer to the C API). Some signals terminate the process if they are generated while the handling is configured to the default disposition.
Note that the certain signals like sig-quit and sig-kill cannot be ignored or handled. Please observe the signal documentation in the IEEE POSIX standard, and your platform.
A signal handling function must take two arguments. It is of the form:
(lambda (signal async-p) ...)
The signal argument is an integer indicating the signal number for which the handler is being invoked. The asyncp-p argument is a Boolean value. If it is t, it indicates that the handler is being invoked asynchronously—directly in a signal handling context. If it is nil, then it is a deferred call. Handlers may do more things in a deferred call, such as terminate by throwing exceptions, and perform I/O.
The return value of a handler is normally ignored. However if it invoked asynchronously (the async-p argument is true), then if the handler returns a non-nil value, it is understood that the handler requesting that it be deferred. This means that the signal will be marked as deferred, and the handler will be called again at some later time in a deferred context, whereby async-p is nil. This is not guaranteed, however; it's possible that another signal will arrive before that happens, possibly resulting in another async call, so the handler must be prepared to deal with an async call at any time.
If a handler is invoked synchronously, then its return value is ignored.
In the current implementation, signals do not queue. If a signal is delivered to the process again, while it is marked as deferred, it simply stays deferred; there is no counter associated with a signal, only a Boolean flag.
(sig-check)
The sig-check function tests whether any signals are deferred, and for each deferred signal in turn, it executes the corresponding handler. For a signal to be deferred means that the signal was caught by an internal handler in TXR and the event was recorded by a flag. If a handler function is removed while a signal is deferred, the deferred flag is cleared for that signal.
Calls to the sig-check function may be inserted into CPU-intensive code that has no opportunity to be interrupted by signals, because it doesn't invoke any I/O functions.
(raise signal)
The raise function sends signal to the process. It is a wrapper for the C function of the same name.
The return value is t if the function succeeds, otherwise nil.
(kill process-id [signal])
The kill function is used for sending a signal to a process group or process. It is a wrapper for the POSIX kill function.
If the signal argument is omitted, it defaults to the same value as sig-term.
The return value is t if the function succeeds, otherwise nil.
(strsignal signal)
The strsignal function returns a character string describing the specified signal number. It is based on the same-named POSIX C library function.
(fork)
(wait [pid [flags]])
The fork and wait functions are interfaces to the Unix functions fork and waitpid.
The fork function creates a child process which is a replica of the parent. Both processes return from the function. In the child process, the return value is zero. In the parent, it is an integer representing the process ID of the child. If the function fails to create a child, it returns nil rather than an integer. In this case, the errno function can be used to inquire about the cause.
The wait function, if successful, returns a cons cell consisting of a pair of integers. The car of the cons is the process ID of the process or group which was successfully waited on, and the cdr is the status. If wait fails, it returns nil. The errno function can be used to inquire about the cause.
The process-id argument, if not supplied, defaults to -1, which means that wait waits for any process, rather than a specific process. Certain other values have special meaning, as documented in the POSIX standard for the waitpid function.
The flags argument defaults to zero. If it is specified as nonzero, it should be a bitwise combination (via the logior function) of the variables w-nohang, w-untraced and w-continued. If w-nohang is used, then wait returns a cons cell whose car specifies a process ID value of zero in the situation that at least one of the processes designated by process-id exist and are children of the calling process, but have not changed state. In this case, the status value in the cdr is unspecified.
Status values may be inspected with the functions w-ifexited, w-exitstatus, w-ifsignaled, w-termsig, w-coredump, w-ifstopped, w-stopsig and w-ifcontinued.
(w-ifexited status)
(w-exitstatus status)
(w-ifsignaled status)
(w-termsig status)
(w-coredump status)
(w-ifstopped status)
(w-stopsig status)
(w-ifcontinued status)
These functions analyze process exit values produced by the wait function.
They are closely based on the POSIX macros WIFEXITED, WEXITSTATUS, and so on.
The status value is either an integer, or a cons cell. In this case, the cons cell is expected to have an integer in its cdr which is used as the status.
The w-ifexited, w-ifsignaled, w-coredump, w-ifstopped and w-ifcontinued functions have Lisp Boolean return semantics, unlike their C language counterparts: they return t or nil, rather than zero or nonzero. The others return integer values.
(exec file [args])
The exec function replaces the process image with the executable specified by string argument file. The executable is found by searching the system path.
The file argument becomes the first argument of the executable, argument zero.
If args is specified, it is a list of strings. These are passed as the additional arguments of the executable.
If exec fails, an exception of type file-error is thrown.
(exit* status)
The exit* function terminates the entire process (running TXR image), specifying the termination status to the operating system. The status argument is treated exactly like that of the exit function. Unlike that function, this one exits the process immediately, cleaning up only low-level operating system resources such as closing file descriptors and releasing memory mappings, without performing userspace cleanup.
exit* is implemented using a call to the POSIX function _exit.
(getpid)
(getppid)
These functions retrieve the current process ID and the parent process ID respectively. They are wrappers for the POSIX functions getpid and getppid.
(daemon nochdir noclose)
This is a wrapper for the function daemon which originated in BSD Unix.
It returns t if successful, nil otherwise, and the errno variable is set in that case.
Unlike in the underlying same-named platform function, the nochdir and noclose arguments are Boolean, rather than integer values.
(open-fileno file-descriptor [mode-string [pid]])
The open-fileno function creates and returns a TXR stream over a file descriptor. The file-descriptor argument must be an integer denoting a valid file descriptor.
For a description of mode-string, see the open-file function.
If the pid argument is present, it must be a positive integer corresponding to a process ID. The open-fileno function will associate the process ID with the returned stream. When the stream is closed with close-stream, special handling takes place, as documented for that function.
(fileno stream)
The fileno function returns the underlying file descriptor of stream, if it has one. Otherwise, it returns nil.
This is equivalent to querying the stream using stream-get-prop for the :fd property.
(dupfd old-fileno [new-fileno])
The dupfd function provides an interface to the POSIX functions dup or dup2, when called with one or two arguments, respectively.
(pipe)
The pipe function, if successful, returns a pair of integer file descriptors as a cons-cell pair. The descriptor in the car field of the pair is the read end of the pipe. The cdr holds the write end.
If the function fails, it throws an exception of type file-error.
(close fileno [throw-on-error-p])
The close function passes the integer descriptor fileno to the POSIX close function. If the operation is successful, then t is returned. Otherwise an exception of type file-error is thrown, unless the throw-on-error-p argument is present, with a true value. In that case, close indicates failure by returning nil.
(poll poll-list [timeout])
The poll function suspends execution while monitoring one or more file descriptors for specified events. It is a wrapper for the same-named POSIX function.
The poll-list argument is a sequence of cons pairs. The car of each pair is either an integer file descriptor, or else a stream object which has a file descriptor (the fileno function can be applied to that stream to retrieve a descriptor). The cdr of each pair is an integer bit mask specifying the events, whose occurrence the file descriptor is to be monitored for. The variables poll-in, poll-out, poll-err and several others are available which hold bitmask values corresponding to the constants POLLIN, POLLOUT, POLLERR used with the C language poll function.
The timeout argument, if absent, defaults to the value -1, which specifies an indefinite wait. A nonnegative value specifies a wait with a timeout, measured in milliseconds.
The function returns a list of pairs representing the descriptors or streams which were successfully polled. If the function times out, it returns an empty list. If an error occurs, an exception is thrown.
The returned list is similar in structure to the input list. However, it holds only entries which polled positive. The cdr of every pair now holds a bitmask of the events which were to have occurred.
(isatty stream)
(isatty fileno)
The isatty function provides access to the underlying POSIX function of the same name.
If the argument is a stream object which has a :fd property, then the file descriptor number is retrieved. The behavior is then as if that descriptor number were passed as the fileno argument.
If the argument is not a stream, it must be a fileno: an integer in the representation range of the C type int.
The POSIX isatty is invoked on this integer. If it that returns 1, then t is returned, otherwise nil.
These variables correspond to the POSIX file mode constants O_ACCMODE, O_RDONLY, O_WRONLY, O_RDWR, O_CREAT, O_NOCTTY, and so forth.
The availability of the variables o-async, o-directory, o-nofollow, o-cloexec, o-direct, o-noatime and o-path depends on the host platform.
Some of these flags may be set or cleared on an existing file descriptor using the f-setfl command of the fcntl function, in accordance with POSIX and the host platform documentation.
These variables correspond to the ISO C constants SEEK_SET, SEEK_CUR and SEEK_END. These values, usually associated with the ISO C fseek function, are also used in the fcntl file locking interface as values of the whence member of the flock structure.
These variables correspond to the POSIX fcntl command constants F_DUPFD, F_GETFD, F_SETFD, and so forth. Availability of the f-dupfd-cloexec depends on the host platform.
The fd-cloexec variable corresponds to the POSIX FD_CLOEXEC constant. It denotes the flag which may be set by the fd-setfd command of the fcntl function.
These variables correspond to the POSIX lock type constants F_RDLCK, F_WRLCK and F_UNLCK. They specify the possible values of the type field of the flock structure.
(defstruct flock nil
type whence
start len
pid)
The flock structure corresponds to the POSIX structure of the same name. An instance of this structure must be specified as the third argument of the fcntl function when the command argument is one of the values f-getlk, f-setlk or f-setlkw.
All slots must be initialized with appropriate values before calling fcntl with the exception that the f-getlk command does not access the existing value of the pid slot.
(fcntl fileno command [arg])
The fcntl function corresponds to the same-named POSIX function. The fileno and command arguments must be integers. The TXR Lisp fileno restricts the command argument to the supported values for which symbolic variable names are provided. Other integer command values are rejected by returning -1 and setting the errno variable to EINVAL. Whether the third argument is required, and what type it must be, depends on the command value. Commands not requiring the third argument ignore it if it is passed.
fcntl commands for which POSIX requires an argument of type long require arg to be an integer.
The file locking commands f-getlk, f-setlk and f-setlkw require arg to be a flock structure.
The fcntl function doesn't throw an error if the underlying POSIX function indicates failure; the underlying function's return value is converted to a Lisp integer and returned.
Itimers ("interval timers") can be used in combination with signal handling to execute asynchronous actions. Itimers deliver delayed, one-time signals, and also periodically recurring signals. For more information, consult the POSIX specification.
These variables correspond to the POSIX constants ITIMER_REAL, ITIMER_VIRTUAL and ITIMER_PROF. Their values are suitable as the timer argument of the getitimer and setitimer functions.
(getitimer timer)
(setitimer timer interval value)
The getitimer function returns the current value of the specified timer, which must be itimer-real, itimer-virtual or itimer-prof.
The current value consists of a list of two integer values, which represents microseconds. The first value is the timer interval, and the second value is the timer's current value.
Like getitimer, the setitimer function also retrieves the specified timer. In addition, it stores a new value in the timer, which is given by the two arguments, expressed in microseconds.
On platforms where a Unix-like syslog API is available, TXR exports this interface. TXR programs can configure logging via the openlog function, control the logging mask via setlogmask and generate logs via syslog, or using special syslog streams.
These variables take on the values of the corresponding C preprocessor constants from the <syslog.h> header: LOG_PID, LOG_CONS, etc. These integer values represent logging options used in the options argument to the openlog function.
Note: LOG_PERROR is not in POSIX, and so log-perror might not be available. See notes about LOG_AUTHPRIV in the documentation for log-authpriv.
These variables take on the values of the corresponding C preprocessor constants from the <syslog.h> header: LOG_USER, LOG_DAEMON, LOG_AUTH and LOG_AUTHPRIV. These are the integer facility codes specified in the openlog function.
Note: LOG_AUTHPRIV is not in POSIX, and so log-authpriv might not be available. For portability use code like (or (symbol-value 'log-authpriv) 0) to evaluate to 0 if log-authpriv doesn't exist, or else check for its existence using (boundp 'log-authpriv).
These variables take on the values of the corresponding C preprocessor constants from the <syslog.h> header: LOG_EMERG, LOG_ALERT, etc. These are the integer priority codes specified in the syslog function.
The *stdlog* variable holds a special kind of stream: a syslog stream. Each newline-terminated line of text sent to this stream becomes a log message.
The stream internally maintains a priority value that is applied when it generates messages. By default, this value is that of log-info. The stream holds the priority as the value of the :prio stream property, which may be changed with the stream-set-prop function.
The latest priority value which has been configured on the stream is used at the time the newline character is processed and the log message is generated, not necessarily the value which was in effect at the time the accumulation of a line began to take place.
Messages sent to *stdlog* are delimited by newline characters. That is to say, each line of text written to the stream is a new log.
(openlog id-string [options [facility]])
The openlog function is a wrapper for the openlog C function, and the arguments have the same semantics. It is not necessary to use openlog in order to call the syslog function or to write data to *stdlog*. The call is necessary in order to override the default identifying string, to set options, such as having the PID (process ID) recorded in log messages, and to specify the facility.
The id-string argument is mandatory.
The options argument is a bitwise mask (see the logior function) of option values such as log-pid and log-cons. If it is missing, then a value of 0 is used, specifying the absence of any options.
The facility argument is one of the values log-user, log-daemon or log-auth. If it is missing, then log-user is assumed.
(setlogmask bitmask-integer)
The setlogmask function interfaces to the corresponding C function, and has the same argument and return value semantics. The bitmask-integer argument is a mask of priority values to enable. The return value is the prior value. Note that if the argument is zero, then the function doesn't set the mask to zero; it only returns the current value of the mask.
Note that the priority values like log-emerg and log-debug are integer enumerations, not bitmasks. These values cannot be combined directly to create a bitmask. Rather, the mask function should be used on these values.
;; Enable LOG_EMERG and LOG_ALERT messages,
;; suppressing all others
(setlogmask (mask log-emerg log-alert))
(syslog priority format format-arg*)
The syslog function is the interface to the syslog C function. The printf formatting capabilities of the function are not used; the format argument follows the conventions of the TXR Lisp format function instead. Note in particular that the %m convention for interpolating the value of strerror(errno) which is available in some versions of the syslog C function is currently not supported.
Note that syslog messages are not newline-terminated.
On platforms where the POSIX glob function is available TXR provides this functionality in the form of a like-named function, and some numeric constants. TXR also provides access the fnmatch function, where available.
These variables take on the values of the corresponding C preprocessor constants from the <glob.h> header: GLOB_ERR, GLOB_MARK, GLOB_NOSORT, etc.
These values are passed as an argument to the optional flags argument of the glob function. They are bitmasks and so multiple values can be combined using the logior function.
Note that the glob-period, glob-altdirfunc, glob-brace, glob-nomagic, glob-tilde, glob-tilde-check and glob-onlydir variables may not be available. They are extensions in the GNU C library implementation of glob.
The standard GLOB_APPEND flag is not represented as a TXR variable. The glob function uses it internally when calling the C library function multiple times, due to having been given multiple patterns.
This value holds an integer bitmask value that may be given as an argument to the optional flags parameter of the glob* function. It may be used alone, combine with the other glob mask values using the logior function.
If used with glob*, it disables brace expansion, which is enabled in glob* by default.
If used with the glob function, it has no effect.
This value is a TXR Lisp extension; it does not appear in the API of the glob C function.
(glob {pattern | patterns} [flags [errfun]])
(glob* {pattern | patterns} [flags [errfun]])
The glob function is a interface to the Unix function of the same name. The first argument must either be a single pattern, which is a string, or else sequence of strings specifying multiple patterns, which are strings. Each string is a glob pattern: a pattern which matches zero or more pathnames, similar to a regular expression. The function tries to expand the patterns and return a list of strings representing the matching pathnames in the file system.
If there are no matches, then an empty list is returned.
The glob* function is a TXR Lisp extension built on glob.
The glob* functions supports a ** ("double star") pattern which matches zero or more path components. The double star match is described in detail below.
The glob* function also supports brace expansion, independently of whether or not glob supports brace expansion. Brace expansion is enabled by default in glob* and can be disabled using the glob-xnobrace flag. Brace expansion is described in detail below.
Lastly, the glob* function performs a path-aware sort of the emerging path names that is not influenced by locale, whereas the sort performed by glob is influenced by locale, defaulting to a lexicographic sort in the "C" locale.
The optional flags argument defaults to zero. If given, it may be a bitwise combination of the values of the variables glob-err, glob-mark, glob-nosort and others. The glob-append
If the errfun argument is specified, it gives a callback function which is invoked when glob encounters errors accessing paths. The function takes two arguments: the pathname and the errno value which occurred for that pathname. The function's return value is Boolean. If the function returns true, then glob will terminate.
The errfun may terminate the traversal by a nonlocal exit, such as by throwing an exception or performing a block return.
The errfun may not reenter the glob function. This situation is detected and diagnosed by an exception.
The errfun may not capture a continuation across the error boundary. That is to say, code invoked from the error may not capture a continuation up to a prompt which surrounds the glob call. Such an attempt is detected and diagnosed by an exception.
If a sequence of patterns is specified instead of a single pattern, glob makes multiple calls to the underlying C library function. The second and subsequent calls specify the GLOB_APPEND flag to add the matches to the result. The following equivalence applies:
(glob (list p0 p1 ...) f e) <--> (append (glob p0 f e)
(glob p1 f e)
...)
Details of the semantics of the glob function, and the meaning of all the flags arguments are given in the documentation for the C function.
The glob* function supports brace expansion, which is enabled by default, and can be disabled with glob-xnobrace.
On some platforms, such as the GNU C Library, the glob function also supports brace expansion. If available, then the glob-brace variable has a nonzero value and must be included in the flags argument.
These two brace expansion features are independent; the TXR Lisp glob* function does not rely on glob for brace expansion, even if it is available.
The brace expansion supported by glob* is a string generation mechanism driven by a syntax which specifies comma-separated elements enclosed in braces.
When a single brace expansion appears in a pattern, that pattern turns into a list of patterns. There are as many elements in the list as there are elements between the braces. Each element replaces the braces with a different element from between the braces.
For instance, "x{a,b}y" denotes the list of strings ("xay" "xby"). The there are two elements in the list because the braces contain two elements. The first string replaces "{a,b}" with "a" and the second replaces it with "b".
When multiple braces occur in a pattern, then all combinations (Cartesian product) of the braces is produced.
Braces may also nest. When the element of a brace itself uses braces, then that element is subject to brace expansion. The elements which emerge then become items of the enclosing brace, as if they were comma-separated elements. For instance "x{a,{b,c}y}z" is equivalent to "x{a,by,cy}z" which then expands to the three strings "xaz", "xbyz" and "xcyz".
Braces may be escaped by a backslash to disable their special meaning. Likewise, the commas may be escaped by a backslash to preserve their special meaning. Brace expansion preserves these backslashes; they appear in the resulting patterns, and must be recognized and removed by subsequent processing.
When the pattern arguments of glob* use brace expansion, those arguments produce multiple patterns. The order of these patterns is preserved: the patterns are matched in that order. For each pattern, the matching path names are sorted, unless the glob-nosort flag is in effect.
The ** ("double star") operator recognized by glob* matches zero or more path components. It may be used more than once. It cannot be combined with other characters or globbing operators.
It is valid for "**" to be an entire pattern. This expands the relative path names of all files, directories and other objects in the current directory and its children.
Otherwise the double star may appear at the start of a pattern if it is followed by a slash; at the end of a pattern if it is preceded by a slash, or in the middle of a pattern if it is surrounded by two slashes. The double star is not recognized in a bracket-enclosed character class.
Thus, the following examples all contain one double star:
**
foo/**
**/bar
here/**/there
These do not contain a double star; the two asterisks in these patterns will be passed to the underlying glob function without being processed as a double star by glob*, with unspecified consequences:
foo**
**bar
here**/there
etc/**conf
foo[/**/]bar
Each double star matches a maximum of ten path components, and all of the double stars in a single pattern together do not match more than 48 components. Using more than three double stars in a pattern is not recommended for performance reasons.
If the double star is followed by a slash, it matches only directories.
The glob* function sorts paths in such a way that the slash character is ranked lower than all other characters. Thus the path "test/" sorts before "test-data/" even though in ASCII and Unicode, the - character has a lower code than the / character.
;; find all jpg and gif paths under the current directory,
;; (up to ten levels deep).
(glob* "**/*.{jpg,gif}")
;; find all "2023" directories under the current directory,
;; which have .jpg or .gif files under them, listing those
;; .jpg and .gif paths:
(glob* "**/2023/**/*.{jpg,gif}")
;; find all "2023" directories under the current directory.
(glob* "**/2023/**/")
These variables take on the values of the corresponding C preprocessor constants from the <fnmatch.h> header: FNM_PATHNAME, FNM_NOESCAPE, FNM_PERIOD, etc.
These values are bit masks which may be combined with the logior function to form the optional third flags argument of the fnmatch function.
Note that the fnm-leading-dir, fnm-casefold and fnm-extmatch functions may not be available. They are GNU extensions, found in the GNU C library.
(fnmatch pattern string [flags]])
The fnmatch function, if available, provides access to the like-named POSIX C library function. The pattern argument specifies a POSIX-shell-style filename-pattern-matching expression. Its exact features and dialect are controlled by flags. If string matches pattern then t is returned. If there is no match, then nil is returned. If the C function indicates that an error has occurred, an exception is thrown.
On platforms where the POSIX nftw function is available TXR provides this functionality in the form of the analogous Lisp function ftw, accompanied by some numeric constants.
Likewise, on platforms where the POSIX functions opendir and readdir are available, TXR provides the functionality in the form of same-named Lisp functions, a structure type named dirent and some accompanying numeric constants.
These variables hold numeric values that may be combined into a single bitmask bitmask value using the logior function. This value is suitable as the flags argument of the ftw function.
These variables correspond to the C constants FTW_PHYS, FTW_MOUNT, etc.
Note that ftw-actionretval is a GNU extension that is not available on all platforms. If the platform's nftw function doesn't have this feature, then this variable is not defined.
These variables provide symbolic names for the integer values that are passed as the type argument of the callback function called by ftw. This argument classifies the kind of file system node visited, or error condition encountered.
These variables correspond to the C constants FTW_F, FTW_D, etc.
Not all of them are present. If the underlying platform doesn't have a given constant, then the corresponding variable doesn't exist in TXR.
These variables are defined if the variable ftw-actionretval is defined.
If the value of ftw-actionretval is included in the flags argument of ftw, then the callback function can use the values of these variables as return codes. Ordinarily, the callback returns zero to continue the search and nonzero to stop.
These variables correspond to the C constants FTW_CONTINUE, FTW_STOP, etc.
(ftw path-or-list callbackfun [flags [nopenfd]])
[callbackfun path type stat-struct level base]
The ftw function provides access to the nftw POSIX C library function.
Note that the flags and nopenfd arguments are reversed with respect to the C language interface. They are both optional; flags defaults to the value of ftw-phys and nopenfd defaults to 20. If an argument is given to flags, then the presence of the ftw-phys is no longer implied; the flag must be explicitly included in the argument in order to be present.
Compatibility Note: the flags parameter defaults to an argument value of zero in TXR versions 283 or lower.
The path-or-list argument may be a string specifying the top-level pathname that ftw shall visit. Or else, path-or-list may be a list. If it is a list, then ftw recursively invokes itself over each of the elements, taking that element as the path-or-name argument of the recursive call, passing down all other argument values as-is. The traversal stops when any recursive invocation of ftw returns a value other than t or nil, and that value is returned. If t or nil is returned, the traversal continues with the application of ftw to the next list element, if any. If the list is completely traversed, and some recursive invocations of ftw return t, then the return value is t. If all recursive invocations return nil then nil is returned. If the list is empty, t is returned.
The ftw function walks the filesystem, as directed by the path-or-list argument and flags bitmask arguments.
For each visited entry, it calls the supplied callbackfun function, which receives five arguments. If this function returns normally, it must return either nil, t, or an integer value in the range of the C type int.
The ftw function can continue the traversal by returning any non-integer value, or the integer value zero. If ftw-actionretval is included in the flags bitmask, then the only integer code which continues the traversal without any special semantics is ftw-continue and only ftw-stop stops the traversal. (Non-integer return values behave like ftw-continue).
The path argument of callbackfun gives the path of the visited filesystem object.
The type argument is an integer code which indicates the kind of object that is visited, or an error situation in visiting that filesystem entry. See the documentation for ftw-f and ftw-d for possible values.
The stat-struct argument provides information about the filesystem object as a stat structure, the same kind of object as what is returned by the stat function.
The level argument is an integer value representing the directory level depth. This value is obtained from the C structure FTW in the nftw C API.
The base argument indicates the length of the directory part of the path argument. Characters in excess of this length are thus the base name of the visited object, and the expression [path base..:] calculates the base name.
The ftw function returns either t upon successful completion, or an integer value returned by callbackfun, as described below. On failure it throws an exception derived from file-error, whose specific type is based on analyzing the POSIX errno value.
The callbackfun may return a value of any type. If it returns a value that is not of integer type, then zero is returned to the nftw function and traversal continues. Similarly, traversal continues if the function returns an integer zero.
If callbackfun returns an integer value, that value must be in the range of the C type int. That int value is returned to nftw. If the value is not zero, and is not -1, then nftw will terminate, and return that value, which ftw then returns. If the value is -1, then nftw is deemed to have failed, and ftw will thrown an exception of type file-error, whose specific type is based on analyzing the POSIX errno value. If the value is zero, then the traversal continues.
The callbackfun may also terminate the traversal by a nonlocal exit, such as by throwing an exception or performing a block return.
The callbackfun may not reenter the ftw function. This situation is detected and diagnosed by an exception.
The callbackfun may not capture a continuation across the callback boundary. That is to say, code invoked from the callback may not capture a continuation up to a prompt which surrounds the ftw call. Such an attempt is detected and diagnosed by an exception.
(defstruct dirent nil
name ino type)
Objects of the dirent structure type are returned by the readdir function.
The name slot is a character string giving the name of the directory entry. If the opendir function's prefix-p argument is specified as true, then readdir operations produce dirent structures whose name slot is a path formed by combining the directory path with the directory entry name.
The ino slot is an integer giving the inode number of the object named by the directory entry.
The type slot indicates the type of the object, which is an integer code. Support for this member is platform-dependent. If the directory traversal doesn't provide the information, then this slot takes on the nil value. In this situation, the dirstat function may be used to backfill the missing information.
These variables give the possible type code values exhibited by the type slot of the dirent structure.
If the underlying host platform does not feature a d_type field in the dirent C structure, then almost all these variables are defined anyway using the values that they have on GNU/Linux. These definitions are useful in conjunction with the dirstat function below.
If the host platform does does not feature a d_type field in the dirent structure, then the variable dt-unknown is not defined. Note: the application can take advantage of this this to detect the situation, in order to conditionally define code in such a way that some run-time checking is avoided.
(opendir dir-path [prefix-p])
The opendir function initiates a traversal of the directory object named by the string argument dir-path, which must be the name of a directory. If opendir is not able to open the directory traversal, it throws an exception of type system-error. Otherwise an object of type dir is returned, which is a directory traversal handle suitable as an argument for the readdir function.
If the prefix-p argument is specified and has a true value, then it indicates that the subsequent readdir operations should produce the value of the name slot of the dirent structure by combining dir-path with the directory entry name using the path-cat function.
(readdir dir-handle [dirent-struct])
The readdir function returns the next available directory entry from the directory traversal controlled by dir-handle, which must be a dir object returned by opendir.
If no more directory entries remain, then readdir returns nil. In this situation, the dir-handle is also closed, as if by a call to closedir.
Otherwise, the next available directory entry is returned as a structure object of type dirent.
The readdir function internally skips and does not report the "." (dot) and ".." (dotdot) directory entries.
If the dirent-struct argument is specified, then it must be a dirent structure, or one which has all of the required slots. In this case, readdir stores values in that structure and returns it. If dirent-struct is absent, then readdir allocates a fresh dirent structure.
(opendir dir-handle)
The closedir function terminates the directory traversal managed by dir-handle, releasing its resources.
If this has already been done before, closedir returns nil, otherwise it returns t.
Further readdir calls on the same dir-handle return nil.
Note: the readdir function implicitly closes dir-handle when the handle indicates that no more directory entries remain to be traversed.
(dirstat dirent-struct [dir-path [struct]])
The dirstat function invokes lstat on the object represented by the dirent structure dirent-struct, sets the type slot of the dirent-struct accordingly, and then returns the value that lstat returned.
If the struct argument is specified, it is passed to lstat.
The dir-path parameter must be specified, if the name slot of dirent-struct is a simple directory entry name, rather than the full path to the object. In that case, the slot's value gives the effective path. If the name slot is already a path (due to, for instance, a true value of prefix-p having been passed to opendir) then dir-path must not be specified. If dir-path is specified, then its value is combined with the name slot of dirent-struct using path-cat to form the effective path.
The lstat function is invoked on the effective path, and if it succeeds, then type information is obtained from the resulting structure to set the value of the type slot of dirent-struct. The same structure that was returned by lstat is then returned.
On platforms where the underlying system interface is available, TXR provides a sockets library for communicating over Internet networks, or over Unix sockets.
Stream as well as datagram sockets are supported.
The classic Version 4 of the Internet protocol is supported, as well as IP Version 6.
Sockets are mapped to TXR streams. The open-socket function creates a socket of a specified type, in a particular address family. This socket is actually a stream (always, even if it isn't used for data transfer, but only as a passive contact point).
The functions sock-connect, sock-bind, sock-listen, sock-accept and sock-shutdown are used for enacting socket communication scenarios.
Stream sockets use ordinary streams, reusing the same underlying framework that is used for file I/O and process types.
Datagram socket streams are implemented using special datagram socket streams. Datagram socket streams eliminate the need for operations analogous to the sendto and recvfrom socket API functions, even in server programs which handle multiple clients. An overview of datagrams is treated in the following section, Datagram Socket Streams.
The getaddrinfo function is provided for resolving host names and services to IPv4 and IPv6 addresses.
Several structure types are provided for representing socket addresses, and options for getaddrinfo.
Various numeric constants are also provided: af-unix, af-inet, af-inet6, sock-stream, sock-dgram and others.
Datagram socket streams are a new paradigm unique to TXR which attempts to unify the programming model of stream and datagram sockets.
A datagram socket stream is created by the open-socket function, when the sock-dgram socket type is specified. Another way in which a datagram socket is created is when sock-accept is invoked on a datagram socket, and returns a new socket.
I/O is performed on datagram sockets using the regular I/O functions. None of the functions take or return peer addresses. There are no I/O functions which are analogous to the C library recvfrom and sendto functions which are usually used for datagram programming. Datagram I/O assumes that the datagram datagram socket is connected to a specific remote peer, and that peer is implicitly used for all I/O.
Datagram streams solve the message framing problem by considering a single datagram to be an entire stream. On input, a datagram stream holds an entire datagram in a buffer. The stream ends (experiences the EOF condition) after the last byte of this buffer is removed by an input operation. Another datagram will be received and buffered if the EOF condition is first explicitly cleared with the clear-error function, and then another input operation is attempted. On output, a datagram stream gathers data into an ever-growing output buffer which isn't subject to any automatic flushing. An explicit flush-stream operation sends the buffer contents to the connected peer as a new datagram, and empties the buffer. Subsequent output operations prepare data for a new datagram. The close-stream function implicitly flushes the stream in the same way, and thus also potentially generates a datagram.
A client-style datagram stream can be explicitly connected to a peer with the sock-connect function. This is equivalent to connecting a datagram socket using the C library connect function. Writes on the stream will be transmitted using the C library function send. A client-style datagram stream can also be "soft-connected" to a peer using the sock-set-peer function. Writes on the stream will transmit data using the C library function sendto to the peer address.
A datagram server program which needs to communicate using multiple peers is implemented by means of the sock-accept function which, unlike the C library accept function, works with datagram sockets as well as stream sockets. The server creates a datagram socket, and uses sock-bind to bind it to a local address. Optionally, it may also call sock-listen which is a no-op on datagram sockets. Supporting this function on datagram sockets allows program code to be more easily changed between datagram and stream operation. The server then uses sock-accept to accept new clients. Note that this is not possible with the C library function accept, which only works with stream sockets.
The sock-accept function receives a datagram from a client, and creates a new datagram socket stream which is connected to that client, and whose input buffer contains the received datagram. Input operations on this stream consume the datagram. Note that clearing the EOF condition and trying to receive another datagram is not recommended on datagram streams returned by sock-accept, since they share the same underlying operating system socket, which is not actually connected to a specific peer. The receive operation could receive a datagram from any peer, without any indication which peer that is. Datagram servers should issue a new sock-accept call for each client datagram, treating it as a new stream.
Datagram sockets ignore almost all aspects of the mode-string passed in open-socket and sock-accept. The only attribute not ignored is the buffer size specified with a decimal digit character; however, it cannot be the only item in the mode string. The string must be syntactically valid, as described under the open-file function. The buffer size attribute controls the size used by the datagram socket for receiving a datagram: the capture size. A datagram socket has obtains a default capture size if one isn't specified by the mode-string. The default capture size is 65536 bytes for a datagram socket created by open-socket. If a size is not passed to sock-accept via its mode-string argument when it is invoked on a datagram socket, that socket's size is used as the capture size of the newly created datagram socket which is returned.
(defstruct sockaddr nil
canonname
(:static family nil))
The sockaddr structure represents the abstract base class for socket addresses, from which several other types are derived: sockaddr-in, sockaddr-in6 and sockaddr-un.
It has a single static slot named family and a single instance slot canonname, both initialized to nil.
Note: the canonname slot is optionally set by the getaddrinfo function on address structures that it returns, if requested via the ai-canonname flag. The slot only provides information to the application, playing no semantic role in addressing.
(defstruct sockaddr-in sockaddr
(addr 0) (port 0) (prefix 32)
(:static family af-inet))
The sockaddr-in address represents a socket address used in the context of networking over IP Version 4. It may be used with sockets in the af-inet address family.
The addr slot holds an integer denoting an abstract IPv4 address. For instance the hexadecimal integer literal constant #x7F000001 or its decimal equivalent 2130706433 represents the loopback address, whose familiar "dot notation" is 127.0.0.1. Conversion of the abstract IP address to four bytes in network order, as required, is handled internally.
The port slot holds the TCP or UDP port number, whose value ranges from 0 to 65535. Zero isn't a valid port; the value is used for requesting an ephemeral port number in active connections. Zero also appears in situations when the port number isn't required: for instance, when the getaddrinfo function is used with the aim of looking up the address of a host, without caring about the port number.
The prefix field is set by the function inaddr-str, when it recognizes and parses a prefix field in the textual representation.
The family static slot holds the value af-inet.
(defstruct sockaddr-in6 sockaddr
(addr 0) (port 0) (flow-info 0) (scope-id 0)
(prefix 128)
(:static family af-inet6))
The sockaddr-in6 address represents a socket address used in the context of networking over IP Version 6. It may be used with sockets in the af-inet6 address family.
The addr slot holds an integer denoting an abstract IPv6 address. IPv6 addresses are pure binary integers up to 128 bits wide.
The port slot holds the TCP or UDP port number, whose value ranges from 0 to 65535. In IPv6, the port number functions similarly to IPv6; see sockaddr-in.
The flow-info and scope-id are special IPv6 parameters corresponding to the sin6_flowinfo and sin6_scope_id slots of the sockaddr_in6 C language structure. Their meaning and use are beyond the scope of this document.
The prefix field is set by the function in6addr-str, when it recognizes and parses a prefix field in the textual representation.
The family static slot holds the value af-inet6.
(defstruct sockaddr-un sockaddr
path
(:static family af-unix))
The sockaddr-un address represents a socket address used for interprocess communication within a single operating system node, using the "Unix domain" sockets of the af-unix address family.
This structure has only one slot, path which holds the rendezvous name for connecting pairs of socket endpoints. This name appears in the filesystem.
When the sockaddr-un structure is converted to the C structure struct sockaddr_un, the path slot undergoes conversion to UTF-8. The resulting bytes are stored in the sun_path member of the C structure. If the resulting UTF-8 byte string is larger than the sun_path array, it is silently truncated.
Note: Linux systems have support for "abstract" names which do not appear in the filesystem. These abstract names are distinguished by starting with a null byte. For more information, consult Linux documentation. This convention is supported in the path slot of the sockaddr-un structure. If path contains occurrences of the pseudo-null character U+DC00, these translate to null bytes in the sun_path member of the corresponding C structure struct sockaddr_un. For example, the path "\xDC00;foo" is valid and represents an abstract address consisting of the three bytes "foo" followed by null padding bytes.
The family static slot holds the value af-unix.
(defstruct addrinfo nil
(flags 0) (family 0) (socktype 0))
The addrinfo structure is used in conjunction with the getaddrinfo function. If that function's hints argument is specified, it is of this type. The purpose of the argument is to narrow down or possibly alter the selection of addresses which are returned.
The flags slot holds a bitwise or combination (see the logior function) of getaddrinfo flags: values given by the variables ai-passive, ai-numerichost, ai-v4mapped, ai-canonname, ai-all, ai-addrconfig and ai-numericserv. These correspond to the C constants AI_PASSIVE, AI_NUMERICHOST and so forth.
If ai-canonname is specified, then every returned address structure will have its canonname member set to a string value rather than nil. This string is a copy of the canonical name reported by the underlying C library function, which that function places only into the first returned address structure.
The family slot holds an address family, which may be the value of af-unspec, af-unix, af-inet or af-inet6.
The socktype slot holds, a socket type. Socket types are given by the variables sock-dgram and sock-stream.
(getaddrinfo [node [service [hints]]])
The getaddrinfo returns a list of socket addresses based on search criteria expressed in its arguments. That is to say, the returned list, unless empty, contains objects of type sockaddr-in and sockaddr-in6.
The function is implemented directly in terms of the like-named C library function. All parameters are optional. Omitting any argument causes a null pointer to be passed for the corresponding parameter of the C library function.
The node and service parameters may be character strings which specify a host name, and service. The contents of these strings may be symbolic, like "www.example.com" and "ssh" or numeric, like "10.13.1.5" and "80".
If an argument is given for the hints parameter, it must be of type addrinfo.
The node and service parameters may also be given integer arguments. An integer argument value in either of these parameters is converted to a null pointer when calling the C getaddrinfo function. The integer values are then simply installed into every returned address as the IP address or port number, respectively. However, if both arguments are numeric, then no addresses are returned, since the C library function is then called with a null node and service.
These variables hold integers which give the values of address families. They correspond to the C constants AF_UNIX, AF_INET and AF_INET6. Address family values are used in the hints argument of the getaddrinfo function, and in the open-socket function. Note that unlike the C language socket addressing structures, the TXR socket addresses do not contain an address family slot. That is because they indicate their family via their type. That is to say, an object of type sockaddr-in is an address which is implicitly associated with the af-inet family via its type.
These variables hold integers which give the values of address families. They correspond to the C constants SOCK_STREAM and SOCK_DGRAM.
These variables hold integers which are bitmasks that combine together via bitwise or, to express the flags slot of the addrinfo structure. They correspond to the C constants AI_PASSIVE, AI_NUMERICHOST, AI_V4MAPPED and so forth. They influence the behavior of the getaddrinfo function.
These integer-valued variables provide constants for commonly used IPv4 and IPv6 address values.
The value of inaddr-any and in6addr-any is zero. This address is used in binding a passive socket to all of the external interfaces of a host, so that it can accept connections or datagrams from all attached networks.
The inaddr-loopback variable is IPv4 loopback address, the same integer as the hexadecimal constant #x7F000001.
The in6addr-loopback is the IPv6 loopback address. Its value is 1.
;; Construct an IPv6 socket address suitable for binding
;; a socket to the loopback network, port 1234:
(new sockaddr-in6 addr in6addr-loopback port 1234)
;; Mistake: IPv4 address used with IPv6 sockaddr.
(new sockaddr-in6 addr inaddr-loopback)
(open-socket family type [mode-string])
The open-socket function creates a socket, which is a kind of stream.
The family parameter specifies the address family of the socket. One of the values af-unix, af-inet or af-inet6 should be used to create a Unix domain, Internet IPv4 or Internet IPv6 socket, respectively.
The type parameter specifies the socket type, either sock-stream (stream socket) or sock-dgram (datagram socket).
The mode-string specifies several properties of the stream; for a description of mode-string parameters, refer to the open-file function. Note that the defaulting behavior for an omitted mode-string argument is different under open-socket from other functions. Because sockets are almost always used for bidirectional data flow, the default mode string is "r+b" rather than the usual "r".
The rationale for including the "b" flag in the default mode string is that network protocols are usually defined in a way that is independent of machine and operating system, down to the byte level, even when they are textual. It doesn't make sense for the same TXR program to see a network stream differently based on what platform it is running on. Line-ending conversion has to do with how a platform locally stores text files, whereas network streams are almost always external formats.
Like other stream types, stream sockets are buffered and marked as non-real-time streams. Specifying the "i" mode in mode-string marks a socket as a real-time stream, and, if it is opened for writing or reading and writing, changes it to use line buffering.
(open-socket-pair family type [mode-string])
The open-socket-pair function provides an interface to the functionality of the socketpair C library function.
If successful, it creates and returns a list of two stream objects, which are sockets that are connected together.
Note: the Internet address families af-inet and af-inet6 are not supported.
The mode-string is applied to each stream. For a description, see open-socket and open-file.
(sock-family socket)
(sock-type socket)
These functions retrieve the integer values representing the address family and type of a socket. The argument to the socket parameter must be a socket stream or a file or process stream. For a file stream, both functions return nil. An exception of type type-error is thrown for other stream types.
(sock-peer socket)
(set (sock-peer socket) address)
The sock-peer function retrieves the peer address has most recently been assigned to socket.
Sockets which are not connected initially have a peer address value of nil. A socket which is connected to a remote peer receives that peer's address as its sock-peer.
If a socket is connected to a remote peer via a successful use of the sock-connect function, then its sock-peer address is set to match that of the peer.
Sockets returned by the sock-accept function are connected, and have the remote endpoint address as their sock-peer address.
Assigning an address to a sock-peer form is equivalent to using sock-set-peer to set the address.
Implementation note: the sock-peer function does not use the getpeername C library function; the association between a stream and sockaddr struct is maintained by TXR.
(sock-set-peer socket address)
The sock-set-peer function stores address into socket as that socket's peer.
Subsequently, the sock-peer function will retrieve that address.
If address is not an appropriate address object in the address family of socket, the behavior is unspecified.
(sock-connect socket address [timeout-usec])
The sock-connect function connects a socket stream to a peer address.
The address argument must be a sockaddr object of type matching the address family of the socket.
If the operation fails, an exception of type socket-error is thrown. Otherwise, the function returns socket.
If the timeout-usec argument is specified, it must be a fixnum integer. It denotes a connection timeout period in microseconds. If the connection doesn't succeed within the specified timeout, an exception of type timeout-error is thrown.
(sock-bind socket address)
The sock-bind function binds a socket stream to a local address after enabling the socket stream's so-reuseaddr option.
The address argument must be a sockaddr object of type matching the address family of the socket.
If the operation fails, an exception of type socket-error is thrown. Otherwise, the function returns t.
(sock-listen socket [backlog])
The sock-listen function prepares socket for listening for connections. The backlog parameter, if specified, requires an integer argument. The default value is 16.
(sock-accept socket [mode-string [timeout-usec]])
The sock-accept function waits for a client connection on socket, which must have been prepared for listening for connections using sock-bind and sock-listen.
If the operation fails, an exception of type socket-error is thrown. Otherwise, the function returns a new socket which is connected to the remote peer.
The peer's address may be retrieved from this socket using sock-peer.
The mode-string parameter is applied to the new socket just like the similar argument in open-socket. It defaults to "r+b".
If the timeout-usec argument is specified, it must be a fixnum integer. It denotes a timeout period in microseconds. If no peer connects for the specified timeout, sock-accept throws an exception of type timeout-error.
The values of these variables are useful as the second argument to the sock-shutdown function.
(sock-shutdown sock [direction])
The sock-shutdown function indicates that no further communication is to take place on socket in the specified direction(s).
If the operation fails, an exception of type socket-error is thrown. Otherwise, the function returns t.
The direction parameter is one of the values given by the variables shut-rd, shut-wr or shut-rdwr. These values shut down communication in the read direction, write direction, or both directions, respectively.
If the argument is omitted, sock-shutdown defaults to closing the write direction.
Notes: shutting down is most commonly requested in the write direction, to perform a "half close". The communicating process thereby indicates that it has written all the data which it intends to write. When the shutdown action is processed on the remote end, that end is unblocked from waiting on any further data, and effectively experiences an "end of stream" condition on its own socket or socket-like endpoint, while continuing to be able to transmit data. Shutting down in the reading direction is potentially abrupt. If it is executed before an "end of stream" indication is received from a peer, it results in an abortive close.
(sock-recv-timeout sock usec)
(sock-send-timeout sock usec)
The sock-recv-timeout and sock-send-timeout functions configure, respectively, receive and send timeouts on socket sock.
The usec parameter specifies the value, in microseconds. It must be a fixnum integer.
When a receive timeout is configured on a socket, then an exception of type timeout-error is thrown when an input operation waits for at least usec microseconds without receiving input.
Similarly, when a send timeout is configured, then an exception of type timeout-error is thrown when an output operation waits for at least usec microseconds for the availability of buffer space in the socket.
These variables represent the protocol levels of socket options and are suitable for use as the level argument of the sock-opt and sock-set-opt functions. The variables correspond to the POSIX C constants SOL_SOCKET, IPPROTO_IP, IPPROTO_IPV6, IPPROTO_TCP and IPPROTO_UDP.
These variables represent socket options at the sol-socket protocol level and are suitable for use as the option argument of the sock-opt and sock-set-opt functions. The variables correspond to the POSIX C constants SO_ACCEPTCONN, SO_BROADCAST, SO_DEBUG, etc.
Note that the sock-recv-timeout and sock-send-timeout are a more convenient interface for setting the value of the so-rcvtimeo and so-sndtimeo socket options.
These variables represent socket options at the ipproto-ipv6 protocol level and are suitable for use as the option argument of the sock-opt and sock-set-opt functions. The variables correspond to the POSIX C constants IPV6_JOIN_GROUP, IPV6_LEAVE_GROUP, IPV6_MULTICAST_HOPS, etc.
This variable represents a socket option at the ipproto-tcp protocol level and is suitable for use as the option argument of the sock-opt and sock-set-opt functions. The variable corresponds to the POSIX C constant TCP_NODELAY.
(sock-opt socket level option [ffi-type])
(set (sock-opt socket level option [ffi-type]) value)
The sock-opt function retrieves the value of the specified socket option, at the specified protocol level, associated with socket, which must be a socket stream.
The level argument should be one of the protocol levels sol-socket, ipproto-ip, ipproto-ipv6, ipproto-tcp and ipproto-udp.
The option argument should be one of the socket options so-acceptconn, so-broadcast, so-debug, ..., ipv6-join-group, ..., ipv6-v6only and tcp-nodelay.
The ffi-type argument, which must be a compiled FFI type, specifies the type of the socket option's value. The type is most commonly int or uint, but it can be any other fixed-size type, including structs. (Variable-size types, such as C char arrays, are unsupported.) The ffi-type argument defaults to (ffi int).
Assigning a value to a sock-opt place is equivalent to calling sock-set-opt with that value.
Note: the sock-opt and sock-set-opt functions call the POSIX C getsockopt and setsockopt functions, respectively. Consult the POSIX specification for more information about these functions and in particular the various socket options (and the types they require).
(sock-set-opt socket level option value [ffi-type])
The sock-set-opt function sets the value of the specified socket option, at the specified protocol level, associated with socket, which must be a socket stream.
See the documentation of the sock-opt function for a description of the level, option and ffi-type arguments. Like the sock-opt function, sock-set-opt's ffi-type argument defaults to (ffi int).
(str-inaddr address [port])
(str-in6addr address [port])
The str-inaddr and str-in6addr functions convert an IPv4 and IPv6 address, respectively, to textual notation which is returned as a character string. The conversion is done in conformance with RFC 5952, section 4.
IPv6 addresses representing IPv6-mapped IPv4 addresses are printed in the hybrid notation exemplified by ::ffff:192.168.1.1.
The address parameter must be a nonnegative integer in the appropriate range for the address type.
If the port number argument is supplied, it is included in the returned character string, according to the requirements in section 6 of RFC 5952 pertaining to IPv6 addresses (including IPv6-mapped IPv6 addresses) and section 3.2.3 of RFC 3986 for IPv4 addresses. In brief, IPv6 addresses with ports are expressed as [address]:port and IPv6 addresses follow the traditional address:port pattern.
(str-inaddr-net address [width])
(str-in6addr-net address [width])
The functions str-inaddr-net and str-in6addr-net convert, respectively, IPv4 and IPv6 network prefix addresses to the "slash notation". For IPv6 addresses, the requirements of section 2.3 of RFC 4291 are implemented. For IPv4, section 3.1 of RFC 4632 is followed.
The condensed portion of the IP address is always determined by measuring the contiguous extent of all-zero bits in the least significant position of the address. For instance an IPv4 address which has at least 24 zero bits in the least significant position, so that the only nonzero bits are in the highest octet, is always condensed to a single decimal number: the value of the first octet.
If the width parameter is specified, then its value is incorporated into the returned textual notation as the width. No check is made whether this width large enough to span all of the nonzero bits in the address.
If width is omitted, then it is calculated as the number of bits in the address, excluding the contiguous all-zero bits in the least significant position: how many times the address can be shifted to the right before a 1 appears in the least significant bit.
(inaddr-str string)
(in6addr-str string)
The inaddr-str and in6addr-str functions recover an IPv4 or IPv6 address from a textual representation. If the parse is successful, the address is returned as, respectively, a sockaddr-in or sockaddr-in6 structure.
If string is a malformed address, due to any issue such as invalid syntax or a numeric value being out of range, an exception is thrown.
The inaddr-str function recognizes the dot notation consisting of four decimal numbers separated by period characters. The numbers must be in the range 0 to 255. Note: superfluous leading zeros are permitted, though this is a nonstandard extension; not all implementations of this notations support this.
A prefix may be specified in the notation as a slash followed by a decimal number, in the range 0 to 32. In this case, the integer value of the prefix appears as the prefix member of the returned sockaddr-in structure. Furthermore, the address is masked, so that any bits not included in the prefix are zero. For instance, the address "255.255.255.255/1" is equivalent to "128.0.0.0", except that the prefix if the returned structure is 1 rather than 32. When a prefix is not specified, the prefix member of the structure retains its default value of 32. When the prefix is specified, the address part need not contain all four octets; it may contain between one and four octets. Thus, "192.168/16" is a valid address, equivalent to "192.168.0.0/16".
A port number may be specified in the notation as a colon, followed by a decimal number in the range 0 to 65535. The integer value of this port number appears as the port member of the returned structure. An example of this notation is "127.0.0.1:23".
A prefix and port number may both be specified; in this case the prefix must appear first, followed by the port number. For example, "127/8:23".
The in6addr-str function recognizes the IPv6 notation consisting of 16-bit hexadecimal pieces separated by colons. If the operation is successful, it returns a sockaddr-in6 structure. Each piece must be a value in the range 0 to FFFF. The hexadecimal digits may be any mixture of uppercase and lowercase. Leading zeros are permitted. Up to eight such pieces must be specified. If fewer pieces are specified, then the token :: (double colon) must appear in the address exactly once. That token denotes the condensation of a sufficient number of zero-valued pieces to make eight pieces. The token must be in one of three positions: it may be the leftmost element of the address, immediately followed by a hexadecimal piece; it may be the rightmost element of the address, preceded by a hexadecimal piece; or else, it may be in the middle of the address, flanked on both sides by hexadecimal pieces.
The in6addr-str also recognizes the special notation for IPv6-mapped IPv4 addresses. This notation consists of the address string "::FFFF" which may appear in any uppercase/lowercase mixture, possibly with leading zeros, followed by an IPv4 address given in the four-octet dot notation. For example, "::FFFF:127.0.0.1".
A prefix may be specified using a slash, followed by a decimal number in the range 0 to 128. The handling of the prefix is similar to that of inaddr-str except that pieces of the address may not be omitted. Condensing the pieces of the IPv6 address is always done by means of the :: token, whether or not a prefix is present. Furthermore, the octets specified in the IPv6-mapped IPv4 notation must all be present, regardless of the prefix.
A port number may be specified in the notation as follows: the entire address, including any slash-separated prefix, must appear surrounded in square brackets. The closing square bracket must be followed by a colon and one or more digits denoting a decimal number in the range 0 to 65535. For instance "[1:2:3::4/64]:1234."
(sockaddr-str string)
The function sockaddr-str analyzes the string argument to determine whether it represents a valid IPv4, IPv6 or Unix domain address. If so, it constructs an object, representing that address, of type sockaddr-in, sockaddr-in or sockaddr-un.
The slash prefix notation, and port numbers are handled, and represented in the returned structures accordingly.
The sockaddr-str function works by applying simple tests to the input, and then invoking the functions inaddr-str or in6addr-str, or constructing a sockaddr-un structure whose path slot is string.
The precise procedure followed is:
sockaddr.(str-addr)
A method named str-addr is defined for the struct types sockaddr-in, sockaddr-in6 and sockaddr-un. It returns a text representation of the address as a string. If the port slot of sockaddr-in or sockaddr-in6 is a nonzero integer, then it is incorporated into the text representation. Likewise if the prefix slot has a non-default value specifying fewer bits than the width of the address, the prefix notation is produced.
The intent is that the representations produced are suitable as input to the sockaddr-str function which will reproducing an address object of the same type, featuring the same addr, port and prefix. In the case of sockaddr-un, the sockaddr-str function will reproduce the same address only if the path slot is a string starting with "/".
TXR provides access to the terminal control "termios" interfaces defined by POSIX, and some of the extensions to it in Linux. By using termios, programs can control serial devices, consoles and virtual terminals. Terminal control in POSIX revolves around a C language structure called struct termios. This is mirrored in a TXR Lisp structure also called termios.
Like-named TXR Lisp functions are provided which correspond to the C functions tcgetattr, tcsetattr, tcsendbreak, tcdrain, tcflush and tcflow.
These have somewhat different argument conventions. The TTY device is specified last, so that it can conveniently default to the *stdin* stream. A TTY device can be specified as either a stream object or a numeric file descriptor.
The functions cfgetispeed, cfgetospeed, cfsetispeed and cfsetospeed are not provided, because they are unnecessary. Device speed (informally, "baud rate") is specified directly as a integer value in the termios structure. The TXR Lisp termios functions automatically convert between integer values and the speed constants (like B38400) used by the C API.
All of the various termios-related constants are provided, including some nonstandard ones. They appear in lowercase. For instance IGNBRK and PARENB are simply known as the predefined TXR Lisp variables ignbrk and parenb.
(defstruct termios nil
iflag oflag cflag lflag
cc ispeed ospeed)
The termios structure represents the kernel level terminal device configuration. It holds hardware related setting such as serial line speed, parity and handshaking. It also holds software settings like translations, and settings affecting input behaviors. The structure closely corresponds to the C language termios structure which exists in the POSIX API.
The iflag, oflag, cflag and lflag slots correspond to the c_iflag, c_oflag, c_cflag and c_lflag members of the C structure. They hold integer values representing bitfields.
The cc slot corresponds to the c_cc member of the C structure. Whereas the C structure's c_cc member is an array of the C type cc_t, the cc slot is a vector of integers, whose values must have the same range as the cc_t type.
These variables specify bitmask values for the iflag slot of the termios structure. They correspond to the C language preprocessor symbols IGNBRK, BRKINT, IGNPAR and so forth.
The imaxbel and iutf8 variables are specific to Linux and may not be present. Portable code should test for their presence with boundp.
The iuclc variable is a legacy feature not found on all systems.
Note: the termios methods set-iflags and clear-iflags provide a convenient means for setting and clearing combinations of these flags.
These variables specify bitmask values for the oflag slot of the termios structure. They correspond to the C language preprocessor symbols OPOST, OLCUC, ONLCR and so forth.
The variable ofdel is Linux-specific. Portable programs should test for its presence using boundp.
The olcuc variable is a legacy feature not found on all systems.
Likewise, whether the following groups of symbols are present is platform-specific: nldly, nl0 and nl1; crdly, cr0, cr1, cr2 and cr3; tabdly, tab0, tab1, tab2 and tab3; bsdly, bs0 and bs1; and ffdly, ff0 and ff1.
Note: the termios methods set-oflags and clear-oflags provide a convenient means for setting and clearing combinations of these flags.
These variables specify bitmask values for the cflag slot of the termios structure. They correspond to the C language preprocessor symbols CSIZE, CS5, CS6 and so forth.
The following are present on Linux, and may not be available on other platforms. Portable code should test for them using boundp: cbaud, cbaudex, cmspar and crtscts.
Note: the termios methods set-cflags and clear-cflags provide a convenient means for setting and clearing combinations of these flags.
These variables specify bitmask values for the lflag slot of the termios structure. They correspond to the C language preprocessor symbols ISIG, ICANON, ECHO and so forth.
The following are present on Linux, and may not be available on other platforms. Portable code should test for them using boundp: iexten, xcase, echoctl, echoprt, echoke, flusho, pendin and extproc.
Note: the termios methods set-lflags and clear-lflags provide a convenient means for setting and clearing combinations of these flags.
These variables specify integer offsets into the vector stored in the cc slot of the termios structure. They correspond to the C language preprocessor symbols VINTR, VQUIT, VERASE and so forth.
The following are present on Linux, and may not be available on other platforms. Portable code should test for them using boundp: vswtc, vreprint, vdiscard, vlnext and veol2.
These variables hold integer values suitable as the action argument of the tcflow function. They correspond to the C language preprocessor symbols TCOOFF, TCOON, TCIOFF and TCION.
These variables hold integer values suitable as the queue argument of the tcflush function. They correspond to the C language preprocessor symbols TCIFLUSH, TCOFLUSH and TCIOFLUSH.
These variables hold integer values suitable as the actions argument of the tcsetattr function. They correspond to the C language preprocessor symbols TCSANOW, TCSADRAIN and TCSAFLUSH.
(tcgetattr [device])
(tcsetattr termios [actions [device]])
The tcgetattr and tcsetattr functions, respectively, retrieve and install the configuration of the terminal driver associated with the specified device.
These functions are wrappers for the like-named POSIX C library functions, but with different argument conventions, and operating using a TXR Lisp structure.
The tcgetattr function, if successful, returns a new instance of the termios structure.
The tcsetattr function requires an instance of a termios structure as an argument to its termios parameter.
A program may alter the settings of a terminal device by retrieving them using tcgetattr, manipulating the structure returned by this function, and then using tcsetattr to install the modified structure into the device.
The actions argument of tcsetattr may be given as the value of one of the variables tcsanow, tcsadrain or tcsaflush. If it is omitted, the default is tcsadrain.
If an argument is given for device it must be either a stream, or an integer file descriptor. In either case, it is expected to be associated with a terminal (TTY) device.
If the argument is omitted, it defaults to the stream currently stored in the *stdin* stream special variable, expected to be associated with a terminal device.
The C termios structure usually does not have members for representing the input and output speed. TXR Lisp does not use such members, in any case, even if they are present. The speeds are encoded in the cc_iflag and cc_lflag bitmasks. When retrieving the settings, the tcgetattr function uses the POSIX functions cfgetispeed and cfgetospeed to retrieve the speed values from the C structure. These values are installed as the ispeed and ospeed slots of the Lisp structure. A reverse conversion takes place when setting are installed using tcsetattr: the speed values are taken from the slots, and installed into the C structure using cfsetispeed and cfsetospeed before the structure is passed to the C tcsetattr function.
On Linux, TTY devices do not have a separate input and output speed. The C termios structure stores only one speed which is taken as both the input and output speed, with a special exception. The input speed may be programmed as zero. In that case, it is independently represented. speed may be programmed as zero.
(tcsendbreak [duration [device]])
The tcsendbreak function generates a break signal on serial devices. The duration parameter specifies the length of the break signal in milliseconds. If the argument is omitted, the value 500 is used.
The device parameter is exactly the same as that of the tcsetattr function.
(tcdrain [device])
The tcdrain function waits until all queued output on a terminal device has been transmitted. It is a direct wrapper for the like-named POSIX C function.
The device parameter is exactly the same as that of the tcsetattr function.
(tcflush queue [device])
The tcflush function discards either untransmitted output data, or received and yet unread input data, depending on the value of the queue argument. It is a direct wrapper for the like-named POSIX C function.
The queue argument should be the value of one of the variables tciflush, tcoflush and tcioflush, which specify the flushing of input data, output data or both.
The device parameter is exactly the same as that of the tcsetattr function.
(tcflow action [device])
The tcflow function provides bidirectional flow control on the specified terminal device. It is a direct wrapper for the like-named POSIX C function.
The action argument should be the value of one of the variables tcooff, tcoon, tcioff and tcion.
The device parameter is exactly the same as that of the tcsetattr function.
termios.(set-iflags flags*)
termios.(set-oflags flags*)
termios.(set-cflags flags*)
termios.(set-lflags flags*)
termios.(clear-iflags flags*)
termios.(clear-oflags flags*)
termios.(clear-cflags flags*)
termios.(clear-lflags flags*)
These methods of the termios structure set or clear multiple flags of the four bitmask flag fields.
The flags arguments specify zero or more integer values. These values are combined together bitwise, as if by the logior function to form a single effective mask. If there are no flags arguments, then the effective mask is zero.
The set-iflags method sets, in the iflag slot of the termios structure, all of the bits which are set in the effective mask. That is to say, the effective mask is combined with the value in iflag by a logior operation, and the result is stored back into iflag. Similarly, the set-oflags, set-cflags and set-lflags methods operate on the oflag, cflag and lflag slots of the structure.
The clear-iflags method clears, in the iflag slot of the termios structure, all of the bits which are set in the effective mask. That is to say, the effective mask is bitwise inverted as if by the lognot function, and then combined with the existing value of the iflag slot using logand. The resulting value is stored back into the iflag slot. Similarly, the clear-oflags, clear-cflags and clear-lflags methods operate on the oflag, cflag and lflag slots of the structure.
Note: the methods go-raw, go-cbreak and go-canon are provided for changing the settings to raw, "cbreak" and canonical mode. These methods should be preferred to directly manipulating the flag and cc slots.
In this example, tio is assumed to be a variable holding an instance of a termios struct:
;; clear the ignbrk, brkint, and various other flags:
tio.(clear-iflags ignbrk brkint parmrk istrip
inlcr igncr icrnl ixon)
;; set the csize and parenb flags:
tio.(set-cflags csize parenb)
termios.(go-raw)
termios.(go-cbreak)
The go-raw and go-cbreak methods of the termios structure manipulate the flag slots, as well as certain elements of the cc slot, in order to prepare the terminal settings for, respectively, "raw" and "cbreak" mode, described below.
Note that manipulating the termios structure doesn't actually put these settings into effect in the terminal device; the settings represented by the structure must be installed into the device using tcsetattr. There is no way to reverse the effect of these methods. To precisely restore the previous terminal settings, the program should retain a copy of the original termios structure.
"Raw" mode refers to a configuration of the terminal device driver in which input and output is passed transparently and without accumulation, conversion or interpretation. Input isn't buffered into lines; as soon as a single byte is received, it is available to the program. No special input characters such as commands for generating an interrupt or process suspend request are processed by the terminal driver; all characters are treated as input data. Input isn't echoed; the only output which takes place is that generated by program output requests to the device.
"Cbreak" mode is named after a concept and function in the "curses" terminal control library. It refers to a configuration of the terminal device driver which is less transparent than "raw" mode. Input isn't buffered into lines, and line editing commands are ordinary input characters, allowing character-at-a-time input. However, most input translations are preserved, except that the conversion of CR characters to NL is disabled. The signal-generating characters are processed in this mode. This latter feature of the configuration is the likely inspiration for the word "cbreak". Unless otherwise configured, the interrupt character corresponds to the Ctrl-C key, and "break" is another term for an interactive interruption.
termios.(string-encode)
termios.(string-decode string)
The string-encode method converts the terminal state stored in a termios structure into a textual format, returning that representation as a character string.
The string-decode method parses the character representation produced by string-encode and populates the termios structure with the settings are encoded in that string.
If a string is passed to string-decode which wasn't produced by string-encode, the behavior is unspecified. An exception may or may not be thrown, and the contents of termios may or may not be affected.
Note: the textual representation produced by string-encode is intended to be identical to that produced by the -g option of the GNU Coreutils version of the stty utility, on the same platform. That is to say, the output of stty -g may be used as input into string-decode, and the output of string-encode may be used as an argument to stty.
(defstruct utsname nil
sysname nodename release
version machine domainname)
The utsname structure corresponds to the POSIX structure of the same name. An instance of this structure is returned by the uname function.
(uname)
The uname function corresponds to the POSIX function of the same name. It returns an instance of the utsname structure. Each slot of the returned structure is initialized with a character string that identifies the corresponding attribute of the host system.
The host system might not support the reporting of the NIS domain name. In this case, the domainname slot of the returned utsname structure will have the value nil.
(defstruct rlim nil
cur max)
The rlim structure is required by the functions getrlimit and setrlimit. It is analogous to the C structure by the same name described in POSIX.
These variables correspond to the POSIX constants RLIM_SAVED_MAX, RLIM_SAVED_CUR and RLIM_INFINITY. They have the same values, and are suitable as slot values of the rlim structure.
Variables @, rlimit-core @, rlimit-cpu @, rlimit-data @, rlimit-fsize @, rlimit-nofile @ rlimit-stack and @ rlimit-as
These variables correspond to the POSIX constants RLIMIT_CORE, RLIMIT_CPU, RLIMIT_DATA and so forth.
(getrlimit resource [rlim])
(setrlimit resource rlim)
The getrlimit function retrieves information about the limits imposed for a particular parameter indicated by the resource integer.
The setrlimit function changes the limit information for a resource parameter.
The resource parameter is the value of one of the variables rlimit-core, rlimit-cpu, rlimit-data and so forth.
The rlim argument is a structure of type rlim. If this argument is given to the getrlimit function, then it fills in that structure with the retrieved parameters. Otherwise it allocates a new structure and fills that one. In either situation, the filled structure is returned, if the underlying call to the host operating system is successful.
In the case of setrlimit, the rlim object must have non-negative integer values which are in the range of the platform's rlim_t type.
If the underlying system call fails, then these functions throw an exception. In the successful case, the getrlimit function returns the rlim structure, and setrlimit returns t.
Further information about resource limits is available in the POSIX standard and platform documentation.
The TXR Lisp interface to the POSIX mmap family of functions is based around the carray type. The mmap function returns a special variant of a carray object which keeps track of the memory mapping. When such an object becomes unreachable and is reclaimed by garbage collection, the mapping is automatically unmapped.
In addition to mmap, the functions munmap, mprotect, madvise and msync are provided, all taking a carray as their leftmost argument.
The TXR Lisp functions do not strictly follow the argument conventions of the same-named, corresponding POSIX functions. Adjustments which are likely to be defaulted are moved to the right. For instance, the msync operation is often applied to the entire memory mapping. Therefore, the first argument is the carray object which keeps track of the mapping. The second argument specifies the flags to be applied, which constitute the last argument of the underlying POSIX function. The remaining two arguments are the size and offset. If these are omitted, then msync applies to the entire region, whose address and size are known to the carray object.
Cautionary note: misuse of mmap and related functions can easily cause the TXR image to receive a fatal signal due to a bad memory access. Care must be taken to prevent such a situation, or else to catch such signals and recover.
(mmap ffi-type length prot flags
[source [offset [addr]]])
The mmap function provides access to the same-named POSIX platform function for creating memory mappings. The POSIX function can be used for creating virtual memory views of files and special devices. Views can be read-only, and they can be mutable. They can be in such a way that changes appear only in the mapping itself, or in such a way that the changes are actually propagated to the mapped object itself. Mappings can be shared among processes, providing a shared memory mechanism: for instance, if fork is called, any map-shared mappings created by the parent are shared with the child: the child process does not get a copy of a shared mapping, but a reference to it. The function can also be used simply for allocating memory: on some platforms, the POSIX mmap function is used as the basis for the malloc function. It behaves as a pure allocator when asked to create a mapping which is private, and anonymous (not backed by an object).
The TXR Lisp mmap function is integrated with the carray type and the FFI type system. A mapping returned by mmap is represented by a carray object.
The required ffi-type argument specifies the element type of the array; it must be a compiled FFI type. Note: this may be produced by the ffi macro. For instance, the type int may be specified using the expression (ffi int). The type must be a complete type suitable as the element type of an array; a type with a zero size such as void is invalid.
The length argument specifies the requested length of the mapping in bytes. Note that mmap allocates or configures virtual memory pages, not bytes. Internally to the system, the length argument is converted to a number of pages. If it specifies a fractional number of pages, it is rounded up. For instance, if the page size is 4096 bytes, and length is specified as 5000, it will be internally rounded up to 8192. The returned TXR Lisp carray object is oblivious to this padding: it works with the given 5000-byte size. Note: the page-size variable holds the system's page size. However, by the use of mmap extensions, it is possible for individual mappings to have their own page size. Mixed page size virtual memory systems exist.
The mmap function determines the number of elements in the array by dividing the length by the size of type, using a division that truncates toward zero. The returned carray shall have that many elements. If the division is inexact, it means that some bytes from the underlying memory mapping are unused, even if length is a multiple of the page size.
The required prot argument must be some bitwise combination of the portable values prot-read, prot-write and prot-exec. Additional system-specific prot- values may be available also for specifying additional properties. If prot is specified as zero, then the mapping, if successfully created, may be inaccessible: prot-read must be present to ensure read access, and prot-write to ensure write access.
The flags argument is a bitwise combination of values given by various map- variables. At the very least, it must contain exactly one of map-shared or map-private, to request a shared or private mapping, respectively. If a mapping is requested which is neither shared nor private, the underlying POSIX function will likely fail.
If a source is specified, indicating a filesystem object to be mapped, the map-anon flag must be omitted. Vice versa, if source is not specified, this means that the mapping will be anonymous. In this situation, the map-anon flag must be present.
The source argument may be an integer file descriptor. If so, this value will be passed to the underlying POSIX function directly. The source argument may be a stream object, in which case the fileno function will be applied to it, which must retrieve an integer file descriptor which will be passed to the POSIX function. The source argument may be a filename. The specified file is opened as if via open-file, with a mode-string which is "r+" if the prot argument includes the prot-write flag, otherwise "r". The integer file descriptor from this open stream is used in the underlying mmap call. The file is immediately closed when mmap returns. In all cases, the integer file descriptor passed to the POSIX function must be a value suitable for conversion to the int type.
Note: in the context of mmap, "anonymous" means "not associated with a filesystem object referenced by a descriptor". It does not mean "without a name", but refers to a pure memory allocation from virtual memory. Memory maps do not have a name, whether anonymous or not. Moreover, the filesystem object associated with a memory map itself does not necessarily have a name. An open file that has been deleted from the directory structure is anonymous, yet a memory mapping can be created using its descriptor, and that mapping is not "anonymous".
The offset argument is used with a non-anonymous mapping. It specifies that the mapping doesn't begin at the start of the file or file-like object, but rather at the specified offset. The offset may not be an arbitrary integer; it must be a multiple of the page size. Unless certain nonportable flags are used to specify an alternative page size, the value of the page-size variable may be relied upon to indicate the page size. If an offset is specified for an anonymous mapping, with a nonzero value, the underlying POSIX function may indicate failure.
If the length and offset values cause one or more pages to be mapped which are beyond the end of the file, then accessing those pages may produce a signal which is fatal if not handled.
The addr argument is used for specifying the address in conjunction with the map-fixed flag. Possibly, certain nonportable values in the flags argument may similarly require addr. If no bit is present in flags which requires addr, then addr should either not be specified, or specified as zero. A nonzero value of addr must be a multiple of the page size.
The mmap function returns a carray object if successful. Upon failure, an exception derived from error is thrown.
Note: when a carray object returned by mmap is identified by the garbage collector as unreachable, and reclaimed, the memory mapping is unmapped. The munmap function can be invoked on the carray to release the mapping before the object becomes garbage. The carray-free function cannot be used on a mapped carray.
(munmap carray)
The munmap function releases the memory mapping tracked by carray, which must be an object previously returned by mmap. An exception is thrown if the object is any other kind of carray.
Note: the memory mapping is released by means of the same-named POSIX function. No provision is made for selectively unmapping the pages of a mapping; the entire mapping associated with a carray is removed.
When the memory mapping is released, munmap returns t. Thereafter, the carray's contents may no longer be accessed, subject to error exceptions being thrown.
If munmap is called again on a carray on which it had previously been successfully called, the additional calls return nil.
(mprotect carray prot [offset [size]])
(madvise carray advice [offset [size]])
(msync carray flags [offset [size]])
The functions mprotect, madvise and msync perform various operations and adjustments on a memory mapping, using the same-named, corresponding POSIX functions.
All functions follow the same argument conventions with regard to the carray argument and the optional offset and size arguments. The respective second arguments prot, advice and flags are all integers. Of these, prot and flags are bitmapped flags, whereas advice specifies an enumerated command.
The prot argument is a bitwise combination of prot- values such as prot-read, prot-write and prot-exec. The mprotect function adjusts the protection bits of the mapping accordingly.
The advice argument of madvise should specify one of the following portable values, or else some system-specific nonportable madv- value: madv-normal, madv-random, madv-sequential, madv-willneed or madv-dontneed.
The flags argument of msync should specify exactly one of the values ms-async and ms-sync. Additional ms- values such as ms-invalidate may be combined in.
If offset and size are omitted, they default to zero and the size of the entire mapping, respectively, so the operation applies to the entire mapping.
If only size is specified, it must not exceed the mapping size, or an error exception is thrown. The offset argument defaults to zero.
If only offset is specified, it must not exceed the length of the mapping, or else an error exception is thrown. The size is calculated as the difference between the offset and the length. It may be zero.
If both offset and size are specified, they must not specify a region any portion of which lies outside the mapping. If size is zero, offset may be equal to the length of the mapping.
The offset must be a multiple of the page size, or else the operation will fail, since these functions work with virtual memory pages, and not individual bytes. The length is adjusted by the system to a multiple of the applicable page size, as noted in the description of mmap.
When any of these three functions succeeds, it returns t. Otherwise, it throws an exception.
The integer values of these variables are bitmasks, intended to be combined with logior to prepare a value for the flags argument of mmap.
Additional nonportable, system-dependent map- variables may be available. Their names are derived by taking the MAP_-prefixed symbol from the platform header file, converting it to lowercase and replacing underscores by hyphen characters. Any such variable which exists, but has a value of zero, is present only for compatibility with other systems. For instance map-huge-shift may be present in non-Linux ports of TXR, but with a zero value; it has a nonzero value on Linux systems to which it is specific. Applications critically relying on certain flags should test the corresponding variables for nonzero to make sure they are actually available.
The integer values of these variable are bitmasks, intended to be combined with logior to prepare a value for the prot argument of mmap and mprotect.
Additional nonportable, system-dependent prot- variables may be available. Their names are derived by taking the PROT_-prefixed symbol from the platform header file, converting it to lowercase and replacing underscores by hyphen characters. Any such variable which exists, but has a value of zero, is present only for compatibility with other systems.
The integer values of these variable are bitmasks, intended to be combined with logior to prepare a value for the advice argument of the madvise function.
Additional nonportable, system-dependent madv- variables may be available. Their names are derived by taking the MADV_-prefixed symbol from the platform header file, converting it to lower case and replacing underscores by hyphen characters. Any such variable which exists, but has a value of zero, is present only for compatibility with another system.
The integer values of these variable are bitmasks, intended to be combined with logior to prepare a value for the advice argument of the msync function.
As described under msync, exactly one of ms-async and ms-sync should be present; ms-invalidate is optional.
(url-encode string [space-plus-p])
(url-decode string [space-plus-p])
These functions convert character strings to and from a form which is suitable for embedding into the request portions of URL syntax.
Encoding a string for URL use means identifying in it certain characters that might have a special meaning in the URL syntax and representing it using "percent encoding": the percent character, followed by the ASCII value of the character. Spaces and control characters are also encoded, as are all byte values greater than or equal to 127 (7F hex). The printable ASCII characters which are percent-encoded consist of this set:
:/?#[]@!$&'()*+,;=%
More generally, strings can consists of Unicode characters, but the URL encoding consists only of printable ASCII characters. Unicode characters in the original string are encoded by expanding into UTF-8, and applying percent-encoding the UTF-8 bytes, which are all in the range \x80-\xFF.
Decoding is the reverse process: reconstituting the UTF-8 byte sequence specified by the URL-encoding, and then decoding the UTF-8 sequence into the string of Unicode characters.
There is an additional complication: whether or not to encode spaces as plus, and to decode plus characters to spaces. In encoding, if spaces are not encoded to the plus character, then they are encoded as %20, since spaces are reserved characters that must be encoded. In decoding, if plus characters are not decoded to spaces, then they are left alone: they become plus characters in the decoded string.
The url-encode function performs the encoding process. If the space-plus-p argument is omitted or specified as nil, then spaces are encoded as %20. If the argument is a value other than nil, then spaces are encoded as the character + (plus).
The url-decode function performs the decoding process. If the space-plus-p argument is omitted or specified as nil, then + (plus) characters in the encoded data are retained as + characters in the decoded strings. Otherwise, plus characters are converted to spaces.
(html-encode text-string)
(html-encode* text-string)
(html-decode html-string)
The html-encode and html-decode functions convert between an HTML and raw representation of text.
The html-encode function returns a string which is based on the content of text-string, but in which all characters which have special meaning in HTML have been replaced by HTML codes for representing those characters literally. The returned string is the HTML-encoded verbatim representation of text-string.
The html-decode function converts html-string, which may contain HTML character encodings, into a string which contains the actual characters represented by those encodings.
The function composition (html-decode (html-encode text)) returns a string which is equal to text.
The reverse composition (html-encode (html-decode html)) does not necessarily return a string equal to html.
For instance if html is the string "<p>Hello, world!</p>", then html-decode produces "<p>Hello, world!</p>". From this, html-encode produces "<p>Hello, world!</p>".
The html-encode* function is similar to html-encode except that it does not encode the single and double quote characters (ASCII 39 and 34, respectively). Text prepared by this function may not be suitable for insertion into a HTML template, depending on the context of its insertion. It is suitable as text placed between tags but not necessarily as tag attribute material.
(base64-encode [string | buf] [column-width])
(base64-decode string)
(base64-decode-buf string)
The base64-encode function converts the UTF-8 representation of string, or the contents of buf, to Base64 and returns that representation as a string. The Base64 encoding is described in RFC 4648, section 5.
The second argument must either be a character string, or a buffer object.
The base64-decode functions performs the opposite conversion; it extracts the bytes encoded in a Base64 string, and decodes them as UTF-8 to return a character string.
The base64-decode-buf extracts the bytes encoded in a Base64 string, and returns a new buffer object containing these bytes.
The Base64 encoding divides the UTF-8 representation of string or the bytes contained in buf into groups of six bits, each representing the values 0 to 63. Each value is then mapped to the characters A to Z, a to z, the digits 0 to 9 and the characters + and /. One or two consecutive occurrences of the character = are added as padding so that the number of non-whitespace characters is divisible by four. These characters map to the code 0, but are understood not to contribute to the length of the encoded message. The base64-encode function enforces this convention, but base64-decode doesn't require these padding characters.
Base64-encoding an empty string or zero-length buffer results in an empty string.
If the column-width argument is passed to base64-encode, then the Base64 encoded string, unless empty, contains newline characters, which divide it into lines which are column-width long, except possibly for the last line.
(base64-stream-enc out in [nbytes [column-width]])
(base64-stream-dec out in)
The base64-stream-enc and base64-stream-dec perform, respectively, bulk Base64 encoding and decoding between streams. This format is described in RFC 4648, section 5.
The in and out arguments must be stream objects. The out stream must support output. In the decode operation, it must support byte output. The in stream must support input. In the encode operation it must support byte input.
The base64-stream-enc function reads a sequence of bytes from the in stream and writes characters to the out stream comprising the Base64 encoding of that sequence. If the nbytes argument is specified, it must be a nonnegative integer. At most nbytes bytes will be read from the in stream. If nbytes is omitted, then the operation will read from the in stream without limit, until that stream indicates that no more bytes are available.
The optional column-with argument influences the formatting of Base64 output, in the same manner as documented for the base64-encode function.
The base64-stream-dec function reads the characters of a Base64 encoding from the in stream and writes the corresponding byte sequence to the out stream. It keeps reading and decoding until it encounters the end of the stream, or a character not used in Base64: a character that is not whitespace according to chr-isspace, isn't any of the Base64 coding characters (not an alphanumeric character, and not one of the characters +, / or =. If the function stops due to a non-Base64 character, that character is pushed back into the in stream.
The base64-stream-enc function returns the number of bytes encoded; the base64-stream-dec function returns the number of bytes decoded.
(base64url-encode [string | buf] [column-width])
(base64url-decode string)
(base64url-decode-buf string)
The base64url-encode, base64url-decode and base64url-decode-buf functions conform, in nearly every respect, to the descriptions of, respectively, base64-encode, base64-decode and base64-decode-buf. The difference is that these functions use the encoding described in section 6 of RFC 4648, rather than section 5. This means that, in the encoding alphabet, instead of the symbols + (plus) and / (slash) the symbols - (minus) and _ (underline) are used.
(base64url-stream-enc out in [nbytes [column-width]])
(base64url-stream-dec out in)
The base64url-stream-enc and base64url-stream-dec functions conform, in nearly every respect, to the descriptions of, respectively, base64-stream-enc and base64-stream-dec. The difference is that these functions use the encoding described in section 6 of RFC 4648, rather than section 5. This means that, in the encoding alphabet, instead of the symbols + (plus) and / (slash) the symbols - (minus) and _ (underline) are used.
The filter module provides a trie (pronounced "try") data structure, which is suitable for representing dictionaries for efficient filtering. Dictionaries are unordered collections of keys, which are strings, which have associated values, which are also strings. A trie can be used to filter text, such that keys appearing in the text are replaced by the corresponding values. A trie supports this filtering operation by providing an efficient prefix-based lookup method which only looks at each input character once, and which does not require knowledge of the length of the key in advance.
(make-trie)
The make-trie function creates an empty trie. There is no special data type for a trie; a trie is some existing type such as a hash table.
(trie-add trie key value)
The trie-add function adds the string key to the trie, associating it with value. If key already exists in trie, then the value is updated with value.
The trie must not have been compressed with trie-compress.
A trie can contain keys which are prefixes of other keys. For instance it can contain "dog" and "dogma". When a trie is used for matching and substitution, the longest match is used. If the input presents the text "doggy", then the match is "dog". If the input is "dogmatic", then "dogma" matches.
(trie-compress trie)
The trie-compress function changes the representation of trie to a representation which occupies less space and supports faster lookups. The new representation is returned.
The compressed representation of a trie does not support the trie-add function.
The trie-compress function destructively manipulates trie, and may return an object that is the same object as trie, or it may return a different object, while at the same time still modifying the internals of trie. Consequently, the program should not retain the input object trie, but use the returned object in its place.
(trie-lookup-begin trie)
The trie-lookup-begin function returns a context object for performing an open-coded lookup traversal of a trie. The tri argument is expected to be a trie that was created by the make-trie function.
(trie-lookup-feed-char trie-context char)
The trie-lookup-feed-char function performs a one character step in a trie lookup. The trie-context argument must be a trie context returned by trie-lookup-begin, or by some previous call to trie-lookup-feed-char. The char argument is the next character to match.
If the lookup is successful (the match through the trie can continue with the given character) then a new trie context object is returned. The old trie context remains valid.
If the lookup is unsuccessful, nil is returned.
Note: determining whether a given string is stored in a trie can be performed looking up every character of the string successively with trie-lookup-feed-char, using the newly returned context for each successive operation. If every character is found, it means that either that exact string is found in the trie, or a prefix. The ambiguity can be resolved by testing whether the trie has a value at the last node using trie-value-at. For instance, if "catalog" is inserted into an empty trie with value "foo", then "cat" will look up successfully, being a prefix of "catalog"; however, the value at "cat" is nil, indicating that "cat" is only a prefix of one or more entries in the trie.
(trie-value-at trie-context)
The trie-value-at function returns the value stored at the node in in the trie given by trie-context. Nodes which have not been given a value hold the value nil.
(filter-string-tree filter obj)
The filter-string-tree function returns a tree structure similar to obj in which all of the string atoms have been filtered through filter.
The obj argument is a string-tree structure: either the symbol nil, denoting an empty structure; a string; or a list of tree structures. If obj is nil, then filter-string-tree returns nil.
The filter argument is a filter: it is either a trie, a function, or nil. If filter is nil, then filter-string-trie just returns obj.
If filter is a function, it must be a function that can be called with one argument. The strings of the string tree are filtered by passing each one into the function and substituting the return value into the corresponding place in the returned structure.
Otherwise if filter is a trie, then this trie is used for filtering, the string elements similarly to a function. For each string, a new string is returned in which occurrences of the keys in the trie are replaced by the values in the trie.
(filter-equal filter-1 filter-2 obj-1 obj-2)
The filter-equal function tests whether two string trees are equal under the given filters.
The precise semantics can be given by this expression:
(equal (filter-string-tree filter-1 obj-1)
(filter-string-tree filter-2 obj-2))
The string tree obj-1 is filtered through filter-1, as if by the filter-string-tree function, and similarly, obj-2 is filtered through filter-2. The resulting structures are compared using equal, and the result of that is returned.
(regex-from-trie trie)
The regex-from-trie function returns a representation of trie as regular-expression abstract syntax, suitable for processing by the regex-compile function.
The values stored in the trie nodes are not represented in the regular expression.
The trie may be one that has been compressed via trie-compress; in fact, a compressed trie results in more compact syntax.
Note: this function is useful for creating a compact, prefix-compressed regular expression which matches a list of strings.
The *filters* special variable holds a hash table which associates symbols with filters. This hash table defines the named filters used in the TXR pattern language. The names are the hash-table keys, and filter objects are the values. Filter objects are one of three representations. The value nil represents a null filter, which performs no filtering, passing the input string through. A filter object may be a raw or compressed trie. It may also be a Lisp function, which must be callable with one argument of string type, and must return a string.
The application may define new filters by associating symbolic keys in *filters* with values which conform to the above representation of filters.
The behavior is unspecified if any of the predefined filters are removed or redefined, and are subsequently used, or if the *filters* variable is replaced or rebound with a hash-table value which omits those keys, or associates them with different values.
Note that functions html-encode, html-encode* and html-decode use, respectively, the HTML-related :tohtml, :tohtml* and :fromhtml.
It is useful to be able to invoke the abilities of the TXR pattern Language from TXR Lisp. An interface for doing this provided in the form of the match-fun function, which is used for invoking a TXR pattern function.
The match-fun function has a cumbersome interface which requires the TXR Lisp program to explicitly deal with the variable bindings emerging from the pattern match in the form of an association list.
To make it the interface easier to use, TXR provides the macros txr-if, txr-when and txr-case.
(match-fun name args [input [files]])
The match-fun function invokes a TXR pattern function whose name is given by name, which must be a symbol.
The args argument is a list of expressions. The expressions may be symbols which will be interpreted as pattern variables, and may be bound or unbound. If they are not symbols, then they are treated as expressions (of the pattern language, not TXR Lisp) and evaluated accordingly.
The optional input argument is an object of one of several types. It may be a stream, character string or list of strings. If it is a string, then it is converted to a list containing that string. A list of strings represents zero or more lines of text to be processed. If the input argument is omitted, then it defaults to nil, interpreted as an empty list of lines.
The files argument is a list of filename specifications, which follow the same conventions as files given on the TXR command line. If the pattern function uses the @(next) directive, it can process these additional files. If this argument is omitted, it defaults to nil.
The match-fun function's return value falls into three cases. If there is a match failure, it returns nil. Otherwise it returns a cons cell. The car field of the cons cell holds the list of captured bindings. The cdr of the cons cell is one of two values. If the entire input was processed, the cdr field holds the symbol t. Otherwise it holds another cons cell whose car is the remainder of the list of lines which were not matched, and whose cdr is the line number.
@(define foo (x y))
@x:@y
@line
@(end)
@(do
(format t "~s\n"
(match-fun 'foo '(a b)
'("alpha:beta" "gamma" "omega") nil)))
Output:
(((a . "alpha") (b . "beta")) ("omega") . 3)
In the above example, the pattern function foo is called with arguments (a b). These are unbound variables, so they correspond to parameters x and y of the function. If x and y get bound, those values propagate to a and b. The data being matched consists of the lines "alpha:beta", "gamma" and "omega". Inside foo, x and y bind to "alpha" and "beta", and then the line variable binds to "gamma". The input stream is left with "omega".
Hence, the return value consists of the bindings of x and y transferred to a and b, and the second cons cell which gives information about the rest of the stream: it is the part starting at "omega", which is line 3. Note that the binding for the line variable does not propagate out of the pattern function foo; it is local inside it.
(match-fboundp symbol)
The match-fboundp function returns t or nil if, respectively, symbol is the name of an existing pattern function.
(txr-if name (argument*) input
then-expr [else-expr])
The txr-if macro invokes the TXR pattern-matching function name on some input given by the input parameter, whose semantics are the same as the input argument of the match-fun function.
If name succeeds, then then-expr is evaluated, and if it fails, else-expr is evaluated instead.
In the successful case, then-expr is evaluated in a scope in which the bindings emerging from the name function are turned into TXR Lisp variables. The result of txr-if is that of then-expr.
In the failed case, else-expr is evaluated in a scope which does not have any new bindings. The result of txr-if is that of else-expr. If else-expr is missing, the result is nil.
The argument forms supply arguments to the pattern function name. There must be as many of these arguments as the function has parameters.
Any argument which is a symbol is treated, for the purposes of calling the pattern function, as an unbound pattern variable. The function may or may not produce a binding for that variable. Also, every argument which is a symbol also denotes a local variable that is established around then-expr if the function succeeds. For any such pattern variable for which the function produces a binding, the corresponding local variable will be initialized with the value of that pattern variable. For any such pattern variable which is left unbound by the function, the corresponding local variable will be set to nil.
Any argument can be a form other than a symbol. In this situation, the argument is evaluated, and will be passed to the pattern function as the value of the binding for the corresponding argument.
@(define date (year month day))
@{year /\d\d\d\d/}-@{month /\d\d/}-@{day /\d\d/}
@(end)
@(do
(each ((date '("09-10-20" "2009-10-20"
"July-15-2014" "foo")))
(txr-if date (y m d) date
(put-line `match: year @y, month @m, day @d`)
(put-line `no match for @date`))))
Output:
no match for 09-10-20
match: year 2009, month 10, day 20
no match for July-15-2014
no match for foo
(txr-when name (argument*) input form*)
The txr-when macro is based on txr-if. It is equivalent to
(txr-if name (argument*) input (progn form*))
If the pattern function name produces a match, then each form is evaluated in the scope of the variables established by the argument expressions. The result of the txr-when form is that of the last form.
If the pattern function fails then the forms are not evaluated, and the result value is nil.
(txr-case input-form
{(name (argument*) form*)}*
[(t form*)])
The txr-case macro evaluates input-form and then uses the value as an input to zero or more test clauses. Each test clause invokes the pattern function named by that clause's name argument.
If the function succeeds, then each form is evaluated, and the value of the last form is taken to be the result value of txr-case, which terminates. If there are no forms, then txr-case terminates with a nil result.
The forms are evaluated in an environment in which variables are bound based on the argument forms, with values depending on the result of the invocation of the name pattern function, in the same manner as documented in detail for the txr-if macro.
If the function fails, then the forms are not evaluated, and control passes to the next clause.
A clause which begins with the symbol t executes unconditionally and causes txr-case to terminate. If it has no forms, then txr-case yields nil, otherwise the forms are evaluated in order and the value of the last one specifies the result of txr-case.
The value of the input input-form is expected to be one of the same kinds of objects as given by the requirements for the input argument of the match-fun functions.
If input-form evaluates to a stream object according to the streamp function, then the stream is converted to a lazy list of lines, as if by invoking the get-lines function on that stream; that list then serves as input to the clauses.
(txr-parse [source [error-stream
[error-retval [name]]]])
The txr-parse function converts textual TXR query syntax into a Lisp data structure representation.
The source argument may be either a character string, or a stream. If it is omitted, then *stdin* is used as the stream.
The source must provide the text representation of one complete TXR query.
The optional error-stream argument can be used to specify a stream to which diagnostics of parse errors are sent. If absent, the diagnostics are suppressed.
The optional name argument can be used to specify the file name which is used for reporting errors. If this argument is missing, the name is taken from the name property of the source argument if it is a stream, or else the word string is used as the name if source is a string.
If there are no parse errors, the function returns the parsed data structure. If there are parse errors, and the error-retval parameter is present, its value is returned. If the error-retval parameter is not present, then an exception of type syntax-error is thrown.
(source-loc form)
(source-loc-str form [alternative])
These functions map an expression in a TXR program to the file name and line number of the source code where that form came from.
The source-loc function returns the raw information as a cons cell whose car/cdr consist of the line number, and file name.
The source-loc-str function formats the information as a string.
Forms which were parsed from a file have source location info tracking to their origin in that file. Forms which are the result of macro-expansion are traced to the form whose evaluation produced them. That is to say, they inherit that form's source location info.
More precisely, when a form is produced by macro-expansion, it usually consists of material which was passed to the macro as arguments, plus some original material allocated by the macro, and possibly literal structure material which is part of the macro code. After the expansion is produced, any of its constituent material which already has source location info keeps that info. Those nodes which are newly allocated by the macro-expansion process inherit their source location info from the form which yields the expansion.
If form is not a piece of the program source code that was constructed by the TXR parser or by a macro, and thus it was neither attributed with source location info, nor has it inherited such info, then source-loc returns nil.
In the same situation, and if its alternative argument is missing, the source-loc-str returns a string whose text conveys that the source location is not available. If the alternative argument is present, it is returned.
(rlcp dest-form source-form)
(rlcp dest-tree source-form)
The rlcp function copies the source code location info ("rl" means "read location") from the source-form object to the dest-form object. These objects are pieces of list-based syntax. If dest-form already has source code location info, then no copying takes place.
The rlcp-tree function copies the source code location info from rlcp into every cons cell in the dest-tree tree structure which doesn't already have location info. It may be regarded as a recursive application of rlcp via car/cdr recursion on the tree structure. However, the traversal performed by rlcp-tree gracefully handles circular structures.
Note: these functions are intended to be used in certain kinds of macros. If a macro transforms source-form to dest-form, this function can be used to propagate the source code location info also, so that when the TXR Lisp evaluator encounters errors in transformed code, it can give diagnostics which refer to the original untransformed source code.
The macro expander already performs this transfer. If a macro call form has location info, the expander propagates that info to that form's expansion. In some situations, it is useful for a macro or other code transformer to perform this action explicitly.
The Boolean special variable *rec-source-loc* controls whether the read and iread functions record source location info. The variable is nil by default, so that these functions do not record source location info. If it is true, then these functions record source location info.
Regardless of the value of this variable, source location info is recorded for Lisp forms which are read from files or streams under the load function or specified on the TXR command line. Source location info is also always recorded when reading the TXR pattern language syntax.
Note: recording and propagating location info incurs a memory and performance penalty. The individual cons cells and certain other literal objects in the structure which emerges from the parser are associated with source location info via a global weak hash table.
(macro-ancestor form)
The macro-ancestor function returns information about the macro-expansion ancestor of form. The ancestor is the original form whose expansion produced form.
If form is not the result of macro-expansion, or the ancestor information is unavailable, the function returns nil.
(prof form*)
The prof operator evaluates the enclosed forms from left to right similarly to progn, while determining the memory allocation requests and time consumed by the evaluation of the forms.
If there are no forms, the prof operator measures the smallest measurable operation of evaluating nothing and producing nil.
If the evaluation terminates normally (not abruptly by a nonlocal control transfer), then prof yields a list consisting of:
(value malloc-bytes gc-bytes milliseconds)
where value is the value returned by the rightmost form, or nil if there are no forms, malloc-bytes is the total number of bytes of all memory allocation requests (or at least those known to the TXR runtime, such as those of all internal objects), gc-bytes is the total number of bytes drawn from the garbage-collected heaps, and milliseconds is the total processor time consumed over the execution of those forms.
Notes:
The bytes allocated by the garbage collector from the C function malloc to create heap areas are not counted as malloc-bytes. malloc-bytes includes storage such as the space used for dynamic strings, vectors and bignums (in addition to their gc-heap-allocated nodes), and the various structures used by the cobj type objects such as streams and hashes. Objects in external libraries that use uninstrumented allocators are not counted: for instance the C FILE * streams.
(pprof form*)
The pprof (pretty-printing prof) macro is similar to progn. It evaluates forms, and returns the rightmost one, or nil if there are no forms.
Over the evaluation of forms, it counts memory allocations, and measures CPU time. If forms terminate normally, then just prior to returning, pprof prints these statistics in a concise report on the *stdout*.
The pprof macro relies on the prof operator.
(sys:gc [full])
The gc function triggers garbage collection. Garbage collection means that unreachable objects are identified and reclaimed, so that their storage can be reused.
The function returns nil if garbage collection is disabled (and consequently nothing is done), otherwise t.
The Boolean full argument, defaulting to nil, indicates whether a full garbage collection should be requested.
Even if this argument is nil, a full garbage collection may occur due to having been scheduled.
(sys:gc-set-delta bytes)
The gc-set-delta function sets the GC delta parameter.
Note: This function may disappear in a future release of TXR or suffer a backward-incompatible change in its syntax or behavior.
When the amount of new dynamic memory allocated since the last garbage collection equals or exceeds the GC delta, a garbage collection pass is triggered. From that point, a new delta begins to be accumulated.
Dynamic memory is used for allocating heaps of small garbage-collected objects such as cons cells, as well as the satellite data attached to some objects: like the storage arrays of vectors, strings or bignum integers. Most garbage collector behaviors are based on counting objects in the heaps.
Sometimes a program works with a small number of objects which are very large, frequently allocating new, large objects and turning old ones into garbage. For instance a single large integer could be many megabytes long. In such a situation, a small number of heap objects therefore control a large amount of memory. This requires garbage collection to be triggered much more often than when working with small objects, such as conses, to prevent runaway allocation of memory. It is for this reason that the garbage collector uses the GC delta.
There is a default GC delta of 64 megabytes. This may be overridden in special builds of TXR for small systems.
(finalize object function [reverse-order-p])
The finalize function registers function to be invoked in the situation when object is identified by the garbage collector as unreachable. A function registered in this way is called a finalizer.
If and when this situation occurs, the finalizer function will be called with object as its only argument.
Multiple finalizer functions can be registered for the same object, up to an internal limit which is not required to be greater than 255. If the limit is exceeded, finalize throws an error exception.
All registered finalizers are called when the object becomes unreachable. Finalizers registered against an object may also be invoked and removed using the call-finalizers function.
If the reverse-order-p argument isn't specified, or is nil, then finalizer is registered at the end of the list.
If reverse-order-p is true, then the finalizer is registered at the front of the list.
Finalizers which are activated in the same finalization processing phase are called in the order in which they appear in the registration list.
After a finalization call takes place, its registration is removed. However, neither object nor function are reclaimed immediately; they are treated as if they were reachable objects until at least the next garbage collection pass. It is therefore safe for function to store somewhere a persistent reference to object or to itself, thereby reinstating these objects as reachable.
A finalizer is itself permitted to call finalize to register the original object or any other object for finalization. Finalization processing can be understood as taking place in one or more rounds. At the start of each round, finalizers are identified that are to be called, arranged in order, and removed from the registration list. If this identification stage produces no finalizers, then finalization ends. Otherwise, those finalizers are processed, and then another round is initiated, to look for eligible finalizers that may have been registered during the previous round.
Note: it is possible for the application to create an infinite finalization loop, if one or more objects have finalizers that register new finalizers, which register new finalizers and so on.
Note: if a finalizer is invoked by the garbage collector rather than explicit finalization via call-finalizers, and that finalizer calls finalize to make a registration, that registration will not be eligible for processing in the same phase, because the criteria for finalization is unreachability.
(call-finalizers object)
The call-finalizers function invokes and removes the finalizers, if any, registered against object. If any finalizers are called, it returns t, otherwise nil.
Finalization performed by call-finalizers works in the manner described under the specification of the finalize function.
It is permissible for a finalizer function itself to call call-finalizers. Such a call can happen in two possible contexts: finalization initiated by by garbage collection, or under the scope of a call-finalizers invocation from application code. Doing so is safe, since the finalization logic may be reentered recursively. When finalizers are being called during a round of processing, those finalizers have already been removed from the registration list, and will not be redundantly invoked by a recursive invocation of finalization.
Under the scope of garbage-collection-driven reclamation, the order of finalizer calls may not be what the application logic expects. For instance even though a finalizer registered for some object A itself invokes (call-finalizers B), it may be the case during GC reclamation that both A and B are identified as unreachable objects at the same time, and some or all finalizers registered against B have already been called before the given A finalizer performs the explicit call-finalizers invocation against B. Thus the call either has no effect at all, or only calls some remaining B finalizers that have not yet been processed, rather than all of them, as the application expects.
The application must avoid creating a dependency on the order of finalization calls, to prevent the situation that the finalization actions are only correct under an explicit call-finalizers but incorrect under spontaneous reclamation driven by garbage collection.
TXR features a rudimentary mechanism for guarding against stack overflows, which cause the TXR process to crash. This capability is separate from and exists in addition to the possibility of catching a sig-segv (segmentation violation) signal upon stack overflow using set-sig-handler.
The stack-overflow guard mechanism is based on TXR, at certain key places in the execution, checking the current position of the stack relative to a predetermined limit. If the position exceeds the limit, then an exception of type stack-overflow, derived from error, is thrown.
The stack-overflow guard mechanism is configured on startup. On platforms where it is possible to inquire the system's actual stack limit, and where the stack limit is at least 512 kilobytes, TXR sets the limit to within a certain percentage of the actual value. If it is not possible to determine the system's stack limit, or if the system indicates that the stack size is unlimited, then a default limit is imposed. If the system's limit is configured below a certain small value, then that small value is used as the stack limit.
The get-stack-limit and set-stack-limit functions are provided to manipulate the stack limit.
The mechanism cannot contain absolutely all sources of stack-overflow threat under all conditions. External functions are not protected, and not all internal functions are monitored. If TXR is close to the limit, but a function is called whose stack growth is not monitored, such as an external function or unmonitored internal function, it is possible that the stack may overflow anyway.
(get-stack-limit)
(set-stack-limit value)
The get-stack-limit returns the current value of the stack limit. If the guard mechanism is not enabled, it returns nil, otherwise it returns a positive integer, which is measured in bytes.
The set-stack-limit configures the stack limit according to value, possibly enabling or disabling the guard mechanism, and returns the previous stack limit in exactly the same manner as get-stack-limit.
The value must be a non-negative integer or else the symbol nil.
The values zero or nil disable the guard mechanism. Positive integer values set the limit. The value may be truncated to a multiple of some denomination or otherwise adjusted, so that a subsequent call to get-stack-limit need not retrieve that exact value.
If value is too close to the system's stack limit or beyond, the effectiveness of the stack-overflow detection mechanism is compromised. Likewise, if value is too low, the operation of TXR shall become unreliable. Values smaller than 32767 bytes are strongly discouraged.
This variable holds the invocation pathname of a TXR program that was specified on the command line.
The value of self-path when TXR Lisp expressions are being evaluated in command-line arguments is the string "cmdline-expr". The value of self-path when a TXR query is supplied on the command line via the -c command-line option is the string "cmdline".
When a file is being compiled using the --compile option, the value of self-path is the source file path.
When the interactive listener is entered, self-path is set to the value "listener", even if prior to that, a file was compiled or executed, for which self-path had been set to the name of that file.
Note that for programs read from a file, self-path holds the resolved name, and not the invocation name. For instance if foo.tl is invoked using the name foo, whereby TXR infers the suffix, then self-path holds the suffixed name.
Note that the functions load, compile-file and compile-update-file have no effect on the value of self-path. The variable is set strictly by command line processing.
The stdlib variable expands to the directory where the TXR standard library is installed. It includes the trailing slash.
Note: there is no need to use the value of this variable to load library modules. Library modules are keyed to specific symbols, and lazily loaded. When a TXR Lisp library function, macro or variable is referenced for the first time, the library module which defines it is loaded. This includes references which occur during the code expansion phase, at "macro time", so it works for macros. In the middle of processing a syntax tree, the expander may encounter a symbol that is registered for autoloading, and trigger the load. When the load completes, the symbol might now be defined as a macro, which the expander can immediately use to expand the given form that is being traversed.
(load target load-arg*)
The load function causes a file containing TXR Lisp or TXR code to be read and processed. The target argument is a string. The function can load TXR Lisp source files as well as compiled files.
Firstly, the value in target is converted to a tentative pathname as follows.
If target specifies a pure relative pathname, as defined by the pure-rel-path-p function, then a special behavior applies. If an existing load operation is in progress, then the special variable *load-path* has a binding. In this case, load will assume that the relative pathname is a reference relative to the directory portion of that pathname. If *load-path* has the value nil, then a pure relative target pathname is used as-is, and thus resolved relative to the current working directory.
Once the tentative pathname is determined, load determines whether the name is suffixed. The name is suffixed if it ends in any of these four suffixes: .tlo, .tlo.gz, .tl, .txr or .txr_profile.
Depending on whether the tentative pathname exists, and whether or not it is suffixed, load tries to make one or more attempts to open several variations of that name. These variations are called actual paths. If any attempt fails due to an error other than non-existence, such as a permission error, then no further attempts are made; the error exception propagates to load's caller.
Regardless of whether the tentative pathname is suffixed, load tries to open a file by that actual pathname first. If that attempt fails for a suffixed pathname, or fails due to a reason other than non-existence, no other names are tried.
If an unsuffixed tentative pathname refers to a nonexistent file, .tlo is appended to the name, and an attempt is made to open a file with the resulting path. If that file is not found, then the suffixes .tlo.gz and .tl are similarly tried.
If the above initial attempts to find the file fail, and the failure is due to the file not being found rather than some other problem such as a permission error, and target isn't an absolute path according to abs-path-p, then additional attempts are made by searching for the file in the list of directories given in the *load-search-dirs* variable. For each directory taken from this variable, the directory is combined with the relative target as if using the path-cat function, and the resulting path is tried, with all the same suffix probing that is performed by the initial attempts. If any such a path is pure relative, it is interpreted relative to the current working directory, and not relative *load-path*: only the initial attempts have that special behavior.
An exception is thrown if a file is not found, or if any attempt to open a file results in an error other than non-existence.
If an unsuffixed file is successfully opened, its contents are treated as interpreted Lisp. Files ending in .txr_profile are also treated as interpreted Lisp. Files ending in .tlo are treated as compiled Lisp, and those ending in .txr are treated as the TXR Pattern Language.
The .tlo.gz suffix denotes a file which is expected to be compressed in the gzip format, and to contain compiled Lisp.
If the file is treated as TXR Lisp, then Lisp forms are read from it in succession. Each form is evaluated as if by the eval function, before the next form is read. If a syntax error is encountered, an exception of type eval-error is thrown.
If a file is treated as a compiled TXR Lisp object file, then the compiled images of top-level forms are read from it, converted into compiled objects, and executed.
If the file treated as TXR Pattern Language code, then its contents are parsed in their entirety. If the parse is successful, the query is executed. Previous TXR pattern variable and function bindings are in effect. If the query binds new variables and functions, these emerge from the load and take effect. If the parse is unsuccessful, an exception of type query-error is thrown.
Parser error messages are directed to the *stderr* stream.
Over the evaluation of either a TXR Lisp, compiled file, or TXR file, load establishes a new dynamic binding for several special variables:
Over the evaluation of either a TXR Lisp, compiled file, or TXR file, load establishes a block named load, which makes it possible for the loaded module to abort the loading using the (return-from load expr) expression. In this situation, the value of expr will appear as the return value of the load function.
When a TXR Lisp file, or compiled file, is executed from the TXR command line in such a way that TXR will terminate when that file's last form has been evaluated, then if that file performs a return-from the load block, the value of expr will turn into the termination status in exactly the same way as if that value were used as an argument to the exit function. However, if TXR has been instructed to enter into the Listener after executing the file, then the value of expr is discarded.
A block named load is also established by the @(load) directive in the pattern language. That directive provides no access to the returned value. The block is also visible to the file processed from the command line. When a such a file aborts the load via return, the returned value is discarded. If the interactive option -i was specified, the interactive listener will be entered, otherwise the process will terminate successfully.
When the load function terminates normally after processing a file, it returns nil. If the file contains a TXR pattern query which is processed to completion, the matching success or failure of that query has no bearing on the return value of load. Note that this behavior is different from the @(load) directive which itself fails if the loaded query fails, causing subsequent directives not to be processed.
A TXR pattern language file loaded with the Lisp load function does not have the usual implicit access to the command-line arguments, unlike a top-level TXR query. If the directives in the file try to match input, they work against the *stdin* stream. The @(next) directive behaves as it does when no more arguments are available.
If the source or compiled file begins with the characters #!, usually indicating a hash-bang script, load reads reads the first line of the file and discards it. Processing of the file then begins with the first byte following that line.
Two or more .tlo files produced by the same version of TXR may be catenated together (for instance, using the cat-files function) to produce a single .tlo file. Such a combined file can be loaded with the load function. The same is true of .tlo.gz files, because the gzip format supports catenation. Mixing is not possible: .tlo and .tlo.gz files cannot be catenated together.
Note: this is a single load operation: all of the binding and unbinding of variables like *load-path* and *package* is performed once over the entire contents of the combined file, and any *load-hooks* are performed one time after the load operation. Therefore it is possible that the load-time behavior differs from that of loading the original files individually. The *load-path* is bound to the name of the combined file.
The *load-path* special variable has a top-level value which is nil.
When a file is being loaded, it is dynamically bound to the pathname of that file. This value is visible to the forms are evaluated in that file during the loading process.
The *load-path* variable is bound when a file is loaded from the command line.
If the -i command-line option is used to enter the interactive listener, and a file to be loaded is also specified, then the *load-path* variable remains bound to the name of that file inside the listener.
The load function establishes a binding for *load-path* prior to processing and evaluating all the top-level forms in the target file. When the forms are evaluated, the binding is discarded and load returns.
The compile-file function also establishes a binding for *load-path*.
The @(load) directive, also similarly establishes a binding around the parsing and processing of a loaded TXR source file.
Also, during the processing of the profile file (see Interactive Profile File), the variable is bound to the name of that file.
The *load-search-dirs* variable holds a list of directories which are searched for a file to be loaded by the load function, the @(load) and @(include) directives, as well as by TXR's command line processing.
Each of these situations first searches for a file in its characteristic way. If that fails due to the file not being found, and the name is a relative path, then the directories in *load-search-dirs* are probed, in order.
The variable is initialized to a list which contains exactly one directory: a lib/ directory dynamically calculated relative to TXR the executable location. Then intent is that third-party library modules may be installed there, and easily found by load. For more information, see the section Deployment Directory Structure.
The *load-search-dirs* isn't influenced by any environment variables, which is deliberate. If a system has multiple installations of different versions of TXR in different locations, an environment variable intended for one installation could be mistakenly used by the others, resulting in chaos.
The *load-hooks* variable is at the centre of a mechanism which associates the deferred execution of actions, associated with a loaded module or program termination.
The application may push values onto this list which are expected to be functions, or objects that may be called as functions. These objects must be capable of being called with no arguments.
In the situations specified below, the list of functions is processed as follows. First *load-hooks* is examined, the list which it holds is remembered. Then the variable is reset to nil, following which the remembered list is traversed in order. Each of the functions in the list is invoked, with no arguments.
The *load-hooks* list is processed, as described above, whenever the load function terminates, whether normally or by throwing an exception. In this situation, the *load-hooks* variable which is accessed is that binding which was established by that invocation of load. The execution of the functions from the *load-hooks* list takes place in the dynamic environment of the load: all of the dynamic variable bindings established by that load are still visible, including that of *load-hooks*.
The *load-hooks* list is also processed after processing a TXR or TXR Lisp file that is specified on the command line. If the interactive listener is also being entered, this processing of *load-hooks* occurs prior to entering the listener. This situation occurs in the context of the top-level dynamic environment, and so the global value of *load-hooks* is referenced.
Lastly, *load-hooks* is also processed if the TXR process terminates normally, regardless of its exit status. This processing executes in whatever dynamic environment is current at the point of exit, using its value of the *load-hooks* variable is used. It is unspecified whether, at exit time, the *load-hooks* functions are executed first, or whether the functions registered by at-exit-call are executed first. However, their executions do not interleave.
Note that *load-hooks* is not processed after the listener reads the .txr_profile file. Hooks installed by the profile file will activate when the process exits.
(load-args-recurse file-list)
(load-args-recurse file*)
The load-args-recurse function loads multiple files, passing down the current *load-args* to each one.
It may be invoked with a single argument which is a list of files, or else it may be given multiple arguments which are files.
Each file is passed to the load function, along with extra arguments coming from the current *load-args* value.
Note: the purpose of load-args-recurse is to support a module organization of system whereby modules have local top-level files that respond to various actions specified via *load-args*, actions such as compiling, loading or cleaning. The load-args-recurse function allows such modules to not only perform the actions requested in *load-args* locally, but also pass it down to submodules which then do the same.
(load-args-process file-list)
(load-args-process file*)
The load-args-process function performs one of several actions over the specified files, those actions being distinguished by the value in *load-args*.
In addition, some of the actions are also performed for the file indicated in the current value of *load-path*.
It may be invoked with a single argument which is a list of files, or else it may be given multiple arguments which are files.
If there is exactly one argument in *load-args*, the function responds to the following values of that argument:
Note: The load-args-process function supports a protocol for organizing a program into library modules.
Suppose a module located in the "path/to/application" path consists of the files "command" "data" "reports" and "main ." Further, suppose that there are two submodules in the "utils" directory relative to this directory: "database" and "date".
Then the application might have a file called "path/to/application/app.tl" with this content:
(compile-only
(load-args-recurse
"utils/database/db"
"utils/date/date")
(load-args-process
"command"
"data"
"reports"
"main"))
Furthermore, the "database" module similarly provides a "path/to/application/utils/database/db.tl" file with this content:
(compile-only
(load-args-process
"postgres"
"mariadb"
"sqlite"))
Lastly, the "date" module provides a file "path/to/application/utils/date/date.tl" with this content:
(compile-only
(load-args-process
"src/date.tl"))
Then, to load the application and the submodules, all that is needed is (load "path/to/application/app").
Furthermore, the modules may be compiled using (load "path/to/application/app" :compile). Now the *load-args* being passed is (:compile) which tells every load-args-process invocation to compile the file in which it occurs as well as its arguments.
First, the app module's load-args-recurse call is executed, causing the "database" and "date" modules to compile.
First, the "database" module's "db.tl" top file is compiled, if necessary, and then likewise the "postgres.tl", "mariadb.tl" and "sqlite.tl" files.
Then the "date" module is similarly processed, due to its own invocation of load-args-process.
Finally the load-args-process call in the "app" module compiles "app.tl", "command.tl", "data.tl" "reports.tl" and "main.tl"
If the :clean keyword is passed via *load-args* instead of :compile, then compiled files are recursively removed. The next time the application is loaded, source files will be loaded rather than compiled files.
Note that the load-args-recurse and load-args-process forms are placed into a compile-only form so that the file compiler refrains from executing them.
(push-after-load form*)
(pop-after-load)
The push-after-load and pop-after-load macros work with the *load-hooks* list.
The push-after-load macro's arguments are zero or more forms. These forms are converted into the body of an anonymous function, which is pushed onto the *load-hooks* list. The return value is the new value of *load-hooks*.
The pop-after-macro removes the first item from *load-hooks*. The return value is the new value of *load-hooks*.
The following equivalences hold:
(push-after-load ...) <--> (push (lambda () ...) *load-hooks*)
(pop-after-load) <--> (set *load-hooks* (cdr *load-hooks*))
(load-for {(kind sym target load-arg*)}*)
The load-for macro takes multiple arguments, each of which is a three-element clause. Each clause specifies that a given target file is to be conditionally loaded based on whether a symbol sym has a certain kind of binding.
Each argument clause has the syntax (kind sym target load-arg*) where kind is one of the five symbols var, fun, macro, struct or pkg. The sym element is a symbol suitable for use as a variable, function or structure name, and target is an expression which is evaluated to produce a value that is suitable as an argument to the load function.
First, all target expressions in all clauses are unconditionally evaluated in left-to-right order. Then the clauses are processed in that order. If the kind symbol of a clause is var, then load-for tests whether sym has a binding in the variable namespace using the boundp function. If a binding does not exist, then the value of the target expression is passed to the load function. Otherwise, load is not called. Similarly, if kind is the symbol fun, then sym is instead tested using fboundp, if kind is macro, then sym is tested using mboundp, if kind is struct, then sym is tested using find-struct-type, and if kind is pkg, then sym is tested using find-package.
When load-for invokes the load function, it confirms whether loading file has had the expected effect of providing a definition of sym of the right kind. If this isn't the case, an error is thrown.
The load function is invoked with any load-arg arguments specified in the clause. The load-arg expressions of all clauses are unconditionally evaluated in order before load-arg performs any other action.
The load-for function returns the value returned by the rightmost load that was actually performed. If no loads are performed, it returns nil.
This variable holds the absolute pathname of the executable file of the running TXR instance.
The *trace-output* special variable holds a stream to which all trace output is sent. Trace output consists of diagnostics enabled by the trace macro.
(trace function-name*)
(untrace function-name*)
The trace and untrace macros control function tracing.
When trace is called with one or more arguments, it considers each argument to be the name of a global function. For each function, it turns on tracing, if it is not already turned on. If an argument denotes a nonexistent function, or is invalid function name syntax, trace terminates by throwing an exception, without processing the subsequent arguments, or undoing the effects already applied due to processing the previous arguments.
When trace is called with no arguments, it lists the names of functions for which tracing is currently enabled. In other cases it returns nil.
When untrace is called with one or more arguments, it considers each argument to be the name of a global function. For each function, it turns off tracing, if tracing is enabled.
When untrace is called with no arguments, it disables tracing for all functions.
The untrace macro always returns nil and silently tolerates arguments which are not names of functions currently being traced.
Tracing a function consists of printing a message prior to entry into the function indicating its name and arguments, and another message upon leaving the function indicating its return value, which is syntactically correlated with the entry message, using a combination of matching and indentation. These messages are posted to the *trace-output* stream.
When traced functions call each other or recurse, these trace messages nest. The nesting is detected and translated into indentation levels.
Tracing works by replacing a function definition with a trace hook function, and retaining the previous definition. The trace hook calls the previous definition and produces the diagnostics around it. When untrace is used to disable tracing, the previous definition is restored.
Methods can be traced; their names are given using (meth struct slot) syntax: see the func-get-name function.
Macros can be traced; their names are given using (macro name) syntax. Note that trace will not show the destructured internal macro arguments, but only the two arguments passed to the expander function: the whole form, and the environment.
The trace and untrace functions return nil.
(dlopen [{lib-name | nil} [flags])
The dlopen function provides access to the POSIX C library function of the same name.
The argument to the optional lib-name parameter may be a character string, or nil.
If it is nil, then the POSIX function is called with a null pointer for its name argument, returning the handle for the main program, if possible.
The flags argument should be expressed as some bitwise combination of the values of the variables rtld-lazy, rtld-now, or other rtld- variables which give names to the dlopen-related flags. If the flags argument is omitted, the default value used is rtld-lazy.
If the function succeeds, it returns an object of type cptr which represents the open library handle ("dlhandle").
Otherwise it throws an exception, whose message incorporates, if possible, error text retrieved from the dlerror POSIX function.
The cptr handle returned by dlopen will automatically be subject to dlclose when reclaimed by the garbage collector.
(dlclose dlhandle)
The dlclose closes the library indicated by dlhandle, which must be a cptr object previously returned by dlopen.
The handle is closed by passing the stored pointer to the POSIX dlclose function. The internal pointer contained in the cptr object is then reset to null.
It is permissible to invoke dlclose more than once on a cptr object which was created by dlopen. The first invocation resets the cptr object's pointer to null; the subsequent invocations do nothing.
The dlclose function returns t if the POSIX function reports a successful result (zero), otherwise it returns nil. It also returns nil if invoked on a previously closed, and hence nulled-out cptr handle.
(dlsym dlhandle sym-name)
(dlvsym dlhandle sym-name ver-name)
The dlsym function provides access to the same-named POSIX function. The dlvsym function provides access to the same-named GNU C Library function, if available.
The dlhandle argument must be a cptr handle previously returned by dlopen and not subsequently closed by dlclose or altered in any way.
The sym-name and ver-name arguments are character strings.
If these functions succeed, they return a cptr value which holds the address of the symbol which was found in the library.
If they fail, they return a cptr object containing a null pointer.
(dlsym-checked dlhandle sym-name)
(dlvsym-checked dlhandle sym-name ver-name)
The dlsym-checked and dlvsym-checked functions are alternatives to dlsym and dlvsym, respectively. Instead of returning a null cptr on failure, these functions throw an exception.
These variables provide the same values as constants in the POSIX C library header <dlfcn.h> named RTLD_LAZY, RTLD_NOW, RTLD_LOCAL, etc.
(json [quote | sys:qquote] object)
The json macro exists in support of the JSON literal and quasiquote #Jjson-syntax and #J^json-syntax notations, which use the macro as their target abstract syntax.
The macro transforms itself by deleting the json symbol, producing either the (quote object) quote syntax, or else the (sys:qquote object) quasiquote syntax, depending on which quoting symbol is present.
If the application produces and expands a json macro form which does not conform to this syntax, or does not specify one of the above two quoting symbols, the behavior is unspecified.
(put-json obj [stream [flat-p]])
(put-jsonl obj [stream [flat-p]])
The put-json function converts obj into JSON notation, and writes that notation into stream as a sequence of characters.
If stream is an external stream such as a file stream, then the JSON is rendered by conversion of the characters into UTF-8, in the usual manner characteristic of those streams.
The behavior is unspecified if obj or any component of obj is an object incompatible with the JSON representation conventions. An exception may be thrown.
An object conforms to the JSON representation conventions if it is:
A list of object is rendered in the same way as vector, in the JSON [] notation. When such JSON notation is parsed, a vector is produced.
A structure object is rendered into JSON using the {} object notation. The keys of the objects are the names of the symbols of the object type's non-static slots, appearing as a string. The values are the values of the slots. They must be JSON-conforming objects. If the special variable *print-json-type* is true, the object includes a key named "__type" whose value is the structure type symbol, appearing as a string. When present, this key occurs first in the printed representation, before any other keys. Both the slot symbols and the type symbol may appear with a package qualifier, depending on the relationship of the symbols to the current package, according to similar rules as if the symbol were printed by the print function.
When integer objects are output, they may not constitute valid JSON, since the JSON specification supports only IEEE 64 bit floating-point numbers. JSON numbers are read as floating-point.
If the flat-p argument is present and has a true value, then the JSON is generated without any line breaks or indentation. Otherwise, the JSON output is subject to such formatting.
The difference between put-json and put-jsonl is that the latter emits a newline character after the JSON output.
When a string object is output as JSON string syntax, the following rules
The put-json and put-jsonl functions return t.
Some of the JSON-related functions carry a mode-opts optional parameter. These functions open a file as if using the open-file function, using a mode-string appropriate to their direction of data transfer. If an argument is given to mode-opts, it specifies the options part to be added to the mode-string.
(tojson obj [flat-p])
The tojson function converts obj into JSON notation, returned as a character string.
The function can be understood as constructing a string output stream, calling the put-json function to write the object into that stream, and then retrieving and returning the constructed string.
The flat-p argument is passed to put-json.
(get-json [source
[err-stream
[err-retval [name [lineno]]]]])
The get-json function closely resembles the read function, and follows the same argument and error reporting conventions.
Rather than reading a Lisp object from the input source, it reads a JSON object, with support for TXR's JSON extensions.
If an object is successfully read, its Lisp representation is returned. JSON numbers produce floating-point number objects. JSON strings produce string objects. The keywords true, false and null map to the Lisp symbols t, nil, and null, respectively. JSON objects map to hash tables, and JSON arrays to vectors.
(put-jsons seq [stream [flat-p]])
The put-jsons function writes multiple JSON representations into stream. The objects are specified by the seq argument, which must be an iterable object. The put-jsons function iterates over seq and writes each element to the stream as if by using the put-jsonl function. Consequently, a newline character is written after each object.
If the stream argument is not specified, the parameter takes on the value of *stdout*.
The flat-p argument has the same meaning as in put-json with regard to the individual elements. If it is specified and true, then exactly as many lines of text are written to stream as there are elements in seq.
The put-jsons function returns t.
(get-jsons [source])
The get-jsons function reads zero or more JSON representations from source until an end-of-stream or error condition is encountered.
If source is a character string, then the input takes place from a stream created from the character string using make-string-byte-input-stream. Otherwise, if source is specified, it must be an input stream supporting byte input; input takes place from that stream. If the source argument is omitted, it defaults to *stdin*.
The objects are read as if by calls to get-json and accumulated into a list.
If the end-of-stream condition is read, then the list of accumulated objects is returned. If an error occurs, then an exception is thrown and the list of accumulated objects is not available.
If an end-of-stream condition occurs before any character is seen other than JSON whitespace, then the empty list nil is returned.
(file-get-json name [mode-opts])
(file-get-jsons name [mode-opts])
The file-get-json and file-get-jsons function open a text stream over the file indicated by the string argument name for reading. The functions ensure that the stream is closed when they terminate.
The file-get-json function invokes get-json to read a single JSON object, which is returned if that function returns normally.
The file-get-jsons function invokes get-jsons to retrieve a list of JSON objects from the stream, which is returned if that function returns normally.
(file-put-json name obj [flat-p [mode-opts]])
(file-put-jsons name seq [flat-p [mode-opts]])
The file-put-json and file-put-jsons functions open a text stream over the file indicated by the string argument name, using the function open-file with a mode-string argument of "w", write the argument object into the stream in their specific manner, and then close the stream.
The file-put-json function writes a JSON representation of obj using the put-json function. The flat-p argument is passed to that function, defaulting to nil. The value returned is that of put-json.
The file-put-jsons function writes zero or more JSON representations of objects from seq, which must be an iterable object, using the put-jsons function. The flat-p argument is passed to that function, defaulting to nil. The value returned is that of put-jsons.
(file-append-json name obj [flat-p [mode-opts]])
(file-append-jsons name seq [flat-p [mode-opts]])
The file-append-json and file-append-jsons are identical in almost all requirements to the functions file-put-json and file-put-jsons.
The only difference is that when these functions open a text stream using open-file, they specify a mode-string argument of "a" rather than "w", in order to append data to the target file rather than overwrite it.
(command-get-json cmd [mode-opts])
(command-get-jsons cmd [mode-opts])
The command-get-json and command-get-jsons functions opens text stream over an input command pipe created for the command string cmd, as if by the open-command function. They ensure that the stream is closed when they terminate.
The command-get-json function calls get-json on the stream, and returns the value returned by that function.
Similarly, command-get-jsons function calls get-jsons on the stream, and returns the value returned by that function.
(command-put-json cmd obj [flat-p [mode-opts]])
(command-put-jsons cmd seq [flat-p [mode-opts]])
The command-put-json and command-put-jsons functions open an output text stream over an output command pipe created for the command specified in the string argument cmd, using the function open-command function, write the argument object into the stream, in their specific manner, and then close the stream.
The command-put-json function writes a JSON representation of obj using the put-json function. The flat-p argument is passed to that function, defaulting to nil. The value returned is that of put-json.
The command-put-jsons function writes zero or more JSON representations of objects from seq, which must be an iterable object, using the put-jsons function. The flat-p argument is passed to that function, defaulting to nil. The value returned is that of put-jsons.
The *print-json-format* variable controls the formatting style exhibited by put-json and related functions. The initial value of this variable is nil.
If the value is the keyword symbol :standard, then a widely-used format is used, in which the opening and closing braces and brackets of vectors and dictionaries are printed on separate lines, as are the elements of those objects.
If the variable has any other value, including the initial value nil, then a default format is used in which braces, brackets and elements appear on the same line, subject to automatic breaking and indentation, similar to the way Lisp nested list structure is printed.
The *print-json-type* variable, whose initial value is t, controls whether the "__type" field is included when a structure object is printed as JSON.
This dynamic variable, initialized to a value of nil, controls whether the parser is tolerant to certain non-conformances in the syntax of JSON data, which are ordinarily syntax errors.
If the value of this variable is true, then the last element in a JSON array or the last element pair in a JSON object may be followed by spurious trailing comma, which is ignored.
Note: in the future, the variable may be extended to enable other instances of tolerance in the area of JSON parsing.
(get-json "{ 3:4, }") -> ;; syntax error
(let ((*read-bad-json* t))
(get-json "{ 3:4, }"))
--> #H(() (3.0 4.0))
This dynamic variable, initialized to a value of nil, controls whether the parser reads some JSON numbers as integer objects.
If the value of the variable is true, then whenever a JSON number is scanned which does not contain a . (decimal point) character or the letters e or E indicating an exponent field, it is converted to an integer object rather than a floating-point value. It is unspecified whether the number is converted to integer or floating-point if the exponent e or E is present, with a positive exponent value.
If this variable is nil, then JSON numbers are all converted to floating point.
On platforms where it is supported, TXR provides a feature called the foreign function interface, or FFI. This refers to the ability to interoperate with programming interfaces which are defined by the binary data type representations and calling conventions of the platform's principal C language compiler.
TXR's FFI module provides a succinct Lisp-based type notation for expressing C data types, together with memory-management semantics pertinent to the transfer of data between software components. The notation is used to describe the arguments and return values of functions in external libraries, and of Lisp callback functions that can be called from those libraries. Driven by the compiled representation of the type notation, the FFI module performs transparent conversions between Lisp data types and C data types, and automatically manages memory around foreign calls and incoming callbacks, for many common interfacing conventions.
The FFI module consists of a library of functions which provide all of its semantics. On top of these functions, the FFI module provides a number of macros which comprise an expressive, convenient language for defining foreign interfaces.
The FFI module supports passing and returning both structures and arrays by value. Passing arrays by value isn't a feature of the C language syntax; from the C point of view, these by-value array objects in the TXR FFI type system are equivalent to C arrays encapsulated in structs.
A carray type is provided for situations when foreign code generates arrays of undeclared, dynamic length, other than strings, and returns these arrays by the usual convention of pointer to the first element. The handling of carray requires more responsibility from the application.
The FFI feature is inherently unsafe. If the FFI type language is used to write incorrect type definitions which do not match the actual binary interface of a foreign function, undefined behavior results. Improper use of FFI can corrupt memory, creating instability and security problems. It can also cause memory leaks and/or use-after-free errors due to inappropriate deallocation of memory.
The implicit memory management behaviors encoded in the FFI type system are convenient, but risky. A minor declarative detail such as writing str instead of str-d in the middle of some nested type can make the difference between correct code and code which causes a memory leak, or instability by freeing memory which is in use.
FFI developers are encouraged to unit test their FFI definitions carefully and use tools such as Valgrind to detect memory misuses and leaks.
When a function call takes place from the TXR Lisp arena into a foreign library function, argument values must be prepared into the foreign representation. This takes place by converting Lisp objects into stack-allocated temporary buffers representing C objects. For aggregate objects containing pointers, additional buffers are allocated dynamically. For instance, suppose a structure contains a string and is passed by value. The structure will be converted to a stack-allocated equivalent C structure, in which the string will appear as a pointer. That pointer may use dynamically allocated (via malloc) string data. The operation which prepares argument material before a foreign function call is the put operation. In FFI callback dispatch, the operation which propagates the callback return value to the foreign caller is also the put operation.
After a foreign function call returns from a foreign library back to the TXR Lisp arena, the arguments have to be examined one more time, because two-way communication is possible, and because some of the material has temporary, dynamically allocated buffers associated with it which must be released. For instance a structure passed by pointer may be updated by the foreign function. FFI needs to propagate the changes which the foreign function performed to the C version of the structure, back to the original Lisp structure. Furthermore, a structure passed by pointer uses a dynamically allocated buffer. This buffer must be freed. The operation which handles the responsibility for propagating argument data back into TXR Lisp objects, and frees any temporary memory that had been arranged by the put operation is the in operation.
The in operation has two nuances: the by-value nuance and the by-pointer nuance. Data passed into a function by value such as function arguments or via ptr-in are subject to the by-value nuance. Updates to the foreign representation of these objects does not propagate back to the Lisp representation; however, those objects may contain pointers requiring the by-pointer nuance of the in operation of those pointers to be invoked.
After a foreign call completes, it is also necessary to retrieve the call's return value, convert it to a Lisp object, and free any dynamic memory. This is performed by the get operation.
The get operation is also used by a Lisp callback function, called from a foreign library, to convert the arguments to Lisp objects.
When a Lisp callback invoked by a foreign library completes, it must provide a return value, and also update any argument objects with new values. The return value is propagated using the put operation. Updates to arguments are performed by the out operation. This operation is like the reverse of the in operation. Like that operation, it has a by-value and by-pointer nuance.
For instance, if a callback receives a structure by value, upon return, there is no use in reconstructing a new version of the structure from the updated Lisp structure; the caller will not receive the change. However, if the structure contains pointers to data that was updated, by the callback, those changes must materialize. This is achieved by triggering the by-value nuance of the structure type's out operation, which will recursively invoke the out operation of embedded pointers, which will in turn invoke the by-pointer nuance.
The FFI type system consists of a notation built using Lisp syntax. Basic, unparametrized types are denoted by symbolic atoms. Similarly to a concept in the C language, typedef names can be globally defined, using the ffi-typedef function, or the typedef macro.
Like in the C language, typedef names are aliases for an existing type, and not distinct types. However, this is of no consequence, since the FFI doesn't perform any type checking between two foreign types, and thus never takes into consideration whether two such types are equal. The main concern in FFI is correspondence between Lisp values and foreign types. For instance, a Lisp string argument will not convert to a foreign function parameter of type int.
Compound expressions denote the construction of derived types, or types which are instantiated with parameters. Each such expression has a type constructor symbol in the operator position, from a limited, fixed vocabulary, which cannot be extended.
Some constituents of compound type syntax are expressions which evaluate to integer values: the dimension value for array types, the size for buffers, the width for bitfields and the value expressions for enumeration constants are such expressions. These expressions allow full use of TXR Lisp. They are evaluated without visibility into any apparent surrounding lexical scope.
Some predefined types which are provided are in fact typedef names. For instance, the size-t type is a typedef name for some other integral type, defined in a platform-specific way. Which type that is may be determined by passing the syntax to the type compiler function using the expression (ffi-type-compile 'size-t). The type compiler converts the size-t syntax to the compiled type object, resolving the typedef name to the type which it denotes. The printed representation of that object reveals the identity of the type. For instance, it might be #<ffi-type uint>, indicating that size-t is an alias for the uint basic type, which corresponds to the C type unsigned int.
If these types are used for representing individual scalar values, there is no difference among char, zchar and bchar.
What is different among these three types is that the array and zarray type constructors treat them specially. Arrays of these types are subject to conversion to and from Lisp strings. The variation among these types expresses different conversion semantics. That is to say, an array of bchar converts between the foreign and native Lisp representation differently from an array of zchar, which in turn converts differently from an array of char.
Note: it is recommended to avoid using the types bchar and zchar other than for expressing the element type of an array or zarray.
Note: this is utterly dangerous. Lisp values that aren't pointers must not be dereferenced by foreign code. Foreign code must not generate Lisp pointer values that aren't objects which came from a Lisp heap. Interpreting a Lisp value in foreign code requires a correct decoding of its type tag, and, if necessary, stripping the tag bits to recover a heap pointer and interpreting the type code stored in the heap object.
The conversion from foreign bit pattern to Lisp value is subject to a validity checks; an exception will be thrown if the bit pattern isn't a valid Lisp object. Nevertheless, the checks has cases which report as false positives: admit some invalid objects may be admitted into the Lisp realm, possibly with catastrophic results.
The cptr type converts between a foreign pointer and a Lisp object of type cptr.
Lisp objects of type cptr are tagged with a symbolic tag, which may be nil.
The unparametrized cptr converts foreign pointers to cptr objects which are tagged with nil.
In the reverse direction, it converts cptr Lisp objects of type cptr to foreign pointer, without regard for their type tag.
There is a parametrized version of the cptr FFI type, which provides a measure of type safety.
Note: the cptr type, in the context of FFI, is particularly useful for representing C pointers that are used in C library interfaces as "opaque" handles. For instance a FFI binding for the C functions fopen and fclose may use the cptr to represent the FILE * type. That is to say, cptr can be specified as the return type for fopen, thereby capturing the stream handle in a cptr object when that function is invoked through FFI. Then, the captured cptr object can be passed as the argument of fclose to close the stream.
The related types bstr, bstr-d, bstr-s, wstr, wstr-d and wstr-s are also provided; these are described in the following sections.
The str type behaves as follows. The put operation allocates, using malloc, a buffer large enough to hold the UTF-8 encoded version of the Lisp string, encodes the string into that buffer, and then stores the char * pointer into the argument space. The in operation deallocates the buffer. If str is passed by pointer, the in operation also takes the current value of the char * pointer, which may have been replaced by a different pointer, and creates a new Lisp string by decoding UTF-8 from that buffer. The get operation retrieves the C pointer and duplicates a new string by decoding the UTF-8 contents. The type has no out operation: a string is not expected to be modified in-place.
The type str-d type differs in behavior from str as follows. Firstly, it has no in operation. Consequently, str-d doesn't deallocate the buffer that had been allocated by put. Under the get operation, the str-d type assumes that ownership over the C pointer has been granted, and after duplicating a new string from the decoded UTF-8 data in the C string, it deallocates that C string by invoking the C library function free on it.
Type type str- is similar to str-d; it also has no in-operation, and doesn't deallocate the buffer allocated in the put operation. Under the get operation, the str-s type does not assume ownership of memory, and therefore does not free the pointer received from the foreign function. The str-s type is intended for receiving strings via a pointer-to-pointer argument, in situations when the string must not be freed.
Like other types, the string types combine with the ptr type family. Because the ptr family has memory management semantics, as does the string family, it is important to understand the memory management implications of the combination of the two.
The derived pointer types (ptr str-d) and (ptr str) are effectively equivalent. They denote a string passed by pointer, with in-out semantics. The effect is that the string is dynamic in both directions. What that means is that the foreign function either must not free the pointer it was given, or else it must replace it with one which the caller can also free (or with a null pointer). The two are equivalent because str-d has no in operation, so its get operation is used instead; but that operation is similar to the in operation of the str type: both decode the string currently referenced by the char * pointer, and then pass that pointer to the C free function.
Receiving a string by pointer from a foreign function is achieved by treating the situation as a pointer to an array of one element. So that is to say, an argument like char **pstr can be treated as either (ptr-out (array 1 str-d)) if the foreign function passes ownership of the string, or else (ptr-out (array 1 str-s)) if the foreign function retains ownership of the string. In either case, the argument is a vector of one element, which will be updated to the returned string, or else nil if the function passes back a null pointer.
The type combination (ptr-in str-d) refers to a string pointer passed to a foreign function by pointer, whereby the foreign function will retain and free the pointer. The type combination (ptr-in str) passes the string pointer in the same way, but the foreign module mustn't use the pointer after returning. FFI will free the pointer that had been passed.
The b prefix in the naming denotes "byte". It indicates that unlike the str family, the bstr family does not use UTF-8 encoding; only Lisp strings which contain strictly code points in the range U+0000 to U+00FF may convert to these types; out-of-range characters trigger an error exception.
Likewise, in the reverse direction, no UTF-8 decoding is performed: every byte value turns into the corresponding character code. The byte 0 is interpreted as the string terminator.
Note: the bstr type may be advantageous in situations when character handling is known to be confined to the ASCII range, since UTF-8 conversion is then unnecessary overhead. Because TXR strings use wide characters internally, converting to and from the bstr type still requires memory management overhead, just like in the case of the str type. The wstr type described in the next section avoids memory management and conversion overhead. Thus, even in situations in which characters are confined to the ASCII range, if wide functions are available in the foreign API, it may be more efficient to use them, particularly if the foreign component uses that representation internally.
Note: because wide characters do not require UTF-8 conversion, the wstr family is more efficient. A wstr string passes into foreign code directly: the Lisp object already contains a null-terminated wide character string, and so the pointer to that is given directly to the foreign code. Similarly, ownership transfer in either direction is a pointer passage with no memory management or conversion overheads. Whenever some foreign API offers a choice between UTF-8 strings, and wide strings, the wide version should be targeted by FFI, particularly if the API is known to works with wide strings internally also.
Under the buf type's put operation, no memory allocation takes place. The pointer to the buffer object's data is written into the argument space, so the foreign function can manipulate the buffer directly. If the object isn't a buffer but rather the symbol nil, then a null pointer is written.
The buf in operation has semantics as follows. In the pass-by-pointer nuance, the buffer pointer currently in the argument space is compared to the original one which had been written there from the buffer object. If they are identical, then the in operation yields the original buffer object. Otherwise, if the altered pointer is non-null, it allocates a new buffer equal in size to the original one and copies in the new data from the new pointer that was placed into the argument space by the foreign function. If the altered pointer is null, then instead of allocating a new buffer, the object nil is returned. The by-value nuance of the in operation does nothing.
The get operation is not meaningful for an unsized buf: it yields a zero length buf object. For this reason, parametrized buf type should be used for retrieving a buffer with a specific fixed size.
The buf-d type has different memory management from buf. The put operation of buf-d allocates a copy of the buffer and writes into the argument space a pointer to the copy. It is assumed that the foreign function takes ownership of the copy.
The in operation of buf-d is also different. The by-value nuance of the in operation is a no-op, like that of buf. The by-pointer nuance doesn't attempt to compare the previously written pointer to the current value. Rather, it assumes that if there is any non-null pointer value in the argument space, then it should take ownership of that object and return it as a new buffer. Thus if two-way dynamic buffer passing is requested using (buf buf-d) it means that the foreign function must replace the pointer with a null to indicate that it has consumed the buffer. Any non-null value in the argument space indicates that the foreign function has either rejected the pointer (not taken ownership), or has replaced it with a new object, whose ownership is being passed.
Unidirectional by-pointer passing of a buf-d can be performed using the types (ptr-out buf-d) or (ptr-int buf-d). The former type will not invoke buf-d's put operation. It will only allocate a pointer-sized space, without initializing it. After the foreign call, the by-pointer semantics of the in operation will be triggered If the foreign function places a non-null pointer into the space, its ownership will be seized by a newly instantiated buffer object. Otherwise the function must place a null pointer, which results in a nil value emerging from the in operation as documented above. The latter type will achieve a transfer of ownership in the other direction, by invoking the buf-d put operation, which places a copy of the buffer into the pointer-sized location prepared in the argument space. After the call, it will invoke the by-value in semantics of buf-d, which is a no-op: thus no attempt is made to extract a buffer, even if the foreign function alters the pointer.
When the nil symbol is converted to a closure type, it becomes a null function pointer.
A cptr object of any kind converts to a closure; the internal pointer is converted to a function pointer.
Instances of the ffi-closure type are produced by the ffi-make-closure function, or by calls to functions defined by the deffi-cb macro. The closure type is useful for passing callbacks to foreign functions: Lisp functions which appear to be C functions to foreign code.
In the reverse direction, when a closure object is converted from the foreign function pointer representation to a Lisp object, it becomes a cptr object whose tag is the closure symbol.
The following following parametrized type operators are available.
(enum name {(sym value) | sym}*)
The type enum specifies an enumerated type, which establishes a correspondence between a set of Lisp symbols and foreign integer values of type int.
The name argument must either be nil or a symbol for which the bindable function returns true. It gives the tag name of the enumerated type. The remaining arguments specify the enumeration constants.
In the enumeration constant syntax, each occurrence of sym They must be a bindable symbol according to the bindable function. The symbols may not repeat within the same enumerated type. Unlike in the C language, different enumerations may use the same symbols; they are in separate spaces.
If a sym is given, it is associated with an integer value which is one greater than the integer value associated with the previous symbol. If there is no previous symbol, then the value is zero. If the previous symbol has been assigned the highest possible value of the FFI int type, then an error exception is thrown.
If (sym value) is given, then sym is given the specified value. The value is an expression which must evaluate to an integer value in range of the FFI int type. It is evaluated in an environment in which the previous symbols from the same enumeration appear as variables whose bindings are their enumeration values, making it possible to use earlier enumerations in the definition of later enumerations.
The FFI enum type converts two kinds of Lisp values to the foreign type int: symbols which are in the set defined by the type, and integer values which are in the range which that foreign type can represent. Out-of-range integer values, symbols not defined in the enumeration, and objects not of symbol or integer type all trigger an exception.
In the reverse direction, the enum type extracts from the foreign representation values of FFI type int, and converts them, if possible, to symbols. If an integer value occurs which is not assigned to any enumeration symbol, then the conversion produces that integer value itself rather than a symbol. If an integer value occurs which is assigned to multiple enumeration symbols, it is not specified which of those symbols is produced.
(enumed type name {(sym value) | sym}*)
The enumed type operator is a generalization of enum which allows the base integer type of the enumeration to be specified. The following equivalence holds:
(enum n a b c ...) <--> (enumed int n a b c ...)
Any integer type or typedef name may be specified for type, including any one of the endian types. The enumeration inherits its size, alignment and other foreign representation details from type.
The values associated with the enumeration symbols must be in the representation range of type, which is not checked until the conversion of a symbol through the enumeration is attempted at run time.
The enumed type is a clone of the underlying type, inheriting most of its properties. In particular, it is possible to derive an enumed type from an underlying bitfield type. The resulting type is still a bitfield, and may only be used as a struct or union member. Moreover, because it is a bitfield type, there is a restriction against creating aliases for it with typedef.
An enumed bitfield allows the values of a bit field to be specified symbolically.
(struct name {(slot type [init-form])}*)
The FFI struct type maps between a Lisp struct and a C struct. The name argument of the syntax gives the structure type's name, known as the tag. If this argument is the symbol nil then the structure type is named by a newly generated uninterned symbol (with gensym).
The name is entered into a global namespace of tags which is shared by structures and unions.
The name also specifies the Lisp struct name associated with the FFI type.
The slot and type pairs specify the structure members. The slot elements must be symbols, and the type elements must be FFI type expressions.
A struct definition with no members refers to a previously defined struct or union type which has the same name in the global struct/union tag space.
If no prior struct or union exists, then a definition with no slots specifies a new, incomplete structure type. A struct definition with no members never causes a Lisp structure type to be created.
A struct definition that specifies one or more members either defines a new structure type, or completes an existing one. If an incomplete structure or union type which has the same name exists, then the newly appearing definition is understood to provide a completion of that type. If the incomplete type is a union, it thereby converted to a struct type.
If a complete structure type which has the same name already exists, then the newly appearing definition replaces that type in the tag namespace.
A struct definition with members is entered into the struct/union tag space immediately as an incomplete type (if it isn't already), before the members are processed. Therefore, the member definitions can refer to the struct type. The type becomes complete when the last member is processed, except in the special situation when that member causes the type to become a flexible structure, described several paragraphs below.
A struct definition that specifies members causes a Lisp struct having the same name to exist, if such a type doesn't already exist. If such a type is created, instance slots are defined for it which correspond to the member definitions in the FFI struct definition.
For any slot which specifies an init-form expression, that expression is evaluated during the processing of the type syntax, in the global environment. The resulting value then becomes the initial value for the slot. The semantics of this value is similar to that of a quoted object appearing as an init-form in the defstruct macro's slot-specifier syntax. For example, if the type expression (struct s (a int expr)), which specifies a slot a initialized by expr, generates a Lisp struct type, the manner in which that type is generated will resemble that of (defstruct s nil (a (quote [value-of-expr]))) where [value-of-expr] denotes the substitution of the value of expr which had been obtained by evaluation in the global environment. Note: if more flexible initialization semantics is required, the application must define the Lisp struct type first with the desired characteristics, before processing the FFI struct type. The FFI struct type will then related to the existing Lisp struct type.
Those members whose slot name is specified as nil is ignored; no instance slots are created in the Lisp type. If a init-form is specified for such a slot, there exists is no situation in which that form will be evaluated.
When a Lisp object is converted to a struct, it must, firstly, be of the struct type specified by name. Secondly, that type must have all of the slots defined in the FFI type. The slots are pulled from the Lisp structure in the order that they appear in the FFI struct definition. They are placed into the target memory area in that order, with all required padding between the members, and possibly after the last member, for alignment.
Whenever a member is defined using nil as the slot name, that member represents anonymous padding. The corresponding type expression is used only to determine the size of the padding only. Its data transfer semantics is completely suppressed. When converting from Lisp, the anonymous padding member simply generates a skip of the number of byte corresponding to the size of its type, plus any necessary additional padding for the alignment of the subsequent member.
Structure members may be bitfields, which are described using the ubit, sbit and bit compound type operators.
A structure member must not be an incomplete or zero-sized array, unless it is the last member. If the last member of FFI structure is an incomplete array, then it is a flexible structure.
A structure member must not be a flexible structure, unless it is the last member; the containing structure is then itself a flexible structure.
Flexible structures correspond to the C concept of a "flexible array member": the idea that the last member of a structure may be an array of unknown size, which allows for variable-length data at the end of a structure, provided that the memory is suitably allocated.
Flexible structures are subject to special restrictions and requirements. See the section Flexible Structures below. In particular, flexible structures may not be passed or returned by value.
See also: the make-zstruct function and the znew macro.
(union name {(slot type)}*)
The FFI union type resembles the struct type syntactically. It provides handling for foreign objects of C union type.
The name argument specifies the name for the union type, known as a tag. If this argument is the symbol nil then the union type is named by a newly generated uninterned symbol (with gensym).
The name is entered into a global namespace of tags which is shared by structures and unions.
The slot and type pairs specify the union members. The slot elements must be symbols, and the type elements must be FFI type expressions.
A union definition with no member refers to a previously defined struct or union type which has the same name in the global struct/union tag space.
If no prior struct or union exists, then a definition with no slots specifies a new, union type that is incomplete.
A union definition that specifies one or more members either defines a new structure type, or completes an existing one. If an incomplete structure type which has the same name exists, then the newly appearing definition is understood to provide a completion of that type. If the prior incomplete type is a struct, it is converted to union type. If a complete structure or union type which has the same name already exists, then the newly appearing definition replaces that type in the tag namespace.
A struct union definition with members is entered into the struct/union tag space immediately as an incomplete type (if it isn't already), before the members are processed. Therefore, the member definitions can refer to the union type. The type becomes complete when the last member is processed.
Unlike the FFI struct type, the union type doesn't provide automatic conversion between C and Lisp data. This is because the union is inherently unsafe, due to its placement of multiple types into the same storage, and lack of any information to discriminate which type is currently stored. Instead, the FFI union creates a correspondence between a C union that is regarded as just a region of memory, and a TXR Lisp data type called union.
An instance of the Lisp union type holds a copy of the C union memory, and also contains type information about the unions members. Functions are provided to store and retrieve the members; it is these functions which provide the conversion between the Lisp types and the foreign representations stored in the C union. This is done under control of the application, because due to the inherent lack of safety of the C union, only the application program knows which member of the union may be accessed.
Conversion between the C union and the Lisp union consists of just a memory copying operation.
The following functions are provided for manipulating unions: make-union instantiates a new union object; union-members retrieves a list of the symbols serving as the union's member names; union-get retrieves a specified member from the union's storage, converting it to a Lisp object; union-put places a Lisp object into a union, using the specified member's type to convert it to a foreign representation; union-in performs the "in semantics" on the specified member of a union, propagating modifications in that member back to a Lisp object; and union-out performs "out semantics" on the specified member of a union, propagating modifications done on a previously retrieved Lisp object back into the union.
(array dim type)
(array type)
The FFI array type creates a correspondence between Lisp sequences and "by value" fixed size arrays in C. It converts Lisp sequences to C arrays, and C arrays to Lisp vectors.
Arrays passed by values do not exist in the C language syntax. Rather, the C type which corresponds to the FFI array is a C array that is encapsulated in a struct. For instance the type (array 3 char) can be visualized as corresponding to the C type struct { char anonymous[3]; }.
Thus, in the FFI syntax, we can specify arrays as function parameters passed by value and as return values.
On conversion from Lisp to the foreign type, the FFI array simply iterates over the Lisp sequence, and performs an element for element conversion to type.
If the sequence is shorter than the array, then the remaining elements are filled with zero bits. If the sequence is longer than the array, then the excess elements in the sequence are ignored.
Since Lisp arrays and C arrays do not share the same representation, temporary buffers are automatically created and destroyed by FFI to manage the conversion.
The dim argument is an ordinary Lisp expression expanded and evaluated in the top-level environment. It must produce a nonnegative integer value.
In addition, several types are treated specially: when type is one of char, zchar, bchar or wchar, the array type establishes a special correspondence with Lisp strings. When the C array is decoded, a Lisp string is created or updated in place to reflect the new contents. This is described in detail below.
The second form, whose syntax omits the dim element, it denotes a variable length array. It corresponds to the concept of an incomplete array in the C language, except that no implicit array-to-pointer conversion concept is implemented in the FFI type system. This type may not be used as an array element or structure member, other than as the last structure member. It also may not be passed or returned by value, only by pointer. If the last member of a structure has this type, then it is a flexible array member; see the Flexible Structures section below.
Since the type has unknown length, it has a trivial get operation which returns nil. It is useful for passing a variable amount of data into a foreign function by pointer.
An array of char represents non-null-terminated UTF-8 character data, which converts to and from a Lisp string. Any null bytes in the data correspond to the pseudo-null character #\xDC00 also notated as #\pnul.
An array of zchar represents a field of optionally null-terminated UTF-8 character data. If a null byte occurs in the data then the text terminates before that null byte, otherwise the data comprises the entire foreign array. Thus, null bytes do not occur in the data. A null byte in the array will not generate a pseudo-null character in the Lisp string.
An array of bchar values represents 8-bit character data that isn't UTF-8 encoded, and is not null terminated. Each byte holds a character whose code is in the range 0 to 255. If a null byte occurs in the data, is interpreted as a string terminator.
(zarray dim type)
(zarray type)
The zarray type is a variant of array. When converting from Lisp to C, it ensures that the array is null-terminated. This means that if the zarray is dimensioned, then the [dim - 1] element of the C array is written out as all zero bytes, ignoring the corresponding Lisp value in the Lisp array. If the zarray is undimensioned, then the size of the C array is deemed to be one greater than the actual length of the Lisp array. The elements in the Lisp array are converted to the corresponding elements of the C array, and then the last element of the C array is filled with null bytes. The zarray type is useful for handling null terminated character arrays representing strings, and for null terminated vectors. Unlike array, zarray allows the Lisp object to be one element short. For instance, when a (zarray 5 int) passed by pointer a foreign function is converted back to Lisp, the Lisp object is required to have only four elements. If the Lisp object has five elements, then the fifth one will be decoded from the C array in earnest; it is not expected to be null. However, when that Lisp representation is converted back to C, that extra element will be ignored and output as a zero bytes.
Lastly, the zarray further extends the special treatment which the array type applies to the types zchar, char, wchar and bchar. The zarray type assumes, and depends on the incoming data being null-terminated, and converts it to a Lisp string accordingly. The regular array type doesn't assume null termination. In particular, this means that whereas (array 42 char) will decode 42 bytes of UTF-8, even if some of them are null, converting those null bytes to the U+DC00 pseudo-null, in contrast, a zarray will treat the 42 bytes as a null-terminated string, and decode UTF-8 only up to the first null. In the other direction, when converting from Lisp string to foreign array, zarray ensures null termination.
Note that the type combination zarray of zchar behaves in a manner indistinguishable from a zarray of char.
The one-argument variant of the zarray syntax which omits the dim argument specifies a null-terminated variant of the variable-length array. Like that type, it corresponds to the concept of an incomplete array in the C language. It may not be used as an array element, and may not be used as a structure member other than the last member. It cannot be passed as an argument or returned as a value. If the last member of a structure has this type, then it is a flexible array member; see the Flexible Structures section below.
Unlike the ordinary variable-length array, the zarray type supports the get operation, which extracts elements, accumulating them into a resulting vector, until it encounters an element consisting of all zero bytes. That element terminates the decoding, and isn't included in the resulting array.
The variable-length zarray also has a special in operation. Like the get operation, the in operation extracts all elements until a terminating null, decoding them to a vector. Then, the entire original vector is replaced with the new vector, even if the original vector is longer.
(ptr type)
The ptr denotes the passage of a value by pointer. The type argument gives the pointer's target type. The ptr type converts a single Lisp value, to and from the target type, using a C pointer as the external representation.
When used for passing a value to a foreign function, the ptr type has in-out semantics: it supports the interfacing concept that the called function can update the datum which has been passed to it "by pointer", thereby altering the caller's object. Since a Lisp value requires a conversion to the FFI external representation, it cannot be directly passed by pointer. Instead, this semantics is simulated. The put semantics of ptr allocates a temporary buffer, large enough to hold the representation of type. The Lisp value is then encoded into this buffer, recursively relying on the type's put semantics. After the foreign call, ptr triggers the in semantics of type to update the Lisp object from the temporary buffer, and releases the buffer.
The get semantics of ptr is used in retrieving a ptr return value, or, in a FFI callback, for retrieving the values of incoming arguments that are of ptr type. The get semantics assumes that the memory referenced by the C pointer is owned by foreign code. The Lisp object is merely decoded from the data area, which is then not touched.
The out semantics of ptr, used by callbacks for updating the values of arguments passed by pointer, assumes that the argument space already contains a valid pointer. The pointer is retrieved from the argument space, and the Lisp value is encoded into the memory referenced by that pointer.
Note that only Lisp objects with mutable slots can be meaningfully passed by pointer with in-out semantics. If a Lisp object without immutable slots, such as an integer, is passed using ptr the incoming updated value of the external representation will be ignored. Concretely, if a C function has the argument signature (int *) with in-out semantics such that it updates the int object which is passed in, this function can be called as a foreign function using a (ptr int) FFI type for the argument. However, the argument of the foreign call on the TXR Lisp side is just an integer value, and that cannot be updated.
On the other hand, if a FFI struct member is declared as of type (ptr int) then the Lisp struct is expected to have an integer-valued slot corresponding to that member. The slot is then subject to a bidirectional transfer. FFI will create an int-sized temporary data area, encode the slot into that area and place that area's pointer into the encoded structure. After the call, the new value of the int will be extracted from the temporary buffer, which will then be released. The Lisp structure's slot will be updated with the new integer. This will happen even if the Lisp structure is being passed as a by-value argument.
(ptr-in type)
ptr-in type is a variation of ptr which denotes the passing of a value by pointer into a function, but not out. The put semantics of ptr-in is the same as that of ptr, but after the completion of the foreign function call, the in semantics differs. The ptr-in type only frees the temporary buffer, without decoding from it.
The out semantics of ptr-in differs also. It effectively treats the object as if it were "by value", since the reverse data transfer is ruled out. In other words, ptr-in simply triggers the by-value nuance of type's out semantics.
The get semantics of ptr-in is the same as that of ptr.
(ptr-out type)
The ptr-out type is a variant of ptr which denotes a by pointer data transfer out of a function only, not into. The put semantics of ptr-out prepares a data area large enough to hold type and stores a pointer to that area into the argument space. The Lisp value isn't encoded into the data area.
The in semantics is the same as that of ptr: the by-pointer nuance of type's in semantics is invoked to decode the external representation to Lisp data.
(ptr-in-d type)
The ptr-in-d type is a variant of ptr-in which transfers ownership of the allocated buffer to the invoked function. That is to say, the in semantics of ptr-in-d doesn't involve the freeing of memory that was allocated by put semantics.
The ptr-in-d type is useful when a function expects a pointer to an object that was allocated by malloc and expects to take responsibility for freeing that object.
Since the function may free the object even before returning, the pointer must not be used once the function is called. This is ensured by the in semantics of ptr-in-d which is the same as that of ptr-in.
The ptr-in-d type also has get semantics which assumes that ownership of the C object is to be seized. FFI will automatically free the C object when get semantics is invoked to retrieve a value through a ptr-in-d.
(ptr-out-d type)
The ptr-out-d type is a variant of ptr-out which is useful for capturing return values or, in a callback producing return values.
The ptr-out-d type has empty put semantics. If it put semantics is invoked, it does nothing: no area is allocated for type and no pointer is stored into the argument space.
The in semantics is the same as that of ptr: a pointer is retrieved from the argument space, the object is subject to type's in semantics to recover the updated Lisp value, and then the object is freed.
The get semantics of ptr-out-d is identical to that of ptr-in-d.
The out semantics is identical to that of ptr.
(ptr-out-s type)
The ptr-out-d type is a variant of ptr-out similar to ptr-out-d, which assumes that the C object being received has an indefinite lifetime, and doesn't need to be freed. The suffix stands for "static".
Like ptr-out-d, the ptr-out-s has no put semantics.
Its in semantics recovers a Lisp value from the external object whose pointer has been stored by the foreign function, but doesn't free the external object.
The get semantics retrieves a Lisp value without freeing.
(bool type)
The parametrized type bool can be derived from any integer or floating-point type. There is also an unparametrized bool which is a typedef for the type (bool uchar).
The bool type family represents Boolean values, converting between a Lisp Boolean and foreign Boolean. A given instance of the bool type inherits all of its characteristics from type, such as its size, alignment and foreign representation. It alters the get and put semantics, however. The get semantics converts a foreign zero value of type to the Lisp symbol nil, and all other values to the symbol t. The put semantics converts the Lisp symbol nil to a foreign value of zero. Any other Lisp object converts to the foreign value one.
The bool types are not integers, and cannot be used as the basis of bitfields: syntax like (bit 3 (bool uint)) is not permitted. However, Boolean bitfields are possible when this syntax is turned inside out: the bool type can be derived from a bitfield type, as exemplified by (bool (bit 3 uint)). This simply applies the above described Boolean conversion semantics to a three-bit field. A zero/nonzero value of the field converts to nil/t and a nil or non-nil Lisp value converts to a 0 or 1 field value.
({ubit | sbit} width)
The ubit and sbit types denote C-language-style bitfields. These types can only appear as members of structures. A bitfield type cannot be the argument or return value of a foreign function or closure, and cannot be a foreign variable. Arrays of bitfields and pointers, of any kind, to bitfields are a forbidden type combination that is rejected by the type system.
The ubit type denotes a bitfield of type uint, corresponding to an unsigned bitfield in the C language.
The sbit type denotes a bitfield of type int. Unlike in the C language, it is not implementation-defined whether such a bitfield represents signed values; it converts between Lisp integers that may be positive or negative, and a foreign representation which is two's complement.
Bitfields based on some other types are supported using the more general bit operator, which is described below.
The width parameter of is an expression evaluated in the top-level environment, indicates the number of bits. It may range from zero to the number of bits in the uint type.
In a structure, bitfields produced by sbit and ubit are allocated out in storage units which have the same width and alignment requirements as a uint. These storage units themselves can be regarded as anonymous members of the structure. When a new unit needs to be allocated in a structure to hold bitfields, it is allocated in the same manner as a named member of type uint would be at the same position.
A zero-length bitfield is permitted. It may be given a name, but the field will not perform any conversions to and from the corresponding slot in the Lisp structure. Note that in situations when the FFI struct definition causes the corresponding Lisp structure type to come into existence, the Lisp structure type will have slots for all the zero width named bitfields, even though those slots don't participate in any conversions in conjunctionwith the FFI type.
The presence of a zero-length bitfield ensures that a subsequent structure member, whether bitfield or not, is placed in a new storage unit of the size of the bitfield's base type.
Details about the algorithm by which bitfields are allocated within a structure are given in the paragraph below entitled Bitfield Allocation Rules.
A ubit field stores values which follow a pure binary enumeration. For instance, a bitfield of width 4 stores values from 0 to 15. On conversion from the Lisp structure to the foreign structure, the corresponding member must be a integer value in this range, or an error exception is thrown.
On conversion from the foreign representation to Lisp, the integer corresponding to the bit pattern is recovered. Bitfields follow the bit order of the underlying storage word. That is to say, the most significant binary digit of the bitfield is the one which is closest to the most significant bit of the underlying storage unit. If a four-bit field is placed into an empty storage unit and the value 8 its stored, then on a big-endian machine, this has the effect of setting to 1 the most significant bit of the underlying storage word. On a little-endian machine, it has the effect of setting bit 3 of the word (where bit 0 is the least significant bit).
The sbit field creates a correspondence between a range of Lisp integers, and a foreign representation based on the two's complement system. The most significant bit of the bitfield functions as a sign bit. Values whose most significant bit is clear are positive, and use a pure binary representation just like their ubit counterparts. The representation of negative values is defined by the "two's complement" operation, which maps each value to its additive inverse. The operation consists of temporarily treating the entire bitfield as unsigned, and inverting the logical value of all the bits, and then adding 1 with "wraparound" to zero if 1 is added to a field consisting of all 1 bits. (Thus zero maps to zero, as expected.) An anomaly in the two's complement system is that the most negative value has no positive counterpart. The two's complement operation on the most negative value produces that same value itself.
A sbit field of width 1 can only store two values: -1 and 0, represented by the bit patterns 1 and 0. An attempt to convert any other integer value to a sbit field of width 1 results in an error.
A sbit field of width 2 can represent the values -2, -1, 0 and 1, which are stored as the bit patterns 10, 11, 00 and 01, respectively.
(bit width type)
The bit operator is more general than ubit and sbit. It allows for bitfields based on on any integer type up to 64 bits wide.
When the character types char and uchar are used as the basis of bitfields, they convert integer values, not characters. In the case of char, the bitfield is signed.
All remarks about ubit and sbit apply to bit also.
Details about the algorithm by which bitfields are allocated within a structure are given in the paragraph below entitled Bitfield Allocation Rules.
Under the bit operator, the endian types such as be-int32 or le-int16 may also be used as the basis for bitfields. If type is an endian type, the bitfield is then allocated in the same way that a bitfield of the corresponding ordinary type would be allocated on a target machine which has the byte order of that endian type.
When a bitfield member follows a member which has a different byte order, the bitfield is placed into a new allocation cell. This is true even if the previous member has the same alignment.
Note: the allocation of bits within a bitfield based on a byte storage cells also differs between different endian systems. However, the FFI type system does not offer one byte endian types such as be-uint8. The workaround is to switch to a wider type.
Note: endian bitfields may be used to match the image of a C structure which contains bitfields, without having to conditionally define the FFI struct type differently based on whether the current machine is big or little endian. Conditionally defining a structure for two different byte orders adds verbiage to the program and is highly error-prone, since the bitfields change order within an allocation unit.
For instance, on a big endian system, the definition of a structure representing an IPv4 packet might begin like this:
(struct ipv4-header
(ver (bit 4 uint16))
(ihl (bit 4 uint16))
(dscp (bit 6 uint16))
(ecn (bit 2 uint16))
(len uint16)
...)
to port this to a little endian system, the programmer has to recognize that the first pair of fields is packed into one byte, and the next pair of fields into a second byte. The bytes stay in the same order, but the pairs are reversed:
(struct ipv4-header
(ihl (bit 4 uint16)) ;; reversed pair
(ver (bit 4 uint16))
(ecn (bit 2 uint16)) ;; reversed pair
(dscp (bit 6 uint16))
(len be-uint16)
...)
Endian bitfields allow this to be defined naturally. The IPv4 header is based on network byte order, which is big-endian, so big endian types are used. The little endian version above already uses be-uint16 for the len field. This just has to be done for the bitfields also:
(struct ipv4-header
(ver (bit 4 be-uint16))
(ihl (bit 4 be-uint16))
(dscp (bit 6 be-uint16))
(ecn (bit 2 be-uint16))
(len be-uint16)
...)
({buf | buf-d} size)
The parametrized buf and buf-d types are variants of the unparametrized buf and buf-d, respectively. The size argument is an expression which is evaluated in the top-level environment, and must produce a nonnegative integer.
Because they have a size, these types have useful get semantics.
The get semantics of buf-d is that a Lisp object of type buf is created which takes direct ownership of the memory.
The get semantics of buf is that a Lisp object is created using a dynamically allocated copy of the memory.
(carray type)
The carray type corresponds to a C pointer, in connection with the concept of representing a variable length array that is passed and returned as a pointer to the base element. On the Lisp side, the carray FFI type corresponds to the carray Lisp type. The carray Lisp type is similar to cptr, but supports array indexing operations, and some other features. It can be regarded as a semantic cross between cptr and buf.
The get semantics of carray is simply that a pointer is retrieved from memory and converted to a freshly allocated carray object which holds that pointer, and is marked as having an unknown size. No copy is made of the underlying array. When the application determines the size of the array, it can inform that object by calling the carray-set-length function.
The put semantics of the carray FFI type is simply to write, into the argument space, the pointer which the object holds. The object must be a carray whose element type matches that of the FFI type.
The carray type has in semantics. When a carray is passed to a foreign function as an argument to a ptr or ptr-out parameter to either a carray or cptr type, what is passed to the function is a pointer to the carray's pointer. The foreign function may update this pointer to a new value, and this value is stored back into the carray object. The array's length is reset to zero. If it is an owned carray, arranged by carray-own, then the current array freed before the new pointer is assigned, and the object's type is reset to borrowed array. The carray object must not be memory mapped carray coming from the mmap function.
The carray type lacks out semantics, since Lisp code cannot change its address; so there is no new pointer to propagate back to a foreign caller which passes a carray to a Lisp callback, and no other memory management tasks to perform.
The carray type is particularly useful in situations when foreign code generates such an array, and the size of that array isn't known from the object itself.
It is also useful, instead of a variable-length zarray, for passing a dynamic array to foreign code in situations when the application benefits from managing the memory for the array. The variable-length zarray FFI type's disadvantage relative to carray is that the zarray converts an entire Lisp sequence to a temporarily allocated array, which is used only for one call. By contrast, the carray object holds the C representation which Lisp code can manipulate; and that representation is passed directly, just like in the case of buf.
Unlike buf, there is no dynamic variant of carray. The transfer of ownership of a carray requires the use of explicit operations like carray-free and carray-own.
It is possible to create a carray view over a buffer, using carray-buf.
Lastly, the carray type is the basis for the TXR Lisp mmap function, which is documented in the section Unix Memory Mapping.
(cptr type-sym)
The parametrized cptr type is similar to the unparametrized cptr. It also converts between Lisp objects of type cptr and foreign pointers. Unlike the unparametrized type, it provides a measure of type safety, and also supports the conversion of carray objects.
When a foreign pointer is converted to a Lisp object under control of the parametrized cptr, the resulting Lisp cptr object is tagged with the type-sym symbol.
In the reverse direction, when a Lisp cptr object is converted to the parametrized type, its type tag must match type-sym, or else the conversion fails with an error exception. This rule contains a slight relaxation: a cptr object with a nil tag can be converted to a foreign representation using any parametrized type, if its value is null. In other situations, the cptr-cast function must be used to coerce the pointer object to the matching type.
Note that if type-sym is specified as nil, then this is precisely equivalent to the unparametrized cptr which doesn't provide the above safety measure.
A carray object may also be converted to a foreign pointer under the control of a parametrized cptr type. The carray object's internal pointer becomes the foreign pointer value. The conversion is only permitted if the following two restrictions are not met, otherwise an error exception is thrown. Firstly, the type-sym of the cptr type must be the name of an FFI type, at the time when the cptr type expression is processed, otherwise the cptr is not associated with a type. Secondly, the carray object being converted must have an element type which matches the FFI type denoted by the cptr object's type-sym.
Pointer type safety is useful, because FFI can be used to create bindings to large application programming interfaces (APIs) in which objects of many different kinds are referenced using pointer handles. The erroneous situation can occur that a FFI call passes a handle of one kind to a function expecting a different kind of handle. If all pointer handles are represented by a single cptr type, then such a situation proceeds without diagnosis. If handles of different types are all mapped to cptr types with different tags, the situation is intercepted and diagnosed with an error exception.
(align [width] type)
(pack [width] type)
The FFI type operators align and pack define a type which is a copy of type, but with adjusted alignment requirements. In some cases, pack (but not align) works by replacing itself with a transformed version of the type syntax.
If the width argument is present, it is an expression which is evaluated in the top-level environment. It must produce a positive integer which is a power of two.
If width is absent, a different default value is used depending on which type operator is specified. For align, it defaults to some platform-specific maximum useful alignment value, typically 16. For pack, a missing width defaults to 1.
The align operator can be used to create a version of type which is aligned at least as strictly as the specified width. That is to say, values of width which are less than or equal to type's existing alignment have no effect on alignment, except when the type is used as a bitfield.
The pack operator can be used to create a version of type which is less strictly aligned than its existing alignment.
Alignment affects the placement of the type as a structure member, and as an array element.
A type with alignment 1, like the default alignment for pack, can be placed at any byte offset, and thus is effectively unaligned. A type with alignment 2 can be placed only at even addresses and offsets.
Alignment can be applied to all types, including arrays and structs. It may also be applied to bitfields, but special considerations have to be observed to obtain the intended effect, described below. However, out of the elementary types, only the integer and floating point types are required to support a weakening of alignment. Whether a type which corresponds to a pointer, such as a str or buf, can be written at an offset which doesn't meet that type's default alignment is machine-dependent.
If a FFI struct type is declared with a weakened alignment, whether or not such a structure can be read or written at the misaligned offsets depends on whether the individual members support it. If they are integer or floating-point types, or aggregates thereof, the usage is supported in a machine-independent manner.
Alignment interacts with the allocation of bitfields in special ways. If width is greater than 1, or regardless of width if the operator is align, the type is marked with a Boolean attribute indicating that it has altered alignment. Then, when a bitfield is based on a type which has altered alignment, then that bitfield isn't packed together with the previous field, even if the allocation rules otherwise call for it. Due to the alignment request, the byte offset is first adjusted according to the requested alignment and the bit offset is reset to zero. The bit field is then allocated at the new alignment. This requirement applies even if the requested alignment is 1, which is possible via a combination of both pack and align, both specified with a width of 1. If the requested alignment for the type of a bitfield is 1, and the previous member is a bitfield which has left a byte partially filled, then the new bitfield starts on a fresh byte, even if it would otherwise be packed with the previous bitfield. If a named bitfield has weakened alignment, other than one byte alignment produced by pack, the bitfield's original type's alignment is used for the purposes of determining its contribution to the alignment of the structure.
When type is one of two kinds of types, the pack type operator exhibits special behaviors, as follows. In these situations, the pack operator has no semantics other than these behaviors.
The rationale for this behavior is that alignment weakening is often required for all members of a structure, rather than select members. Moreover, specifying weak alignment for a structure type itself, while leaving members with strict alignments, rarely makes sense. Weakening the alignment of a structure will not eliminate the padding between the members or at the end; it will only have any useful effect when that structure is itself used as the member of another structure. An important rationale also is that the GNU C packed attribute works this way, and so C structures declarations using that attribute are easier to translate to the TXR Lisp FFI type system.
Deriving a less strictly aligned version of a structure or union type without any effect on the alignment of its members may be obtained by applying the bit operator to either typedef name for a structure or union type, or else to syntax which refers to an existing type without defining members. Given the definition (typedef s (struct s (a int) (b char))), the type s is an eight byte structure with three bytes of padding at the end, which has four byte alignment. The type expression (pack s) produces a version of this type which has one byte alignment. The expression (pack (struct s)), likewise. The resulting unaligned type is still eight bytes wide, and has three padding bytes. In other words, the pack operator does not transform the syntax of a structure which is already defined as an object,
The rationale for this transformation is that when both align and pack are applied to a type, the combination only makes sense when pack is first. For a non-structure type like int, (pack x (align y int)) is equivalent to just (pack x int), because pack will set the alignment to x regardless of the effect of align. Whereas (align y (pack x int)) is meaningful in that the align takes precedence over pack if (> y x). The main rationale is that pack may be applied to structure members via a code transformation. Those members may already have types which use align. This transformation ensures that the semantics is applied in a useful order. For example (pack (struct s (a char) (x (align 2 int)))) is first transformed into (struct s (a (pack 1 char)) (x (pack 1 (align 2 int))))). If this is left as-is, then the align on x is obliterated by the pack, rendering it useless. A further transformation takes place to (struct s (a (pack 1 char)) (x (align 2 (pack 1 int))))). Now the align directive is increasing the alignment of x to 2, so that x will be placed at offset 2, leaving one byte of padding after the a member. This is how attributes work in GNU C also: the aligned attribute on the member of a packed structure can take precedence and increase its alignment.
After these transformations are applied, the nested pack forms which occur in the transformed syntax may perform more such transformations, depending on their operands.
Note that the two-argument form of pack with a width value greater than 1 doesn't directly correspond to any single attribute specifier in GNU C. The GNU C packed attribute is Boolean, implicitly reducing alignment to 1. A combination of the GNU C attributes aligned and packed is used to produce the effect of (pack n type) for values of n > 1. In GNU C, the packed attribute, when applied to a structure, distributes to its members, but isn't capable of distributing an alignment exceeding 1. So the (pack n (struct ...)) expression, for values of n > 1, doesn't correspond to anything in GNU C; its effect can be simulated by attributing the structure type with packed, and then individually applying the required alignment to the member declarations.
These additional FFI types for common C language types are provided as typedef aliases. The intmax-t and uintmax-t types are provided only if the host platform's intmax_t is no wider than 64 bits. If the host platform lacks intmax_t then the above two FFI types are defined as aliases for longlong and ulonglong, respectively.
(qref struct-type member1 [member2 ...])
The FFI type operator qref provides a way to reference the type of a member of a struct or union. The struct-type argument must be a type expression denoting a struct or union. The member1 argument and any additional arguments must be symbols.
If S is a struct or union type, and M is a member, then (qref S M) is a type expression denoting the type of M. Moreover, if M itself is a struct or union, which has a member named N then the type of N can be denoted by the expression (qref S M N). Similarly, additional symbols reference through additional struct/union nestings.
Note: the referencing dot syntax can be used to write qref expressions. For instance, (qref S M N) can be written as S.M.N instead.
(elemtype type)
The FFI type operator elemtype denotes the element type of type, which must be a pointer, array or enum.
Note: there is also a macro elemtype. The macro expression (elemtype X) is equivalent to the expression (ffi (elemtype X)).
In addition to the type system described in the previous section. the FFI type system supports endian types, which are useful for dealing with data formats defined by networking protocols and other kinds of standards, or data structure definitions from other machines.
There are two kinds of endianness: Little endian refers to the least-significant byte of a data type being stored at the lowest address in memory, lowest offset in a buffer, lowest offset in a file, or earlier byte in a communication stream. Big endian is the opposite: it refers to the most significant byte occurring at the lowest address, offset or stream position. For each of the signed integral types
int16 through int64, the corresponding unsigned types uint16 through uint64, and the two floating-point types float and double, the FFI type system provides a big-endian and little-endian version, whose names are derived by prefixing the be- or le- prefix to its related type.
Thus, the exhaustive list of the endian types is: be-int16, be-uint16, be-int32, be-uint32, be-int64, be-uint64, be-float, be-double, le-int16, le-uint16, le-int32, le-uint32, le-int64, le-uint64, le-float and le-double.
These types have the same size and alignment as their plain, unprefixed counterparts. Alignment can be overridden with the align type construction operator to create versions of these types with alternative alignment.
Endian types are supported as arguments to functions, return values, members of structs and elements of arrays.
TXR Lisp's FFI performs the automatic conversion from the abstract Lisp integer representation to the foreign representations exhibiting the specified endianness.
In the TXR Lisp FFI type system, the following types are incomplete: the type void, arrays of unspecified size, and any struct whose last element is of incomplete type.
An incomplete type cannot used as a function parameter type, or a return value type. It may not be used as an array element or union member type. A struct member type may be incomplete only if it is the last member.
An incomplete structure whose last member is an array is a flexible structure.
If a FFI struct type is defined with an incomplete array (an array of unspecified size) as its last member, then it specifies an incomplete type known as a flexible structure. That array is the terminating array. The terminating array corresponds to a slot in the Lisp structure; that slot is the last slot.
A structure which has a flexible structure as its last member is also, effectively, a flexible structure.
When a Lisp structure is being converted to the foreign representation under the control of a flexible structure FFI type, the number of elements in the terminating array is determined from the length of the object stored in the last slot of the Lisp structure. The length includes the terminating null element for zarray types. The conversion is consistent with the semantics of an incomplete array that is not a structure member.
In the reverse direction, when a foreign representation is being converted to a Lisp structure under the control of a flexible structure FFI type, the size of the array that is accessed and extracted is determined from the length of the object stored in the last slot, or, if the array type is a zarray from detecting null-termination of the foreign array. The conversion of the array itself is consistent with the semantics of an incomplete array that is not a structure member. Before the conversion takes place, all of the members of the structure prior to the terminating array, are extracted and converted to Lisp representations. The corresponding slots of the Lisp structure are updated. Then if the Lisp structure type has a length method, that method is invoked. The return value of the method is used to perform an adjustment on the object in the last slot. If the existing object in the last slot is a vector, its length is adjusted to the value returned by the method. If the existing object isn't a vector, then it is replaced by a new nil-filled vector, whose length is given by the return value of length. The conversion of the terminating array to Lisp representation the proceeds after this adjustment, using the adjusted last slot object.
The TXR Lisp FFI type system follows rules for bitfield allocation which were experimentally derived from the behavior of the GNU C compiler on several mainstream architectures.
The allocation algorithm can be imagined to walk through the structure from the first member to the last, maintaining a byte offset O which indicates how many whole bytes have been allocated to members so far, and a bit offset B which indicates, additionally, how many bits have been allocated in the byte which follows these O bytes, between 0 and 7.
When a non-bitfield member is placed, then there are two cases: either B is zero (only O bytes have been allocated, with no fractional byte) or else B is nonzero. In this latter case, B is reset to zero and O is incremented by one. In either case, O is adjusted up to the required alignment boundary for the new member. The member is placed, and O is incremented again by the size of that member.
When a bitfield member is placed, the algorithm considers the structure to be allocated in units of the base type of that bitfield member. For instance if the bitfield is derived from type uint16 then the structure's layout is considered to have been allocated in uint16 units. The algorithm examines the value of O and B to determine the first available unit in which at least one bit of unallocated space remains. Then, if the unit at that offset has enough space to hold the new bitfield, according to the bitfield's width, then the bitfield is placed into that unit. Otherwise, the bitfield is placed into the next available unit.
After a bitfield is placed, the values of O and B are adjusted so that O reflects the whole number of bytes which have been allocated to the structure so far, and B indicates the 0 to 7 additional bits of any bitfield material protruding past those whole bytes.
A zero-width bitfield is also considered with regard to the storage unit size indicated by its type. As in the case of the nonzero-width bitfield, the offset of the first available unit is found which has at least one bit of unallocated space. Then, if that unit is entirely empty, the zero-width bitfield has no effect. If that unit is partially filled, then O is adjusted to point to the next unit after that, and B is reset to zero. Note that according to this semantics, a zero-width bitfield can have an effect even if placed between non-bitfield members, or appears as the last member of a structure. Also, a structure containing only a zero-width bitfield has size zero.
If, after the placement of all structure members, B has a nonzero value, then the offset O is incremented by one to cover that byte.
As the last allocation step, the size of the structure is then padded up to a size which is a multiple of the alignment of the most strictly aligned member.
A named bitfield contributes to the alignment of the structure, according to its type, the same way as a non-bitfield member of the same type. An unnamed bitfield doesn't contribute alignment, or else may be regarded as having the weakest possible alignment, which is byte alignment. If all of the members of a structure are unnamed bitfield members of any type, it exhibits byte alignment.
The description isn't complete without a treatment of byte and bit order. Bitfield allocation follows an imaginary "bit endianness" whose direction follows the machine's byte order: most-significant bits are allocated first on big endian, least significant bits first on little endian.
If a one-bit-wide bitfield is allocated into a hitherto empty structure, it will be placed into the first byte of that structure, regardless of the machine's endianness, and regardless of the underlying storage unit size for that bitfield. Within that first byte, it will be placed into the most significant bit position on a big-endian machine (bit 7); and on a little-endian machine, it will be placed into the least significant bit position (bit 0). If another one-bit-wide is allocated, it is placed into bit 6 on big endian, and bit 1 on little endian.
More generally, whenever a bitfield is allocated for a big-endian machine, and the storage unit is determined into which that bitfield shall be placed, the most significant bits of that storage unit are filled first on a big-endian machine, whereas the least significant bits are filled first on a little-endian machine. From this it follows that on either type of machine, that field shall be placed at the lowest-addressed byte or bytes in which unallocated bits remain.
There are situations in which an a foreign function takes the address of a storage location, and writes a new value into that location. Informally, this referred to as an "out parameter" or "in-out parameter", in the case of bidirectional data transfer. In the C language, the familiar pattern looks like this:
void function(int *ptr);
int val = 0;
function(&val);
In the case of an aggregate type, such as a structure, being an in-out or out parameter, this pattern is easily handled in FFI because the corresponding Lisp object is also an aggregate, and therefore has reference semantics: it can be updated to receive the new value. In the case of a scalar, however, such as int in the above example, this may not be not possible. A Lisp integer doesn't have the referential semantics required to receive a new value by pointer, and there is no "address-of" concept to create a reference to its location.
To understand the following FFI trick, it helps to first rework the C into a different form:
void function(int *ptr);
int val[1] = { 0 };
abc_function(val);
Instead of a scalar value, we can declare an array of 1 element of that same type, and pass the array (which converts into a pointer to that element). This approach inspires a similar trick in the FFI domain:
(with-dyn-lib (...)
(deffi abc-function "abc_function" void ((ptr (array 1 int)))))
(let ((val (vec 0)))
(abc-function val)
;; [vec 0] has updated value coming from function
)
We define the parameter of abc-function as being a pointer to an array of 1 int rather than an int, and then pass a vector as the argument. If the parameter is in-out, then the vector must be constructed or initialized to contain a value that will convert to the C type. If the parameter is out only, then the FFI definition can use ptr-out and the vector can contain the nil value.
The FFI mechanism makes use of a type-like representation called the "call descriptor". A call descriptor is an object which uses FFI types to describe function arguments and return values. A FFI descriptor is required to call a foreign function, and to create a FFI closure to use as a callback function from a foreign function back into TXR Lisp.
A FFI descriptor object can be constructed from a return value type, and a list of argument types, and several other pieces of information using the function ffi-make-call-desc.
This object can then be passed to ffi-call to specify the C type signature of a foreign function, or to ffi-make-closure to specify the C type signature of a FFI closure to bind to a Lisp function.
The FFI macros deffi and deffi-cb provide a simplified syntax for expressing FFI call descriptors, which includes a notation for expressing variadic calls.
A note about variadic foreign functions: although there is support in the call descriptor mechanism for expressing a variadic function, it expresses a particular instance of a variadic function, rather than the variadic function's type per se. To call the same variadic function using different variadic arguments, different call descriptors are required. For instance to perform the equivalent of the C function call printf("hello\n") requires a certain descriptor. To perform the equivalent of printf("hello, %s\n", name) requires a different descriptor.
This group of functions comprises the basic interface to the TXR Lisp's FFI type system module.
(ffi-type-compile syntax)
The ffi-type-compile function produces and returns a compiled type object from a syntax argument which specifies valid FFI syntax. If the type syntax is invalid, or specifies a nonexistent type specifier or operator, an exception is thrown.
Note: whenever a function argument is required to be of FFI type, what it means is that it must be a compiled object, and not a Lisp expression denoting FFI syntax.
(ffi-type-compile 'int) -> #<ffi-type int>
(ffi-type-compile
'(array 3 double)) -> #<ffi-type (array 3 double)>
(ffi-type-compile 'blarg) -> ;; error
(ffi-make-call-desc ntotal nfixed rettype
argtypes [name])
The ffi-make-call-desc function constructs a FFI call descriptor.
The ntotal argument must be a nonnegative integer; it indicates the number of arguments in the call.
If the call denotes a variadic function, the nfixed argument must be an integer at least 1 and less than ntotal, denoting the number of fixed arguments. If the call denotes an ordinary, non-variadic function, then nfixed must either be specified specified as nil or else equal to the ntotal argument.
The rettype parameter must be an FFI type. It specifies the function return type. Functions which don't return a value are specified by the (compiled version of) the return type void.
The argtypes argument must be a list of types, containing at least ntotal elements. If the function takes no arguments, this list is empty. If the function is variadic, then the first nfixed elements of this list specify the types of the fixed arguments; the remaining elements specify the variadic arguments.
The name argument gives the name of the function for which this description is intended, or some other identifying symbol. This symbols is used in diagnostic messages related to errors in the construction of the descriptor itself or its subsequent use. If this parameter is omitted, then the involved FFI functions use their own names in reporting diagnostics.
Note: variadic functions must not be called using a non-variadic descriptor, and vice versa, even if the return types and argument types match.
Note: unlike the deffi and deffi-cb macros, the ffi-make-call-desc function doesn't perform any special treatment of variadic parameter types. When any of the types float, be-float or le-float occur in the variadic portion of argtypes, it is unspecified whether a descriptor is successfully produced and returned or whether an exception is thrown. If a descriptor is successfully produced, and then subsequently used for making or accepting calls, the behavior is undefined.
;;
;; describe a call to the variadic function
;;
;; type void (*)(char *, ...)
;;
;; with these actual arguments
;;
;; (char *, int)
;;
(ffi-make-call-desc
2 ;; two arguments
1 ;; one fixed
(ffi-type-compile 'void) ;; returns nothing
(list (ffi-type-compile 'str) ;; str -> char *
(ffi-type-compile 'int))) ;; int
-->
#<ffi-call-desc #<ffi-type void>
(#<ffi-type str> #<ffi-type int>)>
(ffi-type-operator-p symbol)
The ffi-type-operator-p function return t if symbol is a type operator symbol: a symbol used in the first position of a recognized compound type form in the FFI type system.
Otherwise, it returns nil.
(ffi-type-p symbol)
The ffi-type-p function returns t if symbol denotes a type in the FFI type system: either a built-in type or an alias type name established by typedef.
Otherwise, it returns nil.
(ffi-make-closure lisp-fun call-desc
[safe-p [abort-val]])
The ffi-make-closure function binds a Lisp function lisp-fun, which may be a lexical closure, or any callable object, with a FFI call descriptor call-desc to produce a FFI closure.
A FFI closure is an object of type ffi-closure which is suitable as an argument for the type denoted by the closure type specifier keyword in the FFI type language.
This type appears a C function pointer in the foreign code, and may be called as such. When it is called by foreign code, it triggers a call to lisp-fun.
The optional safe-p parameter controls whether the closure dispatch is "safe", the meaning of which is described shortly. The default value is t so that unsafe closure dispatch must be explicitly requested with a nil argument for this parameter.
A callback closure which is safely dispatched, firstly, does not permit the capture of delimited continuations across foreign code. Delimited continuations can be captured inside a closure dispatched that way, but the delimiting prompt must be within the callback's local stack frame, without traversing across the foreign stack frames. Secondly, a callback closure which is safely dispatched doesn't permit direct nonlocal control transfers across foreign code, such as exception handling. Such transfers, however, appear to work anyway (with caveats): this is because they are specially handled. The closure dispatch mechanism intercepts all dynamic control transfers, converts them to an ordinary return from the callback to the foreign code, and resumes the control transfer when the foreign code itself finishes and returns. If the callback returns a value (its return type is other than void) then in this situation, the callback returns an all-zero-bits return value to the foreign caller. If the abort-val parameter is specified and its value is other than nil, then that value will be used as the return value instead of an all-zero bit pattern.
An unsafely dispatched closure permits the capture of continuations from the callback across the foreign code and direct dynamic control transfers which abandon the foreign stack frames.
Unsafe closure dispatch is only compatible with foreign code which is designed with that usage in mind. For instance foreign code which holds dynamic resources in stack variables will leak those resources if abandoned this way. There are also issues with capturing continuations across foreign code.
Note: the C function pointer is called a "closure" because it carries environment information. For instance, if lisp-fun is a lexical closure, invocations of it through the FFI closure occur in its proper lexical environment, even though its external representation is a simple C function pointer. This requires a special trampoline trick: a piece of dynamically constructed machine code with the closure binding embedded inside it, with the C function pointer pointing to the machine code.
Note: the same call descriptor can be reused multiple times to create different closures. The same Lisp function can be involved in multiple FFI closures.
;; Package the TXR cmp-str function as a string
;; comparison callback compatible with:
;;
;; int (*)(const char *, const char *)
;;
(ffi-make-closure
(fun cmp-str)
(ffi-make-call-desc 2 nil ;; two args, non-variadic
(ffi-type-compile 'int) ;; int return
[mapcar ffi-type-compile '(str str)])) ;; args
(ffi-call fun-cptr call-desc {arg}*)
The ffi-call function invokes a foreign function.
The fun-cptr argument which must be a cptr object. It is assumed to point to a foreign function.
The call-desc argument must be a FFI call descriptor, produced by ffi-make-call-desc.
The call-desc must correctly describe the foreign function.
The zero or more arg arguments are values which are converted into foreign argument values. There must be exactly as many of these arguments as are required by call-desc.
The ffi-call function converts every arg to a corresponding foreign object. If these conversions are successful, the converted foreign arguments are passed by value to the foreign function indicated by fun-cptr. An unsuccessful conversion throws an error.
When the call returns, the foreign function's return value is converted to a Lisp object and returned, in accordance with the return type that is declared inside call-desc.
(ffi-typedef name type)
The ffi-typedef function installs the compiled FFI type given by type as a typedef name under the symbol given by name.
After this registration, whenever the type compiler encounters that symbol being used as a type specifier, it will replace it by the type object it represents.
The ffi-typedef function returns type.
;; define refcount-t as an alias for uint32
(ffi-typedef 'refcount-t (ffi-type-compile 'uint32))
(ffi-size type)
The ffi-size function returns an integer which gives the storage size of the given FFI type: the amount of storage required for the external representation of that type.
Bitfield types do not have a size; it is an error to apply this function to a bitfield.
The size is machine:specific.
(ffi-size '(ffi-type-compile 'double)) -> 8
(ffi-size '(ffi-type-compile 'char)) -> 1
(ffi-size '(ffi-type-compile
'(array 42 char))) -> 42
(ffi-alignof type)
The ffi-alignof function returns an integer which gives the alignment the given FFI type. When an instance of type is placed into a structure as a member, it is placed after the previous member at the smallest available offset which is divisible by the alignment. The bytes skipped from the smallest available offset to the smallest available aligned offset are referred to as padding.
Bitfield types do not have an alignment; it is an error to apply this function to a bitfield. Bitfields are allocated in storage cells, and those cells have alignment which is the same as that of the type int.
The alignment is machine-specific. It may be more strict than what the hardware architecture requires, yet at the same time be smaller than the size of the type. For instance, the size of the type double is commonly 8, yet the alignment is often 4, and this is so even on processors like Intel x86 which can load and store a double at a misaligned address.
The alignment of an array is the same as that of its element type.
The alignment of a structure is that of its member which has the most strict (largest-valued) alignment.
It is a property of arrays, derived from requirements governing the C language, that if the first element of an array is at a correctly aligned address, then all elements are. To ensure that this property holds for for arrays of structures, structures sometimes must include padding at the end. This is because the size of a structure without any padding might not be multiple of its alignment, which is derived from the most strictly aligned member. For instance, if we assume an architecture on which the size and alignment of int is 4, the size of the structure type (struct ab (a int) (b char)) would be 5 if no padding were included. However, in an array of these structures, the second element's a member would be placed at offset 5, rendering it misaligned. To ensure that every a is placed at an offset which is multiple of 4, the struct type is extended with anonymous padding so that its size is 8.
(ffi-alignof (ffi double)) -> 4
(ffi-offsetof type member)
The ffi-alignof function calculates the byte offset of member within the FFI type type.
If type isn't a FFI struct type, or if member isn't a symbol naming a member of that type, the function throws an exception.
An exception is also thrown if member is a bitfield.
(ffi-offsetof (ffi (struct ab (a int) (b char))) 'b) -> 4
(ffi-arraysize type)
The ffi-arraysize function reports the number of elements in type, which must be an array type: an array, zarray or carray.
(ffi-arraysize (ffi (array 5 int))) -> 5
(ffi-elemsize type)
The ffi-elemsize function reports the size of the element type of an array, of the target type of a pointer, or of the base integer type of an enumeration. The type argument must be an array, pointer or enumeration type: a type constructed by one of the operators array, zarray, carray, ptr, ptr-in, ptr-out, enum or enumed.
(ffi-elemsize (ffi (array 5 int))) -> 4 ;; (sizeof int)
(ffi-elemtype type)
The ffi-elemtype function retrieves the element type of an array type, target type of a pointer type, or base integer type of an enumeration. The type argument must be an array, pointer or enumeration type: a type constructed by one of the operators array, zarray, carray, ptr, ptr-in, ptr-out, enum or enumed.
(ffi-elemtype (ffi (ptr int))) -> #<ffi-type int>
This group of macros provides a higher-level language for working with FFI types and defining foreign function bindings. The macros are implemented using the Foreign Function Type API described in the previous section.
(with-dyn-lib lib-expr body-form*)
The with-dyn-lib macro works in conjunction with the deffi, deffi-sym and deffi-var macros.
When a deffi form appears as one of the body-forms of the with-dyn-lib macro, that deffi form is permitted to use the simplified forms of the fun-expr argument, to refer to library functions succinctly, without having to specify the library. The same remark applies to deffi-sym and deffi-var, regarding their var-expr parameter.
A form invoking the with-dyn-lib macro should be a top-level form. The macro creates a global variable named by a symbol generated by gensym whose initializing expression binds it to a dynamic library handle. The macro then creates an environment in which the enclosed deffi, deffi-var and deffi-sym forms can implicitly refer to that library via the global variable.
The lib-expr argument can take on three different forms:
The result value of a with-dyn-lib form is the symbol which names the generated variable which holds the library handle.
;; refer to malloc and free functions
;; in the executable
(with-dyn-lib nil
(deffi malloc "malloc" cptr (size-t))
(deffi free "free" void (cptr)))
;; refer to "draw" function in fictitious
;; "libgraphics" library:
(with-dyn-lib "libgraphics.so.5"
(deffi draw "draw" int (cptr cptr)))
;; refer to "init_foo" function via specific
;; library handle.
(defvarl foo-lib (dlopen "libfoo.so.1"))
(with-dyn-lib foo-lib
(deffi init-foo "init_foo" void (void)))
(deffi name fun-expr rettype argtypes)
The deffi macro arranges for a Lisp function to be defined, via defun, which calls a foreign function.
The name argument must be a symbol suitable as a function name in a defun form. This specifies the function's Lisp name.
The fun-expr parameter specifies the foreign function which is to be called. The syntactic variants permitted for its argument are described below.
The rettype argument must specify the return type, using the FFI type syntax, as an unquoted literal. The macro arranges for the compilation of this syntax via ffi-type-compile.
The argtypes argument must specify a list of the argument types, as an unquoted literal list, using FFI type syntax. The macro arranges for these types to be compiled. Furthermore, a special convention may be used for specifying a variadic function: if the : (colon) keyword symbol appears as one of the elements of argtypes, then the deffi form specifies a fixed call to a foreign function which is variadic. The argument types before the colon keyword are the types of the fixed arguments. The types after the colon, if any, are of the variadic arguments. Special considerations apply to some variadic argument types, described below.
The following syntactic variants are permitted of the fun-expr argument:
When the FFI type float is used as the type of a variadic parameter, deffi replaces it by the FFI type double. This treatment is necessary because the C variadic argument mechanism promotes float values to double. Note: due to this substitution, it is possible to pass floating-point values which are out of range of the float type, without any diagnosis. The behavior of is undefined in the Lisp-to-C direction, if the C function extracts an out-of-range double argument as if it were of type float.
The FFI types be-float and le-float cannot be used for specifying the types of a variadic argument. If any of these occur in that position, deffi throws an error. Rationale: these types are related to the C type float type, which requires promotion in variadic passing. Promotion cannot be performed on floating-point values whose byte order has been rearranged, because promotion is a value-preserving conversion.
The result value of a deffi form is name.
(deffi-cb name rettype argtypes [abort-val])
(deffi-cb-unsafe name rettype argtypes)
The deffi-cb macro defines, using defun a Lisp function called name.
Thus the name argument must be a symbol suitable as a function name in a defun form.
The rettype and argtypes arguments are processed exactly as in the corresponding arguments in the deffi macro.
The deffi-cb macro arranges for rettype and argtypes to be compiled into a FFI call descriptor. The generated function called name then serves as a combinator which takes a Lisp function as its argument, and binds it to the FFI call descriptor to produce a FFI closure. That closure may then be passed to foreign functions as a callback. The deffi-cb macro generates a callback which uses safe dispatch, which is explained in the description of the ffi-make-closure function. The optional abort-val parameter specifies an expression which evaluates to the value to be returned by the callback in the event that a dynamic control transfer is intercepted. The purpose of this value is to indicate to the foreign code that the callback wishes to abort operation; it is useful in situations when a suitable return value will induce the foreign code to cooperate and itself return to the Lisp code which will then continue the dynamic control transfer.
The deffi-cb-unsafe macro is a variant of deffi-cb with the same argument conventions. The difference is that it arranges for ffi-make-closure to be invoked with nil for the safe-p parameter. This macro has no abort-val parameter, since unsafe callbacks do not use it.
;; create a closure combinator which binds
;; Lisp functions to a call descriptor has the C type
;; signature void (*)(int).
(deffi-cb void-int-closure void (int))
;; use the combinator
;; some-foreign-function's second arg is
;; of type closure, specifying a callback:
(some-foreign-function
42
(void-int-closure (lambda (x)
(puts `callback! @x`))))
(deffi-var name var-expr type)
The deffi-var macro defines a global symbol macro which expands to an expression accessing a foreign variable, creating the illusion that the variable is available as a Lisp variable holding a Lisp data type.
The name argument gives the name of the symbol macro to be defined.
The var-expr argument is one of several permitted syntactic forms which specify the address of the foreign variable. They are described below.
The type argument expresses the variable type in FFI type syntax.
Once the variable is defined, accessing the macro symbol name performs a get operation on the foreign variable, yielding the conversion of that variable to a Lisp value. An assignment to the symbol performs a put operation, converting a Lisp object to a value which overwrites the object.
Note: FFI memory management is not helpful in the use of variables. Suppose a string value is stored in a variable of type str. This means that FFI dynamically allocates a buffer which stores the UTF-8 encoded version of the string, and this buffer is placed into the foreign variable. Then suppose another such assignment takes place. The previous value is simply overwritten without being freed.
The following syntactic variants are permitted of the var-expr argument:
(deffi-sym name var-expr [type-sym])
The deffi-sym macro defines a global lexical variable called name whose value is a cptr object that refers to a symbol in a foreign library.
The name argument gives the name for the variable to be defined. This definition takes place place as if by the defparml macro.
The var-expr is syntax which specifies the foreign pointer, using exactly the same conventions as described for the deffi-var macro, allowing for a shorthand notation if this form is enclosed in a with-dyn-lib macro invocation.
The optional type-sym argument must be a symbol. If it is absent, it defaults to nil. This argument specifies the type label for the cptr object which holds the pointer to the foreign symbol.
The result value of deffi-sym is the symbol name.
(typedef name type-syntax)
The typedef macro provides a convenient way to define type aliases.
The type-syntax expression is compiled as FFI syntax, and the name symbol is installed as an alias denoting that type.
The typedef macro yields the compiled version of type-syntax as its value.
(deffi-struct name {(slot type [init-form])}*)
(deffi-union name {(slot type [init-form])}*)
The deffi-struct and deffi-union macros provide a more compact notation for defining FFI structure and union types together with matching typedef names.
The semantics follows from these equivalences:
(deffi-struct S ...) <--> (typedef S (struct S ...))
(deffi-union U ...) <--> (typedef U (union U ...))
(deffi-struct point
(x double)
(y double))
(sizeof type-syntax [object-expr])
The macro sizeof calculates the size of the FFI type denoted by type-syntax.
The type-syntax expression is compiled to a type using ffi-type-compile. The object-expr expression is evaluated to an object value.
If type-syntax denotes an incomplete array or structure type, and the object-expr argument is present, then a dynamic size is computed: the actual number of bytes required to store that object value as a foreign representation.
The sizeof macro arranges for the size calculation to be carried out at macro-expansion time, if possible, so that the sizeof form is replaced by an integer constant. This is possible when the object-expr is omitted, or if it is a constant expression according to the constantp function.
For the type void, incomplete array types, and bitfield types, the one-argument form of sizeof reports zero.
For incomplete structure types, the one-argument sizeof reports a size which is equivalent to the offset of the last member. The size of an incomplete structure does not include padding for the most strictly aligned member.
(alignof type-syntax)
The macro alignof calculates the alignment of the FFI type denoted by type-syntax at macro-expansion time, and produces that integer value as its expansion, such that there is no run-time computation. It uses the ffi-alignof function.
(offsetof type-syntax member-name)
The macro sizeof calculates the offset of the structure member indicated by member-name, a symbol, inside the FFI struct type indicated by type-syntax. This calculation is performed by a macro-expansion-time call to the ffi-offsetof function, and produces that integer value as its expansion, such that there is no run-time computation.
(arraysize type-syntax)
The macro arraysize calculates the number of elements of the array type indicated by type-syntax. This calculation is performed by a macro-expansion-time call to the ffi-arraysize function, and produces that integer value as its expansion, such that there is no run-time computation.
(elemsize type-syntax)
The macro elemsize calculates the size of the element type of an array type, or the size of target type of a pointer type indicated by type-syntax. This calculation is performed by a macro-expansion-time call to the ffi-elemsize function, and produces that integer value as its expansion, such that there is no run-time computation.
(elemtype type-syntax)
The macro elemtype produce the element type of an array type, or the target type of a pointer type indicated by type-syntax. Note: the elemtype macro may be understood in terms of several possible implementations. The form (elemtype X) is equivalent to (ffi-elemtype (ffi-type-compile X)). Since there exists an elemtype type operator, the expression is also equivalent to (ffi-type-compile '(elemtype X)).
(ffi type-syntax)
The ffi macro provides a shorthand notation for compiling a literal FFI type expression to the corresponding type object. The following equivalence holds:
(ffi expr) <--> (load-time (ffi-type-compile 'expr))
Communicating with foreign interfaces sometimes requires representations to be initialized consisting of all zero bits, or mostly zero bits.
TXR provides convenient ways to prepare Lisp objects such that when those objects are converted to a foreign representation, they generate zero-filled representations.
(make-zstruct type {slot-sym init-value}*)
The make-zstruct function provides a convenient means of instantiating a structure for use in foreign function calls, imitating a pattern of initialization often seen in the C language. It instantiates a Lisp struct by conversion of zero-filled memory through FFI, thus creating a Lisp structure which appears zero-filled when converted to the foreign representation.
This simplifies application code, which is spared from providing individual slot initializations which have this effect.
The type argument must be a compiled FFI struct type. The remaining arguments must occur pairwise. Each slot-sym argument must be a symbol naming a slot in the FFI struct type. The init-value argument which follows it specifies the value for that slot.
The make-zstruct function operates as follows. Firstly, the Lisp struct type is retrieved which corresponds to the FFI type given by type. A new instance of the Lisp type is instantiated, as if by a one-argument call to make-struct. Next, each slot indicated by a slot-sym argument is set to the corresponding init-value. Finally, each slot of the struct which is not initialized via slot-sym and init-value pair, and which is known to the FFI type, is reinitialized by a conversion from a foreign object of all-zero bits to a Lisp value. argument. The struct object is then returned.
Note: the znew macro provides a less verbose notation based on make-zstruct.
Note: slots which are not known to the FFI struct type may be initialized by make-zstruct. Each slot-sym must be a slot of the Lisp struct type; but need not be declared as a member in the FFI struct type.
(znew type-syntax {slot-sym init-value}*)
The znew macro provides a convenient way of using make-zstruct, using syntax which resembles that of the new macro.
The znew macro generates a make-zstruct call, arranging for the type-syntax argument to be compiled to a FFI type object, and applies quoting to every slot-sym argument.
The following equivalence holds:
(znew s a i b j ...) <--> (make-zstruct (ffi s)
'a i 'b j ...)
Given the following FFI type definition
(typedef foo (struct foo (a (cptr bar)) (b uint) (c bool)))
the following results are observed:
;; ordinary instantiation
(new foo) -> #S(foo a nil b nil c nil)
;; Under znew, a is null cptr of correct type:
(znew foo) -> #S(foo a #<cptr bar: 0> b 0 c nil)
;; value of b is specified; others come from zeros:
(znew foo b 42) -> #S(foo a #<cptr bar: 0> b 42 c nil)
(zero-fill type obj)
The zero-fill function invokes the by-reference in semantics of FFI type type against a zero-filled buffer, and a Lisp object obj.
This means that if obj is an aggregate such as a vector, list or structure, it is updated as if from an all-zero-bit foreign representation. In that situation, obj is also returned.
An object which has by-value semantics, such as an integer, is not updated. In this case, nevertheless, the return value is a Lisp object produced by converting an all-zero-bit buffer to type.
The following group of functions provides the means for working with foreign unions, in conjunction with the union FFI type.
(make-union type [initval [member]])
The make-union function instantiates a new object of type union, based on the FFI type specified by the type parameter, which must be compiled FFI union type.
The object provides storage for the foreign representation of type, and that storage is initialized to all zero bytes.
Additionally, if initval is specified, but member is not, then initval is stored into the union's via the first member, as if by union-put. If the union type has no members, an error exception is thrown.
If both initval and member are specified, then initval is stored into the union using the specified member, as if by union-put.
(union-members union)
The union-members function retrieves the list of symbols which name the members of union. These are derived from the object's FFI type. It is unspecified whether the list is freshly allocated on each call, or whether the same list is returned; applications shouldn't destructively manipulate this list.
(union-get union member)
The union-get function performs the get semantics (conversion from a foreign representation to Lisp) on the member of union which is specified by the member argument. That argument must be a symbol corresponding to one of the member names.
The union object's storage buffer is treated as an object of the foreign type indicated by that member's type information, and converted accordingly to a Lisp object that is returned.
(union-put union member new-value)
The union-put function performs the put semantics (conversion from a Lisp object to foreign representation) on the member of union which is specified by the member argument. That argument must be a symbol corresponding to one of the member names.
The object given as new-value is converted to the foreign representation according to the type information of the indicated member, and that representation is placed into the union object's storage buffer.
The return value is new-value.
(union-in union memb memb-obj)
(union-out union memb memb-obj)
The union-in and union-out functions perform the FFI in semantics and out semantics, respectively. These semantics are involved in two-way data transfers between foreign representations and Lisp objects.
The union argument must be a union object and the memb argument a symbol which matches one of that object's member names.
In the case of union-in, memb-obj is a Lisp object that was previously stored into union using the union-put operation, into the same member that is currently indicated by member.
In the case of union-out, memb-obj is a Lisp object that was previously retrieved from union using the union-get operation, from the same member that is currently indicated by member.
The union-in performs the by-value nuance of the in semantics on the indicated member: if the member contains pointers to any objects, those objects are updated from their counterparts in memb-obj using their respective by-reference in semantics, recursively.
Similarly union-out performs the by-value nuance of the out semantics on the indicated member: if the member contains pointers to any objects, those objects are updated with their Lisp counterparts in memb-obj using their respective by-reference out semantics, recursively.
Note: union-in is intended to be used after a FFI call, on a union-typed by-value argument, or a union-typed object contained in an argument, in situations when the function is expected to have updated the contents of the union. The union-out function is intended to be used in a FFI callback, on a union-typed callback argument or union-typed object contained in such an argument, in cases when the callback has updated the Lisp object corresponding to a union member, and that change needs to be propagated to the foreign caller.
These functions provide a way to perform I/O on stream using the foreign representation of Lisp objects, performing conversion between the Lisp representations in memory and the foreign representations in a stream.
The stream argument used with these functions must be a stream object which, in the case of input functions, supports get-byte and, in the case of output, supports put-byte.
(put-obj object type [stream])
The put-obj function encodes object into a foreign representation, according to the FFI type type. The bytes of the foreign representation are then written to stream.
If stream is omitted, it defaults to *stdout*.
If the operation successfully writes all bytes of the representation to stream, the value t is returned. A partial write causes the return value to be nil.
All other stream error situations throw exceptions.
(get-obj type [stream])
The get-obj function reads from stream the bytes corresponding to a foreign representation according to the FFI type type.
If stream is omitted, it defaults to *stdin*.
If the read is successful, these bytes are decoded, producing a Lisp object, which is returned.
If the read is incomplete, the value returned is nil.
All other stream error situations throw exceptions.
(fill-obj object type [stream])
The fill-obj function reads from stream the bytes corresponding to a foreign representation according to the FFI type type.
If the read is successful, then object is updated, if possible, from that representation, using the by-value in semantics of the FFI type and returned. If a by-value update of object isn't possible, then a new object is decoded from the data and returned.
If the read is incomplete, the value returned is nil.
All other stream error situations throw exceptions.
Functions in this area provide a way to perform conversion between Lisp objects and foreign representation to and from objects of the buf type.
(ffi-put obj type)
(ffi-put-into dst-buf obj type [offset])
The ffi-put function encodes the Lisp object obj according to the FFI type type and returns a new buffer object of type buf which holds the foreign representation.
The ffi-put-into function is similar, except that it uses an existing buffer dst-buf which must be large enough to hold the foreign representation.
The type argument must be a compiled FFI type.
If type is has a variable length, then the actual size of the foreign representation is calculated from obj.
The obj argument must be an object compatible with the conversions implied by type.
The optional offset argument specifies a byte offset from the beginning of the data area of dst-buf where the foreign representation of obj is stored. The default value is zero.
These functions perform the "put semantics" encoding action similar to what happens to the arguments of an outgoing foreign function call.
Caution: incorrect use of this function, or its use in isolation without a matching ffi-in call, can cause memory leaks, because, depending on type, temporary resources may be allocated, and pointers to those resources will be stored in the buffer.
(ffi-out dst-buf obj type copy-p [offset])
The ffi-out function performs the "out semantics" encoding action, similar to the treatment applied to the arguments of a callback prior to returning to foreign code.
It is assumed that obj is an object that was returned by an earlier call to ffi-get, and that the dst-buf and type arguments are the same objects that were used in that call.
The copy-p argument is a Boolean flag which is true if the buffer represents a datum that is being passed by pointer. If copy-p is true, then obj is converted to a foreign representation which is stored into dst-buf. If it is false, it indicates that the buffer itself is a pass-by-value object. This means that the object itself will not be copied, but if it is an aggregate which contains pointers, the operation will recurse on those objects, invoking their "out semantics" action with pass-by-pointer semantics. The required pointers to these indirect objects are obtained from dst-buf.
The optional offset argument specifies a byte offset from the beginning of the data area of dst-buf where the foreign representation of obj is understood to be stored, and where it is updated if requested by copy-p. The default value is zero.
The ffi-out function returns dst-buf.
(ffi-in src-buf obj type copy-p [offset])
The ffi-in function performs the "in semantics" decoding action, similar to the treatment applied to the arguments of a foreign function call after it returns, in order to free temporary resources and recover the new values of objects that have been modified by the foreign function.
It is assumed that src-buf is a buffer that was prepared by a call to ffi-put or ffi-put-into, and that type and obj are the same values that were passed as the corresponding arguments of those functions.
The ffi-in function releases the temporary memory resources that were allocated by ffi-put or ffi-put-into, which are obtained from the buffer itself, where they appear as pointers. The function recursively performs the in semantics across the entire type, and the entire object graph rooted at the buffer.
The copy-p argument is a Boolean flag which is true if the buffer represents a datum that is being passed by pointer. If it is false, it indicates that the buffer itself is a pass-by-value object. Under pass-by-pointer semantics, either a whole new object is extracted from the buffer and returned, or else the slots of obj are updated with new values from the buffer. Under pass-by-value semantics, no such extraction takes place, and obj is returned. However, regardless of the value of copy-p, if the object is an aggregate which contains pointers, the recursive treatment through those pointers involves pass-by-pointer semantics.
This is consistent with the idea that we can pass a structure by value, but that structure can have pointers to objects which are updated by the called function. Those indirect objects are passed by pointer. They get updated, but the parent structure cannot.
If type is has a variable length, then the actual size of the foreign representation is calculated from obj.
The optional offset argument specifies a byte offset from the beginning of the data area of src-buf from which the foreign representation of obj is taken.
The ffi-in function returns either obj or a new object which is understood to have been produced as its replacement.
(ffi-get src-buf type [offset])
The ffi-get function extracts a Lisp value from buffer src-buf according to the FFI type type. The src-buf argument is an object of type buf large enough to hold a foreign representation of type, at the byte offset indicated by the offset argument. The type argument is compiled FFI type. The optional offset argument defaults to zero.
The external representation in src-buf at the specified offset is scanned according to type and converted to a Lisp value which is returned.
The ffi-get operation is similar to the "get semantics" performed by FFI in order to extract the return value of foreign function calls, and by the FFI callback mechanism to extract the arguments coming into a callback.
The type argument may not be a variable length type, such as an array of unspecified size.
Functions in this area provide a means for working with foreign arrays, in connection with the FFI carray type.
(carray-vec vec type [null-term-p])
(carray-list list type [null-term-p])
The carray-vec and carray-list functions allocate storage for the representation of a foreign array, and return a carray object which holds a pointer to that storage.
The argument type, which must be a compiled FFI type, is retained as the carray object's element type.
Prior to returning, the functions initializes the foreign array by converting the elements of vec or, respectively, list into elements of the foreign array. The conversion is performed using the put semantics of type, which is a compiled FFI type.
The length of the returned carray is determined from the length of vec or list and from the value of the Boolean argument null-term-p.
If null-term-p is nil, then the length of the carray is the same as that of the input vec or list.
A true value of null-term-p indicates null termination. This causes the length of the carray to be one greater than that of vec or list, and the extra element allocated to the foreign array is filled with zero bytes.
(carrayp object)
The carrayp function returns t if object is a carray, otherwise it returns nil.
(carray-blank length type)
The carray-blank function allocates storage for the representation of a foreign array, filling that storage with zero bytes, and returns a carray object which holds a pointer to that storage.
The argument type, which must be a compiled FFI type, is retained as the carray object's element type.
The length argument must be a nonnegative integer; it specifies the number of elements in the foreign array and is retained as the carray object's length.
The size of the foreign array is the product of the size of type as reported by the ffi-size function, and of length.
(carray-buf buf type [offset])
The carray-buf function creates a carray object which refers to the storage provided and managed by the buffer object buf, providing a view of that storage, and manipulation thereof, as an array.
The optional offset parameter specifies an offset from the start of the buffer to the location which is interpreted as the start of the carray, which extends from that offset to the end of the buffer.
The default value is zero: the carray covers the entire buffer.
If a value is specified, it must be in the range zero to the length of buf.
The type argument must be a compiled FFI type whose size is nonzero.
The carray is overlaid onto the storage of buf as follows:
First, offset is subtracted from the bytewise length of buf, as reported by length-buf function to produce the effective length of the storage to be used for the array.
The effective length is divided by the size of type, as reported by ffi-size. The resulting quotient represents the length (number of elements) of the carray object.
Note: the returned carray object holds a reference to buf, preventing buf from being reclaimed by garbage collection, thereby protecting the underlying storage from becoming invalid. A subsequent invocation of carray-own operation releases this reference.
Note: the relationship between the carray object and buf is inherently unsafe: if buf is subsequently subject to operations which reallocate the storage, such as buf-set-length the pointer stored inside the referencing carray object becomes invalid, and operations involving that pointer have undefined behavior.
Note: if the length of the buffer is not evenly divisible by the size of the type, the calculated number of elements is rounded down. The trailing portion of the buffer corresponding to the division remainder, being insufficient to constitute a whole array element, is excluded from the array view.
(carray-buf-sync carray)
The carray-buf-sync function requires carray to be a carray object which refers to a buf object for its storage. Such objects are created by the function carray-buf.
The carray-buf-sync function retrieves and returns the buffer object associated with carray and at the same time also updates the internal properties of carray using the current information: the pointer to the data, and the length of carray are altered to reflect the current state of the buffer.
(buf-carray carray)
The buf-carray function duplicates the underlying storage of carray and returns that storage represented as an object of buf type.
The storage size is calculated by multiplying the carray object's element size by the number of elements. Only that extent of the storage is duplicated.
(carray-cptr cptr type [length])
The carray-cptr function creates a carray object based on a pointer derived from a cptr object.
The cptr argument must be of type cptr. The object's cptr type tag is ignored.
The type argument must specify a compiled FFI type, which will become the element type of the returned carray.
If length is specified as nil, or not specified, then the returned carray object will be of unknown length. Otherwise, length must be a nonnegative integer which will be taken as the length of the array.
Note: this conversion is inherently unsafe.
(cptr-carray carray [type-symbol])
The cptr-carray function returns a cptr object which holds a pointer to a carray object's storage area. The carray argument must be of type carray.
The type-symbol argument should be a symbol. If omitted, it defaults to nil. This symbol becomes the cptr object's type tag.
The lifetime of the returned cptr object is independent from that of carray. If the lifetime of carray reaches its end before that of the cptr, the pointer stored inside the cptr becomes invalid.
(length-carray carray)
The length-carry function returns the length of the carray argument, which must be an object of type carray.
If carray has an unknown length, then nil is returned.
(copy-carray carray)
The copy-carray function returns a duplicate of carray.
The duplicate has the same element type and length, but has its own copy of the underlying storage. This is true whether or not carray owns its storage or not. In either case, the duplicate owns its copy of the storage.
(carray-set-length carray length)
The carry-set-length attempts to change the length of carray, which must be an object of carray type.
The length argument indicates the new length, which must be a nonnegative integer.
The operation throws an error exception if length is negative.
An error exception is also thrown if carray is an object which owns the underlying storage. There is no provision in the carray type to change the storage size.
It is permissible to change the length of a carray object which acts as a view into a buffer (as constructed via the carray-buf operation).
This creates a potentially unsafe situation in which the length requires a larger amount of backing storage than is provided by the buffer.
(carray-ref carray idx)
(set (carray-ref carray idx) new-val)
The carray-ref function accesses an element of the foreign array carray, converting that element to a Lisp value, which is returned.
The idx argument must be a nonnegative integer. If carray has a known length, idx must be less than the length.
If carray has an unknown length, then the access is permitted regardless of how positive is the value of idx. Whether the access has well-defined behavior depends on the actual extent of the underlying array storage.
The validity of any access to the underlying storage depends on the validity of the pointer to that storage.
The access to the array storage proceeds as follows. Every carray object has an element type, which is a compiled FFI type. A byte offset address is calculated by multiplying the size of the element type of carray by idx. Then, the get semantics of the element type is invoked to convert, to a Lisp object, a region of data starting at calculated byte offset in the array storage. The resulting object is returned.
Assigning an a value to a carray-ref form is equivalent to using carray-refset to store the value.
(carray-refset carray idx new-val)
The carray-refset function accesses an element of the foreign array carray, overwriting that element with a new value obtained from a conversion of the Lisp value new-val.
The return value is new-val.
The idx argument must be a nonnegative integer. If carray has a known length, idx must be less than the length.
If carray has an unknown length, then the access is permitted regardless of how positive is the value of idx. Whether the access has well-defined behavior depends on the actual extent of the underlying array storage.
The validity of any access to the underlying storage depends on the validity of the pointer to that storage.
The access to the array storage proceeds as follows. Every carray object has an element type, which is a compiled FFI type. A byte offset address is calculated by multiplying the size of the element type of carray by idx. Then, the put semantics of the element type is invoked to convert new-val to a foreign representation, which is written into the array storage started at the calculated byte offset.
If new-val has a type which is not compatible with the element type, or a value which is out of range or otherwise unsuitable, an exception is thrown.
(carray-dup carray)
(carray-own carray)
The carray-dup function acts upon a carray object which doesn't own its underlying array storage. It allocates a duplicate copy of the array storage referenced by carray, and assigns to carray the new copy. Then it marks carray as owning that storage. Lastly, if carray references another object, that reference is removed; carray no longer prevents the other object from being reclaimed by the garbage collector.
If carray already owns its storage, then this function has no effect.
If carray has an unknown size, then an error exception is thrown.
A carray produced by the functions carray-vec or carray-blank already owns its storage.
A carray object does not own its storage if it is produced by carray-buf or by the conversion of a foreign pointer under the control of the carray FFI type.
Because carray objects derived from foreign pointers via FFI have an unknown size, before using carray-dup, the application must determine the length of the array, and call carray-set-length to establish that length.
After carray-dup, the length may not be altered.
The carray-dup function returns t if it has performed the duplication operation. If it has done nothing, it returns nil.
The carray-own function resembles carray-dup, differing from that function only in two ways. Instead of allocating a duplicate copy of the underlying array storage, carray-own causes carray to assume ownership of the existing storage. Secondly, it is an error to use carray-own on a carray which references a buffer object.
The carray-own function always returns nil.
In all other regards, the descriptions of carray-dup apply to carray-own.
(carray-free carray)
If carray is a carray object which owns the storage to which it refers, then carray-free function liberates that storage by passing the pointer to the C library function free. It then replaces that pointer with a null pointer, and changes the size to zero.
If carray doesn't own the storage, an exception is thrown.
(carray-type carray)
The carray-type function returns the element type of carray, a compiled FFI type.
(vec-carray carray [null-term-p])
(list-carray carray [null-term-p])
The vec-carray and list-carray functions convert the array storage of carray to a freshly constructed object representation: vector, and list, respectively. The new vector or list is returned.
The carray object must have a known size; an error exception is thrown if these functions are invoked on a carray object of unknown size.
The effective length of the new vector or list is derived from the length of carray, taking into account the value of null-term-p.
The null-term-p Boolean parameter defaults to nil. If specified as true, then it has the effect that the effective length of the returned vector or list is one less than that of carray: in other words, a true value of null-term-p indicates that carray holds storage which represents a null-terminated array, and the terminating null element is to be excluded from the conversion.
If null-term-p is true, but the length of carray is already zero, then it has no effect; the effective length remains zero, and a zero-length vector or list is returned.
Conversion of the foreign array to the vector or list is performed by iterating over all of its elements, starting from element zero, up to the element before the effective length.
(carray-get carray)
(carray-getz carray)
The carray-get and carray-getz functions treat the contents of carray as a FFI array and zarray type, respectively.
They invoke the get semantics to convert the FFI array to a Lisp object, and return that object.
If the element type is one of char, bchar or wchar, then the expected string conversion semantics applies.
(carray-put carray new-val)
(carray-putz carray new-val)
The carray-put and carray-putz functions treat the contents of carray as a FFI array and zarray type, respectively.
They invoke the put semantics to convert the Lisp object new-val array to the foreign array representation, which is placed into the array storage referenced by carray.
If the element type is one of char, bchar or wchar, then the expected string conversion semantics applies.
Both of these functions return carray.
(carray-sub carray [from [to]])
(set (carray-sub carray [from [to]]) new-val)
The carray-sub function extracts a subrange of a carray object, returning a new carray object denoting that subrange.
The semantics of from and to work exactly like the corresponding arguments of the sub accessor, following the same conventions.
The returned carray shares the array has the same element type as the original and shares the same array storage. If, subsequently, elements of the original array are modified which lie in the range, then the modifications will affect the previously returned subrange carray. The returned carray references the original object, to ensure that as long as the returned object is reachable by the garbage collector, so is the original. This relationship can be severed by invoking carray-dup on the returned object, after which the two no longer share storage, and modifications in the original are not reflected in the subrange.
If carray-sub is used as a syntactic place, the argument expressions carray, from, to and new-val are evaluated just once. The prior value, if required, is accessed by calling carray-sub and new-val is then stored via carray-replace.
(carray-replace carray item-sequence [from [to]])
The carray-replace function is a specialized version of replace which works on carray objects. It replaces a sub-range of carray with elements from item-sequence. The replacement sequence need not have the same length as the range which it replaces.
The semantics of from and to work exactly like the corresponding arguments of the replace function, following the same conventions.
The semantics of the carray-replace operation itself differs from the replace semantics on sequences in one important regard: the carray object's length always remains the same.
The range indicated by from and to is deleted from carray and replaced by elements of item-sequence, which undergo conversion to the foreign type that defines the elements of carray.
If this operation would make the carray longer, any elements in excess of the object's length are discarded, whether they are the original elements, or whether they come from item-sequence. Under no circumstances does carray-replace write an element beyond the length of the underlying storage.
If this operation would make the carray shorter (the range being replaced is longer than item-sequence) then the downward relocation of items above the replacement range creates a gap at the end of carray which is filled with zero bytes.
The return value is carray itself.
(carray-pun carray type [offset [size-limit]])
The carray-pun creates a new carray object which provides an aliased view of the same data that is referenced by the original carray object.
The type argument specifies the element type used by the returned aliasing array.
If the offset argument is specified, then the aliased view is displaced by that many bytes from the start of the carray object. The offset argument must not be larger than the bytewise length of the array, or an error exception is thrown. The bytewise length of the array is the product of the number of elements and the element size. The default value of offset is zero: no displacement.
If size-limit is specified, it indicates the size, in bytes, of the aliased view. This limit must not be such that the aliased view would extend beyond the array, or an error exception is thrown. If omitted, size-limit defaults to the entire remainder of the array, after the offset. The number of elements of the returned array are then calculated from size-limit.
The carray-pun function calculates how many elements of type fit into size-limit. This value becomes the length of the aliasing array which is returned.
Since the returned aliasing array and the original refer to the same storage, modifications performed in one view are reflected in the other.
The aliasing array holds a reference to the original, so that as long as it is reachable by the garbage collector, so is the original. That relationship is severed if carray-dup is invoked on the aliasing array.
The meaning of the aliasing depends entirely on the bitwise representations of the types involved.
Note: carray-pun does not check whether offset is a value that is suitably aligned for accessing elements of type; on some platforms that must be ensured.
The carray-pun function may be invoked on an object that was itself returned by carray-pun.
(carray-uint number [type])
(carray-int number [type])
The carray-uint and carray-int functions convert number, an integer, to a binary image, which is then used as the underlying storage for a carray.
The type argument, a compiled FFI type, determines the element type for the returned carray. If it is omitted, it defaults to the uchar type, so that the array is effectively of bytes.
Regardless of type, these functions first determine the number of bytes required to represent number in a big endian format. Then the number of elements is determined for the array, so that it provides at least as that many bytes of storage. The representation of number is then placed into this storage, such that its least significant byte coincides with the last byte of that storage. If the number is smaller than the storage provided by the array, it extended with padding bytes on the left, near the beginning of the array.
In the case of carray-uint, number must be a nonnegative integer. An unsigned representation is produced which carries no sign bit. The representation is as many bytes wide as are required to cover the number up to its most-significant bit whose value is 1. If any padding bytes are required due to the array being larger, they are always zero.
The carray-int function encodes negative integers also, using a variable-length two's complement representation. The number of bits required to hold the number is calculated as the smallest width which can represent the value in two's complement, including a sign bit. Any unused bits in the most-significant byte are filled with copies of the sign bit: in other words, sign extension takes place up to the byte size. The sign extension continues through the padding bytes if the array is larger than the number of bytes required to represent number; the padding bytes are filled with the value #b11111111 (255) if the number is negative, or else 0 if it is nonnegative.
(uint-carray carray)
(int-carray carray)
The uint-carray and int-carray functions treat the storage bytes carray object as the representation of an integer.
The uint-carray function simply treats all of the bytes as a big-endian unsigned integer in a pure binary representation, and returns that integer, which is necessarily always nonnegative.
The int-carray function treats the bytes as a two's complement representation. The returned number is negative if the first storage byte of carray has a 1 in the most significant bit position: in other words, is in the range #x80 to #xFF. In this case, the two's complement of the entire representation is calculated: all of the bits are inverted, the resulting positive integer is extracted. Then 1 is added to that integer, and it is negated. Thus, for example, if all of the bytes are #xFF, the value -1 is returned.
(fill-carray carray [pos [stream]])
(put-carray carray [pos [stream]])
The fill-carray and put-carray functions perform stream output using the carray object as a buffer.
The semantics of these functions is as follows. A temporary buffer is created which aliases the storage of carray and this buffer is used as an argument in an invocation of, respectively, the buffer I/O function fill-buf or put-buf.
The value returned by the buffer I/O function is returned.
The pos and stream arguments are defaulted exactly in the same manner as by fill-buf and put-buf, and have the same meaning. In particular, pos indicates a byte offset into the carray object's storage, not an array index.
TXR Lisp supports interfacing with modules that make use of the C setjmp and longjmp feature across their boundaries. It is possible to save a jump location in Lisp code with the setjmp macro, such that a foreign function can perform a longjmp to that saved context.
The jump context buffer, known as the type jmp_buf in C, is modelled as a carray object whose element type is uchar. The function jmp-buf returns such an object. Foreign functions that return a pointer to a jmp_buf may be suitably defined via deffi such that the pointer is mapped to a carray object whose element type is uchar. The resulting object will then be usable as a jump buffer.
The features described here are unsafe. When used in certain incorrect ways, the behavior is undefined.
Using the setjmp macro and longjmp function as control primitives in Lisp code not interacting with foreign functions is strongly discouraged.
There are situations in which the foreign function calling mechanism allocates temporary dynamic memory for converting between Lisp and C objects. These situations occur when objects are referenced by pointers, and so are are outside of the stack-based argument space. In such a situation, if the foreign function performs a longjmp terminating in a setjmp macro in Lisp code, that temporary storage will leak.
(jmp-buf)
The jmp-buf function returns a new carray object suitable for use as a jump buffer with the setjmp macro and longjmp function.
(longjmp jmp-buf value)
The longjmp function restores the context saved into the jmp-buf object by the setjmp macro. If that macro already terminated, the behavior is undefined.
The value must be an integer in range of the FFI type int. That value will be observed in the setjmp form, as described. If value is 0 (zero) the value 1 is used instead. This is a behavior of the underlying longjmp C library function.
Note: a context abandoned via longjmp will not perform unwinding, similarly to sys:abscond*. The form which is abandoned by longjmp should not be using scoped management of resources that relies on unwind-protect for clean-up.
(setjmp jmp-buf result-var main-form longjmp-form*)
The setjmp macro saves the jump context into the jmp-buf object, and evaluates the main-form expression.
If the main-form expression terminates normally then the value it produces becomes the result of setjmp, which terminates.
If the main-form performs a longjmp to the context saved in jmp-buf, then that form is abruptly terminated, without performing any unwinding. Then, the zero or more longjmp-forms are evaluated. The setjmp form terminates, yielding the value of the last longjmp-form or else nil.
The longjmp-forms are evaluated in a scope in which the result-var symbol is bound as a variable, taking on the integer value passed to longjmp, which is never zero.
The jmp-buf argument must be a carray object suitable for use as a jump buffer.
The result-var argument must be a bindable symbol.
Once setjmp terminates, the contents of jmp-buf become indeterminate. Any longjmp attempt using an indeterminate jmp-buf is undefined behavior.
(let ((jb (jmp-buf)))
(setjmp jb result
(progn (put-line "setjmp") ;; "setjmp" is printed
(longjmp jb 42))
(put-line `result is: @result`))) ;; "result is: 42" is printed
Note: this example is for illustration only. Using setjmp and longjmp as Lisp control flow constructs in code not interacting with foreign functions is strongly discouraged.
TXR supports two modes of processing of Lisp programs: evaluation and compilation.
Expressions entered into the listener, loaded from source files via load, processed by the eval function, or embedded into the TXR pattern language, are processed by the evaluator. The evaluator expands all macros, and then interprets the program by traversing its raw syntax tree structure. It uses an inefficient representation of lexical variables consisting of heap-allocated environment objects which store variable bindings as Lisp association lists. Every time a variable is accessed, the chain of environments is searched for the binding.
TXR also provides a compiler and virtual machine for more efficient execution of Lisp programs. In this mode of processing, top-level expressions are translated into the instructions of Lisp-oriented virtual machine. The virtual machine language is traversed more efficiently compared to the traversal of the cons cells of the original Lisp syntax tree. Moreover, compiled code uses a much more efficient representation for lexical variables which doesn't involve searching through an environment chain. Lexical variables are always allocated on the stack (the native one established by the operating system). They are transparently relocated to dynamic storage only when captured by lexical closures, and without sacrificing access speed.
TXR provides the function compile for compiling individual functions, both anonymous and named. File compilation is supported via the function compile-file. The function compile-toplevel is provided for compiling expressions in the global environment. This function is the basis for both compile and compile-file.
The disassemble function is provided to list the compiled code in a more understandable way; disassemble takes a compiled code object and decodes it into an assembly language presentation of its virtual-machine code, accompanied by a dump of the various information tables.
File compilation via compile-file refers to a processing step whereby a source file containing TXR Lisp forms (typically named with a .tl file name suffix) is translated into an object file (named with a .tlo suffix) containing a compiled version of those forms.
The compiled object file can then be loaded via the load function instead of the source file. Usually, loading the compiled file produces the same effect as if the source file were loaded. However, note that the behavior of compiled code can differ from interpreted code in a number of ways. Differences in behavior can be deliberately induced. Certain erroneous or dubious situations can also cause compiled code to behave differently from interpreted code.
Compilation not only provides faster execution; compiled files also load much faster than source files. Moreover, they can be distributed unaccompanied by the source files, and resist reverse engineering.
An important concept in file compilation via compile-file is that of the top-level form, and how that term is defined. The file compiler individually processes top-level forms; for each such form, it emits a translated image.
In the context of file compilation, a top-level form isn't simply any Lisp form which is not enclosed by another one. Rather, in this specific context, it has this specific definition, which allows some enclosed forms to still be considered top-level forms:
Note that the constituent body forms of a macrolet or symacrolet top-level form are not individual top-level forms, even if the expansion of the construct combines the expanded versions of those forms with progn.
Note: the eval function implements a similar concept, specially recognizing progn, compile-only and eval-only top-level forms, taking care to macro-expand and evaluate their constituents separately. In turn, the load function, when processing Lisp source, evaluates each primary top-level form as if by using the eval function. The result is that the behavior of loaded source and compiled files is consistent in this regard.
The file compiler reads each successive forms from a file, performs a partial expansion on that form, then traverses it to identify all of the top-level forms which it contains. Each top-level form is subject to three actions, either of the latter two of which may be omitted: compilation, execution and emission. Compilation refers to the translation to compiled form. Execution is the invocation of the compiled form. Emission refers to appending an externalized representation of the compiled form (its image) to the output which is written into the compiled file.
By default, all three actions take place for every top-level form. Using the operators compile-only or eval-only, execution or emission, or both, may be suppressed. If both are suppressed, then compilation isn't performed; the forms processed in this mode are effectively ignored.
When a compiled file is loaded, the images of compiled forms are read from it and converted back to compiled objects, which are executed in sequence.
Partial expansion means that file compilation doesn't fully expand each form that is encountered. Rather, an incremental expansion is performed, similar to the algorithm used by the eval function:
Programs specify not only code, but also data. Data embedded in a program is called literal data. There are restrictions on what kinds of object may be used as literal data in programs subject to file compilation. Programs which stray outside of these restrictions will produce compiled files which load incorrectly or fail to load.
Literal objects arise not only from the use of literal such as numbers, characters and strings, and not only from quoted symbols or lists. For instance, compiled forms which define or reference free variables or global functions require the names of these variables or functions to be represented as literals.
An object used as a literal in file-compiled code must be externalizable which means that it has a printed representation which can be scanned to produce a similar object. An object which does not have a readable printed representation will give rise to a compiled file which triggers an exception. Literals which are themselves read from program source code naturally meet this restriction; however, with the use of macros, it is possible to embed arbitrary objects into program code.
If the same object appears in two or more places in the code specified in a single file, the file compilation and loading mechanism ensures that the multiple occurrences of that object in the compiled file become a single object when the compiled file is loaded. For example, if macros are used in such a way that the compiled file defines a function which has a name generated by gensym, and there are calls to that function throughout that file, this will work properly: the multiple occurrences of the gensym will appear as the same symbol. However: that symbol in the loaded file will not be identical to any other symbol in the TXR image; it will be newly allocated each time the compiled file is loaded.
Interned symbols are recorded in a compiled file by means of their textual names and package prefixes. When a compiled file is loaded, the interned symbols which occur as literals in it are entered into the specified packages under the specified names. The value of the *package* special variable has no influence on this.
Circular structures in compiled literals are preserved; on loading, similar circular structures are reproduced.
TXR supports the hash-bang mechanism in compiled .tlo files, thereby allowing compiled scripts to be executable.
When a source file begins with the #! (hash bang) character sequence, the file compiler propagates that line (all characters up to and including the terminating newline) to the compiled file, subject to the following transformation steps:
Furthermore, certain permissions are propagated from a hash-bang source file to the target file. If the source file is executable to its owner, then the target file is made executable as if by using chmod with the +x mode: all the executable permissions that are allowed by the current umask are are enabled on the target file. If the target file is thus being marked executable, then additional permissions are also treated as follows. If the target file has the same owner as the source file, and the source file's setuid permission bit is set, then this is propagated to the target file. Similarly, if the target file has the same group owner as the source file, and the source file's group execute bit and setgid permission bit are set, then the setgid bit is set on the target file.
TXR's virtual-machine architecture for executing compiled code is evolving, and that evolution has implications for the compatibility between compiled files and the TXR executable image.
The basic requirement is that a given version of TXR can load and execute the compiled files which that same version has produced.
Furthermore, these files are architecture-independent, except that their encoding is in the local byte order ("endianness") of the host machine. The byte order is explicitly indicated in the files, and the load function resolves it. Thus a file produced by TXR running on a 64-bit big-endian Power PC can be loaded by TXR running on 32-bit x86, which is little endian.
A given TXR version may also be capable of loading files produced by an older version, or even ones produced by a newer version. Whether this is possible depends on the versions involved. Furthermore, there is a general issue at play: code compiled by newer versions of TXR may require functions that are not present in older versions, preventing that code from running. Newer TXR may support new syntax not recognized by older TXR, and that syntax may end up in compiled files.
Compiled files contain a minor and major version number (which is independent of the TXR version). The load function examines these numbers and decides whether the file is loadable, or whether it must be rejected.
The first version of TXR which featured the compiler and virtual machine was 191. Older versions therefore cannot load compiled files.
Versions 191 and 192 produce version 1 compiled files, and load only that version.
Versions 193 through 198 produce version 2 compiled files and load only that version.
Version 199 produces version 3 files and loads versions 2 and 3.
Versions 200 through 215 produce version 4 files and load versions 2, 3 and 4.
Versions 216 through 243 produce version 5.0 files and load versions 2, 3, 4 and 5, regardless of minor version.
Versions 244 through 251 produce version 5.1 files and load versions 2, 3, 4 and 5, regardless of minor version.
Versions 252 through 259 produce version 6.0 files and load only version 6, regardless of minor version.
Versions 260 through 296 produce version 7.0 files and load versions 6 and 7, regardless of minor version. Version 261 introduces JSON #J syntax. Compiled code which contains embedded JSON literals is not loadable by TXR 260 and older.
By default, the unused diagnostic option is enabled in *compile-opts*, causing unused variables to be diagnosed.
The first step in resolving an unused variable diagnostic is to determine whether it is caused by a bug in the code. If so, the resolution is to address the bug.
If the situation isn't a bug, then the diagnostic is a false positive, and may be silenced. There are multiple ways to do that, six of which are given here:
(with-compile-opts (nil unused)
(compile-file "foo.tl"))
(defun stub-function (arg1 arg2)
(ignore arg1 arg2))
Note that an ignore call may be elided if it occurs in dead code, in which case it won't have the right effect:
(defun unused-arg (arg) ;; diagnosed as unused
(when (= (+ 2 2) 5)
(ignore arg) ;; wrongly placed
(dead-code)))
(defun unused-arg (arg) ;; no diagnostic
(ignore arg) ;; correctly placed
(when (= (+ 2 2) 5)
(dead-code)))
(defun platform-specific-action (arg)
(use arg)
(if (eql (sizeof wchar) 2)
(do-something arg)))
However, unlike ignore, use takes exactly one argument, and returns that argument rather than nil.
(lambda (x y) y) ;; unused x diagnosed
(lambda (#:x y) y) ;; no diagnostic
Examples:
(tree-case obj
((a b) (calculate-something a b))
(else (transform obj))) ;; unused else
(tree-case obj
((a b) (calculate-something a b))
(else (transform else))) ;; diagnostic gone
(match-case obj
((@a @b) (calculate-something a b))
(@else (transform obj))) ;; unused else
(match-case obj
((@a @b) (calculate-something a b))
(@else (transform else))) ;; diagnostic gone
(match-case obj
((@a @nil) (calculate-something a))
(@nil (transform obj)))
(defmacro foo (x y) ;; y unused
^(a b c ,x))
(defmacro foo (x t) ;; no diagnostic
^(a b c ,x))
(tree-bind (a . (b . c)) obj ;; a, b, unused
c)
(tree-bind (t . (t . c)) obj ;; no diagnostic
c)
The compile-only and eval-only operators can be used to deliberately produce code which behaves differently when compiled and interpreted. In addition, unwanted differences in behavior can also occur. The situations are summarized below.
Forms evaluated by load-time are treated differently by the compiler. When a top-level form is compiled, its embedded load-time forms are factored out such that the compiled image of the top-level form will evaluate these forms before other evaluations take place. The interpreter doesn't perform this factoring; it evaluates a load-time form when it encounters it for the first time.
The compiler identifies multiple occurrences of equivalent strings and bignum integers that occur as literals, and condenses each one to a single instance, within the scope of the compilation. The scope is possibly as wide as a file.
If the literal "abc" appears in multiple places in the same file that is processed by compile-file, in the resulting compiled file, there may be just a single "abc" object. For instance, if the file contains two functions:
(defun f1 () "abc")
(defun f2 () "abc")
when compiled, these will return the same object such that
(eq (f1) (f2)) -> t
No such de-duplication is performed for interpreted code.
Consequently, code which depends on multiple occurrences of these objects to be distinct objects may behave correctly when interpreted, but misbehave when compiled. Or vice versa. One example is code which modifies a string literal. Under compilation, the change will affect all occurrences of that literal that have been merged into one object. Another example is an expression like (eq "abc" "abc"), which yields nil under interpretation because the two strings are distinct object in spite of appearing side by side in the syntax, but t when compiled, since they denote the same string object.
In the future, objects other than strings and bignums may be similarly consolidated, such as lists and vectors, which means that interpreted code which works today when compiled may misbehave in the future.
Note that objects which are literally notated in source code are not the only kinds of objects considered to be literals. Objects which are constructed by macros and inserted into macro-expansions are also literals. Literals are self-evaluating objects that appear as expressions in the syntax which remains after macro-expansion, as well as arguments of the quote operator. If a macro calculates a new string each time it is expanded, and inserts it into the expansion as a literal, the compiler will identify and consolidate groups of such strings that are identical.
A source file may contain unqualified symbol tokens which are interned in the current package.
In contrast, a compiled file encodes symbols with full package qualification. When a compiled file is loaded, the current package at that time has no effect on the symbols in the compiled file, even if those symbols were specified as unqualified in the original source file.
This difference can lead to surprising behaviors. Suppose a source file contains references to functions or variables or other entities which do not exist. Furthermore, suppose the entities were referenced, in that file, using unqualified symbols which didn't exist, and were expected to come from a different package from the one where they ended up interned. For instance, supposed the file is being processed in a package called abc and is expecting to use a function calc which should come from the xyz package. Unfortunately, no such symbol exists. Therefore, the symbol is interned as abc:calc and not xyz:calc. In that case, it should be sufficient to ensure that the xyz:calc function exists, and then reload the source file. The unqualified symbol token calc in that file will be correctly resolved to xyz:calc that time. However, if the file is compiled, reloading will not be sufficient. Even though the symbol xyz:calc exists, the file will continue to try to refer a function using the symbol abc:calc which comes from a fully qualified representation stored in the compiled file. The file will have to be recompiled to fix the issue.
Unbound variables are treated differently by the compiler. A reference to an unbound variable is treated as a global lexical access. This means that if a variable access is compiled first and then a defvar is processed which introduces the variable as a dynamically scoped ("special") variable, the compiled code will not treat the variable as special; it will refer to the global binding of the variable, even when a dynamic binding for that variable exists. The interpreter treats all variable references that do not have lexical bindings as referring to dynamic variables. The compiler treats a variable as dynamic if a defvar has been processed which marked that variable as special.
Arguments of a dwim form (or the equivalent bracket notation) which are unbound symbols are treated differently by the compiler. The code is compiled under the assumption that all such symbols refer to global functions. For instance, if neither f nor x are defined, then [f x] will be compiled under the assumption that they are functions. If they are later defined as variables, the compiled code will fail because no function named x exists. The interpreter resolves each symbol in a dwim form at the time the form is being executed. If a symbol is defined as a variable at that time, it is accessed as a variable. If it defined as a function, it is accessed as a function.
The symbolic arguments of a dwim form that refer to global bindings are also treated differently by the compiler. For each such symbol, the compiler determines whether it refers to a function or variable and, further, whether the variable is global lexical or special. This treatment of the symbol is then cemented in the compiled code; the compiled code will treat that symbol that way regardless of the run-time situation. By contrast, the interpreter performs this classification each time the arguments of a dwim form are evaluated. The rules are otherwise the same: if the symbol is bound as a variable, it is treated as a variable. If it is bound as a function, it is treated as a function. If it has both bindings, it is treated as a variable. The difference is that this is resolved at compile time for compiled code, and at evaluation time for interpreted code.
The following degenerate situation occurs, illustrated by example. Suppose the following definitions are given:
(defvarl %gensym%)
(defmacro define-secret-fun ((. args) . body)
(set %gensym% (gensym))
^(defun ,%gensym% (,*args) ,*body))
(defmacro call-secret-fun (. args)
^(,%gensym% ,*args))
The idea is to be able to define a function whose name is an uninterned symbol and then call it. An example module might use these definitions as follows:
(define-secret-fun (a) (put-line `a is @a`))
(call-secret-fun 42)
The effect is that the second top-level form calls the function, which prints 42 to standard out. This works both interpreted and compiled with compile-file. Each of these two macro calls generates a top-level form into which the same gensym is inserted. This works under file compilation due to a deliberate strategy in the layout of compiled files, which allows such uses. Namely, the file compiler combines multiple top-level forms into a single object, which is read at once, and which uses the circle notation to unify gensym references.
However, suppose the following change is introduced:
(define-secret-fun (a) (put-line `a is @a`))
(defpackage foo) ;; newly inserted form
(call-secret-fun 42)
This still works when interpreted, and compiles successfully. However, when the compiled file is loaded, the compiled version of the call-secret-fun form fails with an error complaining that the #:g0039 (or other gensym name) function is not defined.
This is because for this modified source file, the file compiler is not able to combine the compiled forms into a single object. It would not be correct to do so in the presence of the defpackage form, because the evaluation of that form affects the subsequent interpretation of symbols. After the package definition is executed, it is possible for a subsequent top-level form to refer to a symbol in the foo package such as foo:bar to occur, which would be erroneous if the package didn't exist.
The file compiler therefore arranges for the compiled forms after the defpackage to be emitted into a separate object. But that division in the output file consequently prevents the occurrences of the gensym to resolve to the same symbol object.
In other words, the strategy for allowing global gensym use is in conflict with support for forms which have a necessary read-time effect such as defpackage.
The solution is to rearrange the file to unravel the interference, or to use interned symbols instead of gensyms.
There are differences in behavior between compiled and interpreted code with regard to delimited continuations. This is covered in the Delimited Continuations section of the manual.
(compile-toplevel form expanded-p)
The compile-toplevel function takes the Lisp form form and compiles it. The return value is a virtual-machine description object representing the compiled form. This object isn't of function type, but may be invoked as if it were a function with no arguments.
Invoking the compiled object is expected to produce the same effect as evaluating the original form using the eval function.
The expanded-p argument indicates that form has already been expanded and is to be compiled without further expansion.
If expanded-p is nil, then it is subject to a full expansion.
Note: in spite of the name, compile-toplevel makes no consideration whether or not form is a "top-level form" according to the definition of that term as it applies to compile-file processing.
Note: a form like (progn (defmacro foo ()) (foo)) will not be processed by compile-toplevel in a manner similar to the processing by eval or compile-file. In this example, defmacro form will not be evaluated prior to the expansion of (foo) (and in fact not evaluated at all) and so the latter expression isn't correctly referring to that macro. The form (progn (macro-time (defmacro foo ())) (foo)) can be processed by compile-toplevel; however, the macro definition now takes place during expansion, and isn't compiled. The compile-file function has no such issue when it encounters such a form at the top-level, because that function will consider a top-level progn form to consist of multiple top-level forms that are compiled individually, and also executed immediately after being compiled.
;; compile (+ 2 2) form and execute to calculate 4
;;
(defparm comp (compile-toplevel '(+ 2 2)))
(call comp) -> 4
[comp] -> 4
(compile function-name)
(compile lambda-expression)
(compile function-object)
The compile function compiles functions.
It can compile named functions when the argument is a function-name. A function name is a symbol denoting an existing interpreted function, or compound syntax such as (meth type name) to refer to methods. The code of the interpreted function is retrieved, compiled in a manner which produces an anonymous compiled function, and then that function replaces the original function under the same name.
If the argument is a lambda expression, then that function is compiled.
If the argument is a function object, and that object is an interpreted function, then its code and lexical environment are retrieved and compiled.
In all cases, the return value of compile is the compiled function.
Note: when an interpreted function object is compiled, the compiled environment does not share bindings with the original interpreted environment. Modifications to the bindings of either environment have no effect on the other. However, the objects referenced by the bindings are shared. Shared bindings may be arranged using the hlet or hlet* macros.
(compile-file input-path [output-path])
(compile-update-file input-path [output-path])
The compile-file function reads forms from an input file, and produces a compiled output file.
First, input-path is converted to a tentative pathname as follows.
If input-path specifies a pure relative pathname, as defined by the pure-rel-path-p function, then a special behavior applies. If an existing load operation is in progress, then the special variable *load-path* has a binding. In this case, load will assume that the relative pathname is a reference relative to the directory portion of that pathname.
If *load-path* has the value nil, then a pure relative input-path pathname is used as-is, and thus resolved relative to the current working directory.
The tentative pathname is converted to an actual input pathname as follows. Firstly, if the tentative pathname ends with one of the suffixes .tl or .txr then it is considered suffixed, otherwise it is considered unsuffixed. If it is suffixed, then the actual pathname is the same as the tentative pathname. In the unsuffixed case, two possible actual input pathnames are considered. First, if the unsuffixed path refers to a file that can be opened, then that unsuffixed path is taken as actual path. Otherwise, the suffix .tl is added to the tentative pathname, and that becomes the actual path.
If the actual path ends in the suffix .txr then the behavior is unspecified.
If the output-path parameter is given an argument, then that argument specifies the output path. Otherwise the output path is derived from the tentative input path as follows. If the tentative input path is unsuffixed, then .tlo is added to it to produce the output path. Otherwise, the suffix is removed from the tentative input path and replaced with the .tlo suffix.
The compile-file function binds the variables *load-path* and *package* similarly to the load function.
Over the compilation of the input file, compile-file establishes a new dynamic binding for several special variables. The variable *load-path* is given a new binding containing the actual input pathname. The *package* variable is also given a new dynamic binding, whose value is the same as the existing binding. Thus if the compilation of the file has side the effect of altering the value of *package*, that effect will be undone when the binding is removed after the compilation completes.
Compilation proceeds according to the File Compilation Model.
If the compilation process fails to produce a successful translation for each form in the input file, the output file is removed.
The compile-update-file function differs from compile-file in the following regard: compilation is performed only if the input file is newer than the output file, or else if the output file doesn't exist.
The compile-file always returns t if it terminates normally, which occurs if it successfully translates every form in the input file, depositing the translation into the output file. If compilation fails, compile-file terminates by throwing an exception.
The compile-update-file function returns t if it successfully compiles, similarly to compile-file. If compilation is skipped, the function returns nil.
Note: the following idiom may be used to load a file, compiling it if necessary:
(or (compile-update-file "file")
(load-file "file"))
However, note that it relies on the effect of compiling a source file being the same as the effect of loading the compiled file. This can only be true if the source file contains no compile-only or eval-only top-level forms.
Two or more compiled files that are compiled by the same version of TXR may be catenated together to produce a single .tlo file. Such a file may be loaded by the load function. The behavior of loading such a file may differ from loading the individual files, because such a load is treated as a single operation.
The special variable *opt-level* provides control over compiler optimizations.
The variable takes on integer values. If the value is nil, it is interpreted as zero. The meaningful range is from 0 to 7. The initial value of the variable is 7.
The meanings of the values are as follows:
(compile-file path)
The clean-file function removes a previously compiled file associated with path, if such a file exists. In situations when it successfully removes a file, it returns t, otherwise nil. The function may also throw an exception, in situations such as encountering a nonexistent directory component or permission problem.
First, if path specifies a pure relative pathname, as defined by the pure-rel-path-p function, and if the *load-path* variable contains a value other than nil, then clean-file calculates the directory name of *load-path* as if by using dir-name and catenates that directory name with path to produce an intermediate path. Otherwise path is considered to be the intermediate path.
Next, the suffix of the intermediate path is examined. If it ends with ".tlo" or ".tlo.gz", then an attempt is made to remove that path, and the function terminates.
If the intermediate path ends with ".tl" or ".txr", then two attempts are made to remove a file: first, the suffix is replaced with ".tlo" and that is attempted to be removed. If that fails due to non-existence, then the suffix ".tlo.gz" is tried.
Otherwise, if the intermediate path doesn't have any of the above suffixes, then an attempt is made to remove the path with the ".tlo" suffix added, and then with the ".tlo.gz" suffix added.
Note: no attempt is made to remove the unmodified intermediate path itself, except in the cases when it ends with ".tlo" or ".tlo.gz", because that risks removing a source file rather than a compiled file.
(with-compilation-unit form*)
When a file is processed by compile-file, certain actions, such as the issuance of diagnostics about undefined functions and variables, are delayed until the file is completely processed.
The with-compilation-unit macro allows these actions to be collectively deferred until multiple files are completely processed.
The macro evaluates each enclosed form in a single compilation environment. After the last form is evaluated, deferred actions of any enclosed compile-file forms are performed, and then the value of the last form is returned.
It is permissible to nest with-compilation-unit forms, lexically or dynamically. The outermost invocation of with-compilation-unit dominates; all deferred compile-file actions are held until the outermost enclosing with-compilation-unit terminates.
(compile-only form*)
(eval-only form*)
These operators take on a special behavior only when they appear as top-level forms in the context of file compilation. When a compile-only or eval-only form is processed by the evaluator rather than the compiler, or when it is processed outside of file compilation, or when it is appears as other than a top-level form even under file compilation, then these operators behave in a manner identical to progn.
When a compile-only form appears as a top-level form under file compilation, it indicates to the file compiler that the forms enclosed in it are not to be evaluated. By default, the file compiler executes each top-level form after compiling it. The compile-only operator suppresses this evaluation.
When an eval-only form appears as a top-level form under file compilation, it indicates to the file compiler that the forms enclosed in it are not to be emitted into the output file. By default, the file compiler includes the compiled image in the output written to the output file. The eval-only operator suppresses this inclusion.
Forms which are surrounded by both an eval-only form and a compile-only form are neither executed nor emitted into the output file. In this situation, the forms are skipped entirely; no compilation takes place.
The compile-file function not only compiles, but also executes every form for the following reason: the correct compilation of forms can depend on the execution of earlier forms. For instance, code may depend on macros. Macros may in turn depend on functions and variables. All those definitions are required in order to compile the dependent code. Those dependencies may be in a separate file which is loaded by a load form; that load form must be executed.
Note that execution of a form implies that the load-time forms that it contains are evaluated (prior to other evaluations). Suppression of the execution of a form also suppresses the evaluation of load-time forms.
Situations in which compile-only is useful are those in which it is desirable to stage the execution of some top-level form into the compiled file, and not have it happen during compilation. For instance:
;; in a main module
(compile-only (start-application))
It is not desirable to have the file compiler try to start the application as a side effect of compiling the main module. The right behavior is to compile the (start-application) top-level form so that this will happen when that module is loaded.
Situation in which eval-only is useful is for specifying forms which have a compile-time effect only, but are not propagated into the compiled file.
For example, since the correct treatment of literal symbols occurring in a compiled file does not depend on the *package* variable, in many cases, the in-package invocation in the file can be wrapped with eval-only:
(eval-only (in-package app))
The in-package form must be evaluated during compilation so that the remaining forms are read in the correct package. However the loading of the compiled versions of those forms doesn't require that package to be in effect; thus a compiled image of the in-package form need not appear in the compiled file.
Macros definitions may be treated with eval-only if the intent is only to make the expanded code available in the compiled file, and not to propagate compiled versions of the macros which produced it.
(defstruct compile-opts ()
shadow-fun shadow-var shadow-cross unused
log-level constant-throws)
The compile-opts structure represents compiler options: its slots are variables which affect compiler behavior. The compiler expects the special variable *compile-opts* to hold a compile-opts structure. It is recommended to manipulate options using the with-compile-opts macro.
Currently, all of the options are diagnostic. In the future, there may be other kinds of options.
Diagnostic options which are Boolean take on the values nil, t, :warn or :error. Numeric options take integer values. The t and :warn value are synonyms. A value of nil means that the option is disabled. The t and :warn values mean that the diagnostic controlled by the option will be emitted as a warning. The :error value indicates that the diagnostic will be an error.
The slots of compile-opts are as follows:
The special variable *compile-opts* holds a value of type compile-opts which is a structure type. It is recommended to manipulate options using the with-compile-opts macro.
(with-compile-opts {(value option*) | form}*)
The with-compile-opts macro takes zero or more arguments. Each argument is either a clause which affects compiler options or else an ordinary form which is processed in a context in which the *compile-options* variable has been affected by all of the previous clauses.
It is unspecified whether the clauses operate destructively on *compile-options* or freshly bind it. However, the macro dynamically binds *compile-options* at least once, so that when it terminates, its previous value is restored. This binding is performed using compiler-let.
When with-compile-opts occurs in code processed by the compiler, all of the clause-driven compile option manipulation is performed in the compiler's own context. The changes to the *compile-options* variable are not visible to the code being compiled. Thus the macro may be used to transparently change compiler options over individual subexpressions in compiled code.
When with-compile-opts occurs in interpreted code, the manipulations of *compile-options* are visible to the forms. This allows interpreted build steps to configure compiler options around functions such as compile-file.
The clauses which operate on options have list syntax consisting of a value followed by one or more symbols which must be the names of options which are compatible with that value. The clause indicates that all those options take on that value.
The possible values are: nil, t, :warn and :error. These values are documented under the description of the compile-opts structure.
The following expression specifies that the file "foo.tl" is to be compiled with function and variable shadowing treated as error, but unused variable checking disabled. Then compile "bar.tl" with unused variable checking enabled.
;; this form must be interpreted in order for
;; the compile-file call to "see" the effect of the
;; option manipulation.
(with-compile-opts
(:error shadow-var shadow-fun)
(nil unused)
(compile-file "foo.tl")
(:warn unused)
(compile-file "bar.tl"))
;; when the following form is compiled, the unused
;; variable warning will be disabled just around
;; the (let (y) x).
(lambda (x)
(with-compile-opts (nil unused)
(let (y) x)))
;; Show detailed traces of what forms are
;; compiled in these two files.
(with-compile-opts
(2 log-level)
(compile-file "foo.tl")
(compile-file "bar.tl"))
(compiler-let ({(sym init-form)}*) body-form*)
The compiler-let operator strongly resembles let* but has different semantics, relevant to compilation. It also has a stricter syntax in that variables may not be symbols without a init-form: only variable binding specifications of the form (sym init-form) are allowed.
Symbols bound using compiler-let are expected to be special variables. For every sym, the expression (special-var-p sym) should be true. The behavior is unspecified for any sym which doesn't name a special variable.
When the compiler encounters the compiler-let construct, the compiler itself establishes a dynamic scope in which the implied special variable bindings are in effect. This effect is not incorporated into the compiled code. The compiler then implicitly places the body-forms, into a progn from, and compiles that form. While the implicit progn is being compiled, the dynamic bindings established by compiler-let are in scope.
Thus compiler-let may be used to bind special variables which influence compiler behavior.
The compiler-let form is treated like let* by the interpreter, provided that every sym names a special variable.
(load-time form)
The load-time macro makes it possible for a program to evaluate a form, such that, subsequently, the value of that form is then treated as if it were a literal object.
Literals are pieces of the program syntax which are not evaluated at all. On the other hand, the values of expressions are not literals.
From time to time, certain situations benefit from the program being able to perform an evaluation, and then have the result of that evaluation treated as a literal.
The macro-time macro makes this possible in its particular manner: that macro allows one or more expressions to be evaluated during macro expansion. The result of the macro-time is then quoted and substituted in place of the expression. That result then appears as a true literal to the executing code.
The load-time macro similarly arranges for the single form form to be evaluated. However, this evaluation doesn't take place at expansion time. It is delayed until the program executes.
What exactly "delayed until the program executes" means depends on whether load-time is used in compiled or interpreted code, and in what situation is it compiled.
If the load-time form appears in interpreted code, then the exact time when form is evaluated is unspecified. The evaluator may identify all load-time forms which occur anywhere in a top-level expression, and perform their evaluations immediately, before evaluating the form itself. Then, when the load-time forms are encountered again during the evaluation of the form, they simply retrieve the previously evaluated values as if they were literal. Or else, the evaluation may be performed late: when the load-time form itself is encountered during normal evaluation. In that case, form will still be evaluated only once and then its value will be be inserted as a literal in subsequent reevaluations of that load-time form, if any.
If a load-time form appears in a non-top-level expression which is compiled, the compiler arranges for the compiled version of form to be executed when compiled version of the entire expression is executed. This execution occurs early, before the execution of forms that are not wrapped in load-time. The value produced by form is entered into the static data vector associated with the compiled top-level expression, which also holds ordinary literals. Whenever the value of that load-time form is required, the compiled code references it from the data vector as if it were a true literal.
When a load-time top-level form is processed by compile-file, it has no unusual semantics; the effect is that it is replaced by its argument form, which is in that case also considered a top-level form.
The implications of the translation scheme may be understood separately from the perspective of code processed with compile-toplevel, compile and compile-file.
A load-time form appearing in a form passed to compile-toplevel is translated such that its embedded form will be executed each time the virtual-machine description returned by compile-toplevel is executed, and the execution of all such forms is placed ahead of other code.
A load-time form appearing in an interpreted function which is processed by compile is evaluated immediately, and its value becomes a literal in the compiled version of the function.
A load-time form appearing as a non-top-level form inside a file that is processed by compile-file is compiled along with that form and deposited into the object file. When the object file is loaded, each compiled top-level form is executed. Each compiled top-level form's load-time calculations are executed first, and the corresponding form values become literals at that point. This execution order is individually ensured for each top-level form. Thus, the load-time forms in a given top-level form may rely on the side-effects of prior top-level forms having taken place. Note that, by default, compile-file also immediately executes each top-level form which it compiles and deposits into the output file. This execution is equivalent to a load; it causes load-time forms to be evaluated. The compile-only operator must be used around load-time forms which must be evaluated only when the compiled file is loaded, and not at compile time.
In all situations, the evaluation of form takes place in the global environment. Even if the load-time form is surrounded by constructs which establish lexical bindings, those lexical bindings aren't visible to form. Which dynamic bindings are visible to form depends on the exact situation. If a load-time form occurs in code that had been processed by compile-file and is now being loaded by load, then the dynamic environment in effect is the one in which the load occurred, with any modifications to that environment that were performed by previously executed forms. If a load-time form occurs in code that had been processed by compile-toplevel, then form is evaluated in the dynamic environment of the caller which invokes the execution of the resulting compiled object. When a load-time form occurs in the code of a function being processed by compile, then form is evaluated in the dynamic environment of the caller which invokes compile. If a load-time form occurs in a form processed processed by the evaluator, it is unspecified whether it takes place in the original dynamic environment in which the evaluator was invoked, or whether it is in the dynamic environment of the immediately enclosing form which surrounds the load-time form.
A load-time form may be nested inside another load-time form. In this situation, two cases occur.
If the two forms are not embedded in a lambda, or else are embedded in the same lambda, then the inner load-time form is superfluous due to the presence of the outer load-time. That is to say, the inner (load-time form) expression is equivalent to form, because the outer form already establishes its evaluation to be in a load-time context.
If the inner load-time form occurs in a lambda, but the outer form occurs outside of that lambda, then the semantics of the inner load-time form is relevant and necessary. This is because expressions occurring in a lambda are evaluated when the lambda is called, which may take place from a non-load-time context, even if the lambda itself was produced in a load-time context.
An expression being embedded in a lambda means that it appears either in the lambda body, or else in the parameter list as the initializing expression for an optional parameter.
When interpreted code containing load-time is evaluated, a mutating side effect may take place on the tree structure of that code itself as a result of the load-time evaluation. If that previously evaluated code is subsequently compiled, the compiled translation may be different from compiling the original unevaluated code. Specifically, the compiler may take advantage of the load-time evaluation which had already taken place in the interpreter, and simply take that value, and avoid compiling form entirely. This also has implications on the dynamic environment that is in effect when form is evaluated. If form is evaluated by the interpreter, then it interacts with the dynamic environment which as in effect in that situation; then when the compiler later just takes the result of that evaluation, the compiler's dynamic environment is irrelevant since form isn't being evaluated any more.
If form, when evaluated multiple times, potentially produces a different value on each evaluation, this has implications for the situation when an object produced by compile-toplevel is invoked multiple times. Each time such an object is invoked, the load-time forms are evaluated. If they produce different values, then it appears that the values of literals are changing. All lexical closures derived from the same compiled object share the same literal data. The load function never evaluates a compiled expression more than once. If the same compiled file is loaded more than once, a new compiled object instance is produced from each compiled expression, carrying its own storage area for literals. The compile function also never evaluates a compiled expression more than once; it produces a compiled object, and then executes it once in order to obtain a lexical closure which is returned. Invoking the closure doesn't cause the load-time expressions to be evaluated.
The load-time form is subject to compiler optimizations. A top-level expression is assumed to be evaluated at load time, so load-time does nothing in a top-level expression. It becomes active inside forms embedded in a lambda expressions. Since load-time may be used to hoist calculations outside of loops, load-time is also active in those parts of loops which are repeatedly evaluated.
The use of load-time is similar to defining a variable and then referring to the variable. For instance, a file containing this:
(defvarl a (list 1 2))
(defun f () (cons 0 a))
is similar to
(defun f () (cons 0 (load-time (list 1 2))))
When either file is loaded, in source or compiled form, list expression is evaluated at load time, and then when f is invoked, it retrieves the list.
Both approaches have advantages. The variable-based approach gives the value a name. The semantics of the variable is straightforward. The variable a can easily be assigned a new value. Using its name, the variable can be inspected from the interactive listener. The variable can be referenced from multiple top-level forms directly; it is not a static datum tied to a table of literal values that is tied to a single top-level form. Furthermore, the use of defvar/defvarl versus defparm/defparml controls whether the variable gets replaced with a new value when the file is reloaded.
The advantage of load-time is that it doesn't require a separate top-level form to achieve its load-time effect: the expression is simply nested at the point where it is needed. The load-time form can therefore be generated by macros, whose expansions cannot inject extra top-level forms into the site where they are invoked. If a macro writer would like some form to be evaluated at load time and its value accessible in a macro expansion that appears arbitrarily nested in code, then load-time may provide the path to a straightforward implementation strategy. Access to a load-time value is fast because it doesn't involve referencing through a variable binding; compiled code accesses the value directly via its fixed position in the static data table associated with that code. This advantage is insignificant, however, because access to lexical variables in compiled code is similarly fast, and a value can easily be propagated from a global variable to a lexical for the sake of speed. That said, load-time eliminates that copying step too.
A load-time is also useful when the value is not required, and instead the form produces a useful effect, which should be hoisted to load time. For instance, consider a macro which produces the following expansion:
(progn (load-time (defvar #:g0025)) (other-logic ... #:g0025))
no matter where this expansion is inserted, compile-file and load will ensure that the defvar is executed once, when the compiled file is loaded, as if that defvar appeared on its own as a top-level form. Then the other-logic form can refer to the variable, without the defvar being evaluated on each execution of the progn.
The author of a macro can use load-time to stage the evaluation of global effects that the macro expansion depends on simply by bundling these effects into the expansion, wrapped in load-time.
The load-time macro is similar to the ANSI Common Lisp load-time-value special operator. It doesn't support the read-only-p argument featured in the ANSI CL operator. The semantics of load-time is somewhat more precisely specified in terms of concrete implementation concepts. The ANSI CL load-time-value may evaluate form more than once in interpreted code; effectively, the ANSI CL implementation may treat (load-time-value x) as (progn x). This is not true of TXR Lisp's load-time which requires once-only evaluation even in interpreted code. The name load-time is used instead of load-time-value for several reasons. Firstly, load-time is useful for staging effects, like definitions, to load time, even when the resulting value is not used. Secondly, unlike TXR Lisp, ANSI CL features multiple values: a form can yield zero or more values. The ANSI CL load-time-value operator, however, is restricted to yielding a single value, and its name may have been chosen to emphasize this aspect/restriction. That doesn't apply in the context of TXR Lisp in which all expressions which terminate normally yield exactly one value, making "-value" a suffix that adds no value. Lastly, load-time is shorter, and harmonizes with macro-time, which existed earlier.
(disassemble function-name)
(disassemble function)
(disassemble compiled-expression)
The disassemble function presents a disassembly listing of the virtual-machine code of a compiled function or form. It also presents the literal data contained in that compiled object in a tabular form which is readily cross-referenced with the disassembly listing.
If the argument is a function-name then the function object is retrieved from the binding indicated by the name, in the global namespace. That object is then treated as if it were the function argument.
A function argument is one that is a function object. Only compiled virtual-machine functions can be disassembled; other kinds of functions are rejected by disassemble.
The disassemble function will also process the complied-expression object that is returned by the compile-toplevel function.
In the case of function, the entire compiled form containing function is disassembled. That form usually contains code which is external to the function, even possibly other functions. The disassembly listing indicates the entry point in the code block where the execution of function begins.
The disassemble function returns its argument.
(dump-compiled-objects stream object*)
The dump-compiled-objects function writes compiled objects into stream in the same format as the compile-file function.
Unlike under compile-file, the output is written into an arbitrary stream rather than a named file. The objects aren't specified by the to-be-compiled syntax processed from a source file, but rather as zero or more arguments which specify objects that are already compiled.
Each object must be be one of three kinds of values:
If exactly one call to dump-compiled-objects is used to populate an initially empty file, and no other data are written into the file, then that file is a valid compiled file. If that file is processed by load-file then each of the externalized forms is converted to a virtual-machine description and executed.
Note that virtual-machine descriptions are not functions. A function's virtual-machine description is the compiled version of the top-level form whose evaluation produced that function.
For example, if the following top-level form is compiled and executed, two functions are defined:
(let ()
(defun a ())
(defun b ()))
Then, the following two expressions all have the same effect on stream s:
(dump-compiled-objects s 'a)
(dump-compiled-objects s 'b)
Whether the a or b symbol is used to specify the object to be dumped, the same virtual-machine description is externalized and deposited into the stream. That machine description, when loaded and executed, defines two functions.
On some target platforms, TXR provides an interactive listener, which is invoked using the -i command-line option, or by executing txr with no arguments. The interactive listener provides features like visual editing of the command line, tab completion on TXR Lisp symbols, and history recall.
The interactive listener prints a numbered prompt. The number in the prompt increments with every command. The first command line is numbered 1, the second one 2 and so forth.
The listener accepts input characters from the terminal. Characters are either interpreted as editing commands or other special characters, or else are inserted into the editing buffer. However, control characters which don't correspond to commands are silently rejected.
The carriage return character generated by the Enter key indicates that a complete line has been entered, and it is to be interpreted. The listener parses the line as a TXR Lisp expression, evaluates it, and prints the resulting value. If the evaluation of the line throws an exception, the listener intercepts the exception and prints information about it preceded by two asterisks and a space. These asterisks distinguish an exception from a result value.
If an empty line is entered, or a line containing only spaces, tabs or embedded carriage returns or linefeeds, the prompt is repeated without incrementing the number. Such a line is not entered into the history.
A line which only contains a TXR Lisp comment (optional spaces, tabs or embedded carriage returns or linefeeds, followed by a semicolon), also causes the prompt to be repeated without incrementing the number. However, such a line is entered into the history.
The listener does not allow lines containing certain bad syntax to be submitted with Enter. If the buffer contains an expression with unbalanced parentheses or brackets, or unterminated literals, then Enter generates a newline character which is inserted into the buffer. In that situation, if that newline character is being added at the very end of the buffer, the listener flashes the exclamation mark character (!) two times to warn the user that line has not been submitted: no computation is taking place, and the listener is waiting for more input. It is possible to force the submission of an unbalanced line using the sequence Ctrl-X Ctrl-F.
The interactive listener can only accept up to 4095 abstract characters of input in a single command line.
Though the edit buffer is referred as the "command line", it may contain multiline input. The carriage return characters which separate multiple lines count as one abstract character each, and are understood to occupy two display positions.
Until TXR 286, the command line had to contain exactly one complete TXR Lisp expression, or a comment. Multiple expressions were not evaluated. This restriction has been lifted: multiple expressions in the command line are parsed as one unit, and evaluated as if they were placed into a progn form. If all the expressions evaluate and terminate normally, the value of the last expression is printed.
In multiline mode, if the number of lines exceeds the number of lines of the terminal display, the editing experience is adversely affected in unspecified ways.
The screen updating logic in the listener is based on the assumption that the display terminal uses ANSI emulation. No other terminal emulation is supported. The TERM environment variable is ignored.
Pressing Ctrl-D in a completely empty command line terminates the listener. Another way to quit is to enter the :quit keyword symbol. When the form input into the listener consists of this symbol, the listener will terminate:
1> (+ 2 2)
4
2> :quit
os-shell $
Another way to terminate is to evaluate a call to the exit function. This method allows a termination status to be specified:
1> (exit 1)
os-shell $
However, if a TXR interactive session is terminated this way, it will not save the listener history.
Raising a fatal signal with the raise function is another way to quit:
1> (raise sig-abrt)
Aborted (core dumped)
os-shell $
The previous remark about not saving the listener history applies here also.
Ctrl-C typed while editing a command line is interpreted as an editing command which causes that command line to be canceled. The listener prints the string "** intr" and repeats the same prompt.
If a command line is submitted for evaluation, the evaluation might take a long time or block for input. In these situations, typing Ctrl-C will issue an interrupt signal. The listener has installed a handler for this signal which generates an exception of type error which is caught by the listener. The exception's message is the string "intr" so that the listener ends up printing "intr **" like in the case of the Ctrl-C editing command. In this situation, though, a new command-line prompt is issued with an incremented number, and the exception is recorded as a value.
The listener provides useful variables which allow commands to reference the results of previous commands. As noted previously, the commands are enumerated with an incrementing number. Each command's number, modulo 100, corresponds to one of the variables *0, *1, *2, ..., *99. Thus, up to the previous hundred results can be referenced:
...
99> (+ 2 2) ;; stored in *99
4
100> (* 3 2) ;; stored in *0
6
101> (+ *99 *0) ;; i.e. (+ 4 6)
10
Note: each of these macros expands to a reference to the *r vector, according to the following pattern:
*-1 --> [*r (mod (- *v 1) 100)]
*-2 --> [*r (mod (- *v 2) 100)]
...
*-20 --> [*r (mod (- *v 20) 100)]
The listener variable *n evaluates to the current command-line number: the number of the command in which the variable occurs:
5> *n
5
6> (* 2 *n)
12
The listener variable *v evaluates to the current variable number: the command number modulo 100:
103> *v
3
104> *v
4
The listener variable *r evaluates to a hash table which associates variable numbers with command results:
213> 42
42
214> [*r 13]
42
The result hash allows relative addressing. For instance the expression [*r (mod (pred *v) 100)] refers to the result of the previous command.
The interactive listener catches all exceptions. Each caught exception is associated with the command's variable number, and stored as a value in the appropriate listener variable as well as the *r result hash. Exceptions are turned into values by creating a cons cell whose car is the exception symbol and whose cdr holds the exception's arguments.
For each caught exception, a message is printed beginning with the sequence "** ". Exactly how the message appears depends on the type and content of the exception.
The following sections describe the interactive editing commands available in the listener.
Terminals can often be configured with different choices of cursor shape: such as a block-shaped cursor, an underline cursor or a vertical line or "I-beam" cursor. In the following sections, the phrase "character under the cursor" refers to the character that is currently covered by a block cursor, underlined by an underline cursor, or that is immediately to the right of an I-beam cursor.
Moving within the line is achieved using the left and right arrow keys ← and →. In addition, Ctrl-B ("back") and Ctrl-F ("forward") perform this movement.
The Ctrl-A command moves to the beginning of the line. ("A" is the beginning of the alphabet). The Ctrl-E ("end") command jumps to the end of the line, such that the last character of the line is to the left of the cursor position. On terminals which have the Home and End keys, these may also be used instead of Ctrl-A and Ctrl-E.
In line mode, these commands move the cursor to the beginning or end of the edit buffer.
In multiline mode, if the cursor is not already at the beginning of a physical line, then Ctrl-A moves it to the first character of the physical line. Otherwise, Ctrl-A moves the cursor to the beginning of the edit buffer.
Similarly, in multiline mode, if the cursor not already at the end of a physical line, Ctrl-E moves it there. Otherwise, the cursor moves to the end of the edit buffer.
If the cursor is on an opening or closing parenthesis, brace or bracket, the Ctrl-] command tries to jump to the matching character. The logic for finding the matching character is identical to that of the Parenthesis Matching feature. If no matching character is found, then no movement takes place.
If the cursor is not on an opening or closing parenthesis, brace or bracket, then the closest such character is found. The cursor is moved to that character and then an attempt is made to jump to the matching one from that new position.
If the cursor is equidistant to two such characters, then one of them is chosen as follows. If the two characters are oriented in the same way (both are opening and closing), then that one is chosen whose convex side faces the cursor position. Thus, effectively, an inner enclosure is favored over an outer one. Otherwise, if the two characters have opposite orientation (one is opening and the other closing), then the one which is to the right of the cursor position is chosen.
Note: the Ctrl-] character can be produced on some terminals using Ctrl-5 (using the keyboard home row 5, not the numeric keypad 5). This the same key which produces the % character when Shift is used. The % character is used in the Vi editor for parenthesis matching.
The Ctrl-T (twiddle) command exchanges the character under the cursor with the previous character.
The Backspace key erases the character to the left of the cursor, and moves the cursor to the position which that character occupied.
It doesn't matter whether this key generates ASCII characters 8 (BS) or 127 (DEL): either one is acceptable. The Ctrl-H command also performs the same action, since it corresponds to ASCII BS.
The Ctrl-D command deletes the character under the cursor, if the cursor is block-shaped, or to the right of the cursor if the cursor is an I-beam. the cursor maintains its current character position relative to the start of the line. In multiline mode, if Ctrl-D is at the end of a line that is not the last line, it deletes the newline character, causing the following line to be joined to the end of the current line. If the cursor is at the end of the buffer, then Ctrl-D does nothing, except if the buffer is completely empty, in which case it is a quit indication. The Delete key, if available on the terminal, is a near synonym of Ctrl-D. It performs all the same functions, except that it does not act as a quit indication; Delete has no effect when the buffer is empty.
When a visual selection is in effect, then Ctrl-D and Del delete that selection, and copy it to the clipboard.
The Ctrl-W ("word") command deletes the word to the left of the cursor position. More precisely, this command first deletes any consecutive whitespace characters (spaces or tabs) to the left of the cursor. Then, it deletes consecutive non-whitespace characters. Material under the cursor or to the right remains. The deleted material is copied into the clipboard.
The Ctrl-U ("undo typing") command is a "super backspace" operation: it deletes all characters to the left of the cursor position. The cursor is moved to the leftmost position. In multiline mode, Ctrl-U deletes only to the beginning of the current physical line, not all the way to the first position of the buffer. Ctrl-U copies the deleted material into the clipboard.
The Ctrl-K ("kill") command deletes the character under the cursor position and all subsequent characters. The cursor position doesn't change. In multiline mode, Ctrl-K deletes only until the end of the current physical line, not the entire buffer. The material deleted by Ctrl-K is copied into the clipboard.
The Ctrl-V ("verbatim") command places the listener's input editor into a mode in which the next character is interpreted literally and inserted into the line, even if that character is a special character such as Enter, or a command character.
The Ctrl-X Ctrl-K command sequence may be used in multiline mode to delete the entire physical line under the cursor. Any lines below that line move up to close the gap. In line mode, the command has no effect, other than canceling select mode. The deleted line, including the terminating newline character, if it has one, is copied into the clipboard.
By default, the most recent 500 lines submitted to the interactive listener are remembered in a history. This history is available for recall, making it convenient to repair mistakes, or compose new lines which are based on previous lines. Note that the history suppresses consecutive, duplicate lines. The number of lines retained may be customized using the *listener-hist-len* variable.
If the ↑ key is used while editing a line, the contents of the line are placed into a temporary save area. The line display is then updated to show the most recent line of history. Using ↑ will recall successively less recent lines.
The ↓ key navigates in the opposite direction: from older lines to newer lines. When ↓ is invoked on the most recent history line, then the current line is restored from the temporary save area.
Instead of ↑ and ↓, the commands Ctrl-P ("previous") and Ctrl-N ("next") may be used.
If the Enter key is pressed while a recalled history line is showing, then that line will be submitted as if it were a newly composed line. The originally edited line which had been placed in the save area is discarded.
When a recalled line is showing, it may be edited. There are two important behaviors to note here. If a recalled history line is edited, and then ↑ or ↓ or a navigation command is used to show a different history line, or to restore the original current line, then the edit is made permanent: the edited line replaces its original version in the same position in the history. This feature allows corrections to be made to the history.
The edit is recorded in the line's undo history as a single change; if the edited line is visited again, then a single Ctrl-O command will revert all the edits that were made.
However, if a recalled line is edited and submitted without navigating to another line, then it is submitted as a newly composed line, without replacing the original in the history.
Each submitted line is entered into the history, if it is different from the most recent line already in history. This is true whether it is a freshly composed line, a recalled history line, or an edited history line.
In search mode, characters may be typed. They accumulate inside the search box, and constitute the string to search for. The listener instantly navigates to the most recent line which contains a substring match for the search string, and places the cursor on the first character of the match. Control characters entered directly are ignored. The Ctrl-V command be used to add a character verbatim, as in edit mode.
To remove characters from the search box, Backspace can be used. The search is not repeated with the shortened search text: the same line continues to show until a character is added, at which point a new search is issued.
Search mode has a "home position": a starting point for searches. The initial home position is whatever line of history is selected when search mode is initiated. Searches work backward in history from that line. If search text is edited by deleting characters and then adding new ones, the new search proceeds from the home position.
The Ctrl-R command can be used in search mode. It registers the currently showing line as the new home position, and then repeats the search using the existing search text backwards from the new position. If the search text is empty, Ctrl-R has no effect.
The Ctrl-C command leaves search mode at any time and causes the listener to resume editing the original input at the original character position. The Enter key accepts the result of a search and submits it as if it were a newly composed line.
Navigation and editing keys may be used in search mode. A navigation or editing key immediately cancels search mode, and is processed in edit mode, using whatever line was located by the search, at the matching character position.
The Ctrl-L (Clear Screen and Refresh), as well as Ctrl-Z (Suspend to Background) commands are available in search mode. Their effects takes place without leaving search mode.
Navigating to a history line manually using ↑ or ↓ (or Ctrl-P and Ctrl-N) has the same net effect same as locating that line using Ctrl-R search.
Normally when the Enter key is used on a recalled history line, the next time the listener is reentered, it jumps back to the newest history position where a new line is about to be composed.
The alternative command sequence Ctrl-X Enter provides a useful alternative behavior. After the submitted line is processed, the listener doesn't jump to the newest history position. Instead, it stays in the history, advancing forward by one position to the successor of the submitted line.
Ctrl-X Enter can be used to conveniently submit a range of lines from the history, one by one, in their original order.
The equivalent command sequences Ctrl-X w and Ctrl-X Ctrl-W insert a word from the previous line at the cursor position. A word is defined as a sequence of non-whitespace characters, separated from other words by whitespace. By default, the last word of the previous line is inserted. Between the Ctrl-X and the following Ctrl-W or w, a decimal number can be entered. The number 1 specifies that the last word is to be inserted, 2 specifies the second last word, 3 the third word from the right and so on. Only the most recent three decimal digits are retained, so the number can range from 0 to 999. A value of 0, or a value which exceeds the number of words causes the Ctrl-W or w to do nothing. Note that "previous line" means relative to the current location in the history. If the 42nd most recent history line is currently recalled, this command takes material from the 43rd history line.
The equivalent command sequences Ctrl-X a and Ctrl-X Ctrl-A insert an atom from the previous line at the cursor position. A line only makes atoms available if it expresses a valid TXR form, free of syntax errors. A line containing only whitespace or a comment makes no atoms available. For the purposes of this editing feature, an atom is defined as the printed representation of a Lisp atom taken from the Lisp form specified in the previous line. The line is flattened into atoms as if by the flatcar function. By default, the last atom is extracted. A numeric argument typed between the Ctrl-X and Ctrl-A or a can be used to select a atoms by position from the end. The number 1 specifies the last atom, 2 the second last and so on. Only the most recent three decimal digits are retained, so the number can range from 0 to 999. A value of 0, or a value which exceeds the number of words causes the Ctrl-A or a to do nothing. Note that "previous line" has the same meaning as for the Ctrl-X Ctrl-W (insert previous word) command.
The command sequences Ctrl-X Ctrl-R ("repeat") and Ctrl-X r, which are equivalent, insert an entire line of history into the current buffer. By default, the previous line is inserted. A less recent line can be selected by typing a numeric argument between the Ctrl-X and the Ctrl-R or r. The immediately previous history line is numbered 1, the one before it 2 and so on. If this command is used during history navigation, it references previous lines relative to the currently recalled history line.
If the Tab key is pressed while editing a line, it is interpreted as a request for completion. There is a second completion command: the sequence Ctrl-X Tab.
When completion is invoked with Tab or Ctrl-X Tab, the listener looks at a few of the trailing characters to the left of the cursor position to determine the applicable list of completions. Completions are determined from among the TXR Lisp symbols which have global variable, function, macro and symbolic macro bindings, as well as the static and instance slots of structures. Symbols which have operator bindings are also taken into consideration. If a package-qualified symbol is completed, then completion is restricted to that package. Keyword symbol completion is restricted to the contents of the keyword package. The namespaces which are searched for symbols are restricted according to preceding character syntax. For instance if the characters .( or .[ immediately precede the prefix, then only those symbols are considered which are methods: that is, each is the static slot of at least one structure, in which that static slots holds a function.
The difference between Tab and Ctrl-X Tab is that Tab completion looks only for prefix matches among the eligible identifiers. Thus it is a pure completion in the sense that it suggests additional material that may follow what has been typed. If the buffer contains (list it will only suggest completions which can be endings for list such as list*, listp, and list-str. It will not suggest identifiers which rewrite the list prefix. By contrast, the Ctrl-X Tab completion suggests not only pure completions but also alternatives to the partial identifier, by looking for substring matches. For instance copy-list is a possible completion for list, as is proper-list-p.
If no completions are found, then the BEL character is sent to the terminal to generate a beep or a visual alert indication. The listener returns to editing mode.
If completions are found, listener enters into completion selection mode. The first available completion is placed into the line as if it had been typed in. The other completions may be viewed one by one using the Tab key. (Note that the Ctrl-X is not used, only Tab, even if completion mode had been entered via Ctrl-X Tab). When the completions are exhausted, the original uncompleted line is shown again, and Tab can continue to be used to cycle through the completions again. In completion mode, the Ctrl-C character acts as a command to cancel completion mode and return to editing the original uncompleted line. Any other input character causes the listener to keep the currently shown completion, and return to edit mode, where that character is immediately processed as if it had been typed in edit mode.
The two character command Ctrl-X Ctrl-E launches an external editor to edit the current command line. The command line is stored in a temporary file first, and the editor is invoked on this file. When the editor terminates, the file is read into the editing buffer.
The editor is determined from the EDITOR environment variable. If this variable is unset or empty, the command does nothing.
The temporary file is created in the home directory, if that can be determined. Otherwise it is created in the current working directory. If the creation of the file fails, then the command silently returns to edit mode. The home directory is determined from the HOME environment variable in POSIX environments. On MS Windows, the USERPROFILE variable is probed for the user's directory.
If the command line contains embedded carriage returns (which denote line breaks in multiline mode) these are replaced with newline characters when written out to the file. Conversely, when the edited file is read back, its newlines are converted to carriage returns, so that multiline content is handled properly. (See the following section, Multiline Mode.)
The listener provides an undo feature. The Ctrl-O command ("old", "oops") restores the edit buffer contents and cursor position to a previous state.
There is a single undo history which records up the 200 most recent edit states. However, the states are associated with history lines, so that it appears that each line has its own, independent undo history. Undoing the edits in one line has no effect on the undo history of another line.
Undo also records edits for lines that have been canceled with Ctrl-C and are not entered into the history, making it possible to recall canceled lines.
The undo history is lost when TXR terminates.
Undo doesn't save and restore previous contents of the clipboard buffer.
There is no redo. When undo removes an edit to restore to a prior edit state, the removed edit is permanently discarded.
Note that if undo is invoked on a historic line, each undo step updates that history entry instantly to the restored state, not only the visible edit buffer. This is in contrast to the way new edits work. New edits are not committed to history until navigation takes place to a different history line.
Also note that when new edits are performed on a historic line and it is submitted with Enter without navigating to another line, the undo information for those edits is retained, and belongs to the newly submitted line. The historic line hasn't actually been modified, and so it has no new undo information. However, if a historic line is edited, and then navigation takes place to a different historic line, then the undo information is committed to that line, because the modifications to the line have been placed back in the history entry.
The interactive listener supports visual copy and paste operation. Text may be visually selected for copying into a clipboard or or for deletion. In visual selection mode, the actions of some editing commands are modified so that they act upon the selection instead of their usual target, or upon both the target and the selection.
The Ctrl-S command enters into visual selection mode and marks the starting point of the selection, which is considered the position immediately to the left of the current character.
While in visual selection mode, it is possible to move around using the usual movement commands. The ending point of the selection tracks the movement.
The selected text is displayed in reverse video.
Typing Ctrl-S again while in visual selection mode cancels the mode.
Tab completion, history navigation, history search and editing in an external editor all cancel visual selection mode.
By default, the selection excludes the character which lies to the right of the rightmost endpoint. Thus, the selection simply consists of the text between these two positions, whether or not they are reversed. This style of selection pairs excellently with an I-beam style cursor, and has clear semantics. The endpoints are referenced to the positions between the characters, and everything between them is selected.
The selection behavior may be altered using the Boolean configuration variable *listener-sel-inclusive-p*. This variable is nil by default. If it is changed to true, then the selection includes the character to the right of the rightmost endpoint, if there is such a character within the current line. This style of selection pair well with a block-shaped cursor. It creates the apparent semantics that the endpoints of the selection are characters, rather than points between characters, and that these characters are included in the selection.
In visual selection, the starting point of the selection remains fixed, while the ending point tracks the movement of the cursor. The Ctrl-^ command will exchange the two points. The effect is that the cursor jumps to the opposite end of the selection. That end is now the ending point which tracks the cursor movement.
The Ctrl-Y command ("yank") copies the selected text into a clipboard buffer. The previous contents of the clipboard buffer, if any, are discarded.
Unlike the history, the clipboard buffer is not persisted. If TXR terminates, it is lost.
If the Ctrl-D command is invoked while a selection is in effect, then instead of deleting the character under the cursor, it deletes the selection, and copies it to the clipboard. The Delete key has the same effect.
Ctrl-D and Del have no effect on the clipboard when visual selection is not in effect, and they operate on just one character.
The Ctrl-Q command ("quote the clipboard") inserts text from the clipboard at the current cursor position. The cursor position is updated to be immediately after the inserted text. The clipboard text remains available for further pasting.
If nothing has been yet been copied to the clipboard in the current session, then this command has no effect.
The Ctrl-X Ctrl-Q command sequence ("exchange quote") exchanges the selected text with the contents of the clipboard. The selection is copied into the clipboard as if by Ctrl-Y and replaced by the previous contents of the clipboard.
If the clipboard has not yet been used in the current session,
If nothing has been yet been copied to the clipboard in the current session, then this command behaves like Ctrl-Y: text is yanked into the clipboard, but not deleted.
In visual selection mode, an editing commands may be used which insert new text, or a character may be typed in order to insert it. When this happens, the selection is first deleted and visual mode is canceled. Then the insertion takes place and visual mode is canceled. The effect is that the newly inserted text replaces the selected text.
This applies to the Clipboard Paste Ctrl-Q command also. If a selection is effect when Ctrl-Q is invoked, the selected text is replaced with the clipboard buffer contents.
When a selection is replaced in this manner, the contents of the clipboard are unaffected.
In visual mode, it is possible to issue commands which delete text.
One such command is Ctrl-D. Its special behavior in selection mode, Visual Cut, is described above.
The Backspace key and Ctrl-H also have a special behavior in select mode. If the cursor is at the rightmost endpoint of the selection, then these commands delete the selection and nothing else. If the cursor is at the leftmost endpoint of the selection, then these commands delete the selection, and take their usual effect of deleting a character also. In both cases, selection mode is canceled. The clipboard is not affected.
The Ctrl-W command for deleting the previous word, when used in visual selection mode, deletes the selection and cancels selection mode, and then deletes the word before the selection. Only the deleted selection is copied into the clipboard, not the deleted word.
All other deletion commands such as Ctrl-K simply cancel visual selection mode and take their usual effect.
The listener operates in one of two modes: line mode and multiline mode. This is determined by the special variable *listener-multi-line-p* whose default value is t (multiline mode). It is possible to toggle between line mode and multiline mode using the Ctrl-J command.
In line mode, all input given to a single prompt appears to be on a single line. When the line becomes longer than the screen width, it scrolls horizontally. In line mode, carriage return characters embedded in a line are displayed as ^M.
In multiline mode, when the input exceeds the screen width, it simply wraps to take up additional lines rather than scrolling horizontally. Furthermore, multiline mode not only wraps long lines of input onto multiple lines of the display, but also supports true multiline input. In multiline mode, carriage return characters embedded in input are treated as line breaks rather than being rendered as ^M.
Because carriage returns are not line terminators in text files, lines which contain embedded carriage returns are correctly saved into and retrieved from the persistent history file.
When Enter is typed in multiline mode, the listener tries to determine whether the current input, taken as a whole, is an incomplete expression which requires closing punctuation for elements like compound expressions and string literals.
If the input appears incomplete, then the Enter is inserted verbatim at the current cursor position, rather than signaling that the line is being submitted for evaluation. The Ctrl-X Enter command sequence also has this behavior.
In addition to multiline mode, the listener provides support for directly parsing input from the terminal, suitable for processing large amounts of pasted material.
If the :read keyword is entered into the listener, it will temporarily suspend interactive editing and allow the TXR Lisp parser to read directly from standard input. The reading stops when an error occurs, or EOF is indicated by entering Ctrl-D.
In direct parsing mode, each expression which is read is evaluated, but its value is not printed. However, the value of the last form evaluated is returned to the interactive listener, which prints the value and accepts it as if it as the result value of the :read command.
Note that none of the material read from the terminal is entered into the interactive history. Only the :read command which triggers this parsing mode appears in the history.
The Ctrl-L command clears the screen and redraws the line being edited. This is useful when the display is disturbed by the output of some background process, or serial line noise.
The Ctrl-Z ("Zzzz... (sleep)") command causes TXR to be placed into the background in a suspended, and control returned to the system shell.
Bringing the suspended TXR back into the foreground is achieved with a shell job-control command such as the fg command in GNU Bash.
When TXR is resumed, the interactive listener will redisplay the edited line and restore the previous cursor position.
Making full use of this feature requires a POSIX job control shell, in the sense that without job control support in the shell, there may not be a way to restore TXR into the terminal session's foreground, causing the user to lose interactive control over that TXR instance.
The Ctrl-X ? command shows a summary of commands, in a four-line display which temporarily replaces the editing area. The help text is divided into several pages. Ctrl-C dismisses the display, and returns to editing. The Ctrl-P, ← and ↑ keys return to the previous screen. The Ctrl-Z and Ctrl-L commands are available, having their usual meaning of suspending and refreshing the display. Any other key advances to the next screen. Advancing from the last screen, dismisses the display, and returns to editing. Navigating to the previous screen when the first screen is being shown also dismisses the display and returns to editing.
The :prompt command prints the current prompt, followed by a newline, without incrementing the prompt number. The :p command prints just the current prompt number, followed by a newline, without incrementing the number.
In plain mode, the :prompt-on command enables the printing of prompts. The full prompt is printed before reading each new command line. An abbreviated prompt is printed before reading the continuation lines of an incomplete expression. The printing of prompts is automatically enabled if the input device is an interactive terminal.
None of these prompt-related commands are entered into the history.
When the input device isn't an interactive terminal, or if the -n or --noninteractive command-line operations are used when invoking TXR, the listener operates in plain mode. It reads input without providing any of the editing features of visual mode: no completion, history recall, selection, or copy and paste. Only the line editing features provided by the operating system are available. Prompts appear if standard input is an interactive terminal, or if explicitly enabled. There is still an incrementing counter, and the numbered variables *1, *2, ... for accessing evaluation results are established. Lines are still entered into the history, and the interactive profile is still processed, as usual.
Plain mode reads whole lines of input, yet recognizes multi-line expressions. Whenever a line of input is read which represents incomplete syntax, another line of input is read and appended to that line. This repeats until the accumulated input represents complete syntax, and is then processed as a unit.
Like in visual mode, each unit of input may contain multiple expressions. These are parsed as a unit and evaluated as if they were the elements of a progn expression. The resulting value which is printed is that of the last expression.
Unless the --noprofile option has been used, when the listener starts up, it looks for file called .txr_profile in the user's home directory, as determined by the HOME environment variable in POSIX environments or the USERPROFILE environment variable on MS Windows. If that variable doesn't exist, no further attempt is made to locate this file.
If the file exists, it is subject to security checks. First, the path-components-safe is applied to its path name. The function validates that no component of the path name is a directory that is writable to another user, or a symbolic link that could be rewritten by another user. If that check passes, the file is then checked with the function path-strictly-private-to-me-p which requires that other users have no read or write permission. If the checks fail, then an error message is displayed and the file is not loaded.
If the file passes the security check, it is expected to be readable and to contain TXR Lisp forms, which are read and evaluated. Syntax errors encountered while reading the profile file are displayed on standard output, and any exceptions thrown that are derived from error are caught and displayed. The interactive listener starts in spite of these situations. Exceptions not derived from error will terminate the process.
The profile file is not read by noninteractive invocations of TXR: that is, when the -i option isn't present.
The history is maintained in a text file called .txr_history in the user's home directory. Whenever the interactive listener terminates, this file is updated with the history contents stored in the listener's memory. The next time the listener starts, it first reloads the history from this file, making the most recent *listener-hist-len* expressions of a previous session available for recall.
The history file is maintained in a way that is somewhat robust against the loss of history arising from the situation that a user manages multiple simultaneous TXR sessions. When a session terminates, it doesn't blindly overwrite the history file, which may have already been updated with new history produced by another session. Rather, it appends new entries to the history file. New entries are those that had not been previously read from the history file, but have been newly entered into the listener.
An effort is made to keep the history file trimmed to no more than twice the number of entries specified in *listener-hist-len*. The terminating session first makes a temporary copy of the existing history, which is trimmed to the most recent *listener-hist-len* entries. New entries are then appended to this temporary file. Finally, the actual history file is replaced with this temporary file by a rename-path a rename operation. This algorithm doesn't use locking, and is therefore not robust against the situation when a two or more multiple interactive TXR sessions belonging to the same user terminate at around the same time.
The home directory is determined from the contents of the HOME environment variable in POSIX environments or USERPROFILE on MS Windows. If this variable doesn't exist, or the user doesn't have permissions to write to this directory or to an existing history file in that directory, then the history isn't saved.
It is possible to save the history without terminating the interactive session, using the :save command. This saves the history in the manner described above. Each invocation of :save only adds to the history file new input since the most recent :save command.
When the history file is loaded, security checks take place, in exactly the same way that the ".txr_profile" file is validated. First the path of the history file is checked using the function path-components-safe, which determines that no component of the path name can be subverted by another user, other than the superuser. If that check passes, then the file is checked using path-strictly-private-to-me-p which requires that other users have no read or write permission. If the checks fail, then an error message is displayed and the history file is not loaded.
A feature of the listener is visual parenthesis matching in the form of a brief forward or backward jump of the cursor. This provides a hint to the programmer, helping to prevent avoid parenthesis balancing errors.
When any of the three closing characters ), ] or } is inserted, the listener scans backward for the matching opening character. Likewise, if any of the three opening characters (, [ or { is inserted in the middle of text, the listener scans forward for the matching closing character.
If the matching character is found, the cursor jumps to that character and then returns to the original position a brief moment later. If a new character is typed during the brief time delay, the delay is immediately canceled, so as not to hinder rapid typing.
This back-and-forth jump behavior also occurs when a character is erased using Backspace, and the cursor ends up immediately to the right of a parenthesis.
Note that the matching is unsophisticated; it doesn't observe the lexical conventions and syntax of the TXR Lisp programming language. For instance, a closing parenthesis outside a string literal may match match an opening one inside a string literal.
The listener's behavior can be influenced through values of certain global variables. The settings can be made persistent by means of setting these variables in the interactive profile file.
This special variable determines how many lines of history are retained by the listener. Changing this variable from within the listener has an instant effect. If the number is reduced from its current value, history lines are immediately discarded. The default value is 500.
This is a Boolean variable which indicates whether the listener is in multiline mode. The default value is nil.
Changing this variable from within the listener takes effect immediately for the next line of input.
If multiline mode is toggled interactively from within the listener, the variable is updated to reflect the latest state. This happens when the command is submitted for evaluation.
This Boolean variable controls the behavior of visual selection. It is nil by default.
A visual selection is determined by endpoints, which are abstract positions understood as being between characters. When a visual selection begins, it marks an endpoint immediately to the left of a block-shaped cursor, or precisely at the in-between position of an I-beam cursor. The end of the visual selection is similarly determined from the ending cursor position. The selection consists of those characters which lie between these positions. This style of selection pairs well with an I-beam style cursor shape.
If the *listener-sel-inclusive-p* variable is set true, then the selection also includes one more character to the right of the rightmost endpoint, if there is such a character within the current line, giving rise to the appearance that the selection is determined by the starting and ending character, and includes them. This type of selection pairs well with a block-shaped cursor.
This Boolean variable controls how the listener prints the results of evaluations. It is nil by default.
When the variable is nil, the evaluation result of each line entered into the listener is printed using the prinl function. Thus values are rendered in a machine-readable syntax, ensuing read/print consistency.
If the variable is set true, the evaluation result of each line is printed using the pprinl function.
The special variable *listener-greedy-eval-p* controls whether or not a "greedy evaluation" feature is enabled in the listener. The default value is nil, disabling the feature.
Greedy evaluation means that after the listener evaluates the input expressions successfully and prints the value of the last one, it then checks whether that value is an expression that may be further subject to nontrivial evaluation. If so, it evaluates that expression, and prints the resulting value. The process is then repeated with the resulting value. It keeps repeating until evaluation throws an error, or produces a self-evaluating object.
These additional evaluations are performed in such a way that all warnings are suppressed and all other exceptions are intercepted.
Greedy evaluation doesn't affect the state of the listener. Only the original expression is entered into the history. Only the value of the original expression is saved in the result hash or a numbered variable. The command-line number *n is incremented by one. The additional evaluations are only performed for the purpose of producing useful output. The evaluations may have side effects.
1> (set *listener-greedy-eval-p* t)
t
2> 'a
a
3> (defvar b 2)
b
2
4> (defvar c '(+ 2 2))
c
(+ 2 2)
4
5> (defvar d '(list '+ 2 2))
d
(list '+ 2 2)
(+ 2 2)
4
The (defvar d ...) form produces d symbol as its result value. That symbol has a variable binding as a result of that defvar and so evaluates; that evaluation produces (list '+ 2 2), the contents of d. That object is a Lisp expression and is evaluated, producing (+ 2 2) and that is also an expression, which reduces to 4. The object 4 is self-evaluating, and so the greedy evaluation process stops.
The special variable *listener-auto-compound-p* controls whether or the listener is operating in "auto compound expression" mode. The default value is nil, disabling the feature.
Normally, an input line can contain multiple expressions, which are treated as if they were combined into a single expression by progn. Thus all the expressions are evaluated, and the value from the last one is printed.
In auto compound mode, the behavior changes. An input line which consists of multiple expressions is turned into a compound form whose constituents are those items. Thus, for instance, the input + 2 2 is treated as the compound expression (+ 2 2) resulting in 4 being calculated.
When a single expression is input, it is evaluated as-is, and thus in that case auto compound expression mode makes no difference.
The special variable *doc-url* holds a character string representing a web URL intended to point to the HTML version of this document. The initial value points to the publicly hosted document on the Internet. The user may change this to point to another location, such as a locally hosted copy of the document.
This variable is used by the doc function.
(doc [symbol])
The doc function provides help for the library symbol symbol. If information about symbol is available in the HTML version of this document, and is indexed, then this function causes that document to be opened using a web browser, such that the browser navigates to the appropriate section of the manual.
If the symbol argument is omitted, then the document is opened without navigating to a particular section.
The base URL for the document is configured by the *doc-url* variable.
If symbol is successfully found, or else not specified, and doc successfully invokes the URL-opening mechanism, it returns t. Otherwise, it throws an error exception.
The web browser is invoked using a system-dependent strategy. On MS Windows, the ShellExecuteW function is relied upon to open the URL.
On other platforms, if the BROWSER environment variable exists and is nonempty, its value is assumed to indicate the name or path of the web-browsing program which can accept the URL as an argument. If this variable doesn't exist or is empty, then doc searches for a system-dependent URL-opening utility, such as xdg-open. If this utility is not found, then doc falls back to searching for a browser using one of several names. If no URL-opening mechanism is identified using the above strategies, an error exception is thrown. However, if the mechanism is identified, but does not successfully dispatch the URL to a browser, there is no requirement to throw an error exception. It may appear that the doc function returns t but has no effect.
(quip)
The quip function returns a randomly selected string containing a humorous quip, quote or witticism. The following code may be added to .txr_profile to produce a random quip on startup:
The quip function was introduced in TXR 244. If the .txr_profile is used with installations of older TXR versions, it is recommended to use the following, to avoid calling the undefined function, as well as to prevent a warning:
(if (fboundp 'quip)
(put-line (quip))
(defun quip ()))
On platforms with the Unix filesystem and process security model, TXR has support for executing setuid/setgid scripts, even on platforms whose operating system kernel does not honor the setuid/setgid bit on hash-bang scripts. On these systems, taking advantage of the feature requires TXR to be installed as a setuid/setgid executable. For this reason, TXR is aware when it is executed setuid and takes care to manage privileges. The following description about the handling of setuid applies to the parallel handling of setgid also.
When TXR starts, early in its execution it determines whether or not is is executing setuid. If so, it temporarily drops privileges, as a precaution. This is done before processing the command-line arguments. When TXR determines that it is executing a setuid script (a file marked executable to its owner and attributed with the set-user-ID bit), it then attempts to impersonate the owner of the script file by changing to effective user ID to that owner just before executing the file. It retains the real and saved user ID. If the attempt to assume that user ID is unsuccessful, then TXR permanently drops setuid privileges before executing the script. Likewise, before executing any code other than a setuid script, TXR also drops privileges.
TXR tries to honor and implement the setuid permissions on a script whether or not it is running setuid. When not running setuid, it nevertheless tries to change its effective user ID to that of the owner of the setuid script. This will succeed if it has sufficient permissions to do so.
To rephrase: in order for TXR to execute a file which is setuid root, it has to be running with a root effective user ID somehow. In order to execute a file which is setuid to a non-root user, TXR has to be running effectively as root or else as that user. It doesn't matter whether these privileges are achieved effectively using the setuid mechanism, or whether TXR is running with the required user ID as its real ID. However, if TXR is running setuid, it takes special care to temporarily drop the privileges as early as possible, and eventually to drop the privileges permanently before executing any code, other that the setuid script. If the setuid script cannot be executed with the privileges it calls for, TXR also drops privileges and executes it anyway, strictly as the real user who invoked the TXR executable.
What it means to drop privileges is to change the effective user ID and the saved user ID to be equal to the real user ID. On platforms where the setresuid function is available, TXR uses that function to drop privileges. On platforms where setresuid is not available, TXR tries to drop privileges using the C language function call setuid(r), where r is the previously noted real user ID obtained from getuid(). On some platforms, this only works for dropping root privileges: it overwrites the real and saved ID only if the caller is effectively root. On those platforms, this approach does not drop non-root privileges. TXR tries to detect whether this approach worked by evaluating the C language expression seteuid(e), where e is the previously noted effective user ID. In other words, it attempts to regain the dropped privilege by recovering the previous effective ID. If this attempt succeeds, TXR immediately aborts. Dropping setgid privileges is similar. Where setresgid is available it is used, otherwise an attempt is made with setegid(r) where r is the previously noted real group ID. Then a test using setegid(e) is performed using the original effective group ID as e. This is done after dropping any setuid root user ID privilege which would allow such a test to succeed.
If TXR is running both setuid and setgid, and execute a script which is setuid only, it will still drop group privileges, and vice versa: if it executed a setgid script, it will drop user privileges. For instance, if a root-owned TXR runs a setgid script which is owned by user 10 and group-owned by group 20, that script will run with an effective group ID of 20. The effective user ID will be that of the user who invoked the script: TXR will drop the root privilege to the original real ID of the user, and while for the setgid operation, it will change to the group ID of the script.
The setuid/setgid privilege machinery in TXR does not manipulate the list of supplementary ("ancillary", in the language of POSIX) group IDs. It is unnecessary for security because the list does not change while running with setuid privilege. No group IDs are added to the list which need to be retracted when privileges are dropped. The supplementary groups also persist across the execution of a setuid/setgid script.
The TXR executable image supports a general mechanism by means of which a custom program can be packaged as an apparent standalone executable.
The TXR executable contains a 128 byte data area preceded by the seven-byte ASCII character sequence "@(txr):". The 128 byte data area which follows this identifying prefix represents a null-terminated UTF-8 string. In the stock executable, this area is filled with null bytes.
If the TXR executable is edited such that this area is replaced with a nonempty, null-terminated UTF-8 string, the program will, for the purposes of command-line-argument processing, treat this string as if it were the one and only command-line argument. (The original command-line arguments are still retained in the *args* and *args-full* variables).
The function save-exe creates a copy of the TXR executable with a custom internal argument.
Suppose that TXR is copied to an executable in the same directory called myapp (or myapp.exe on an operating system which requires the .exe suffix). Also suppose that in the same directory, there exists a file called myscript.tl.
This myapp executable can then be edited so that the data area which follows the @(txr): bytes contains the following string:
--args|-e|(load (path-cat (dir-name txr-exe-path) "main.tl"))
When the myapp executable is invoked, it will process the above string as a single command-line argument, causing the main.tl TXR Lisp source file to be loaded. Any arguments passed to myapp are ignored and available to main.tl via the *args* variable.
The TXR executable may require library files, depending on the functionality invoked by the program code. Library files are located relative to the installation directory, called the sysroot. The executable tries to dynamically determine the sysroot from its own location, according to the following directory structure. The executable may be renamed, it need not be called txr:
/path/to/sysroot/bin/txr
.../share/txr/stdlib/cadr.tl
.../stdlib/cadr.tlo
.../stdlib/except.tl
...
.../share/txr/lib/...
The above structure is assumed if the executable finds itself in a directory named "bin".
Otherwise, if the executable finds itself in a directory not named "bin", the following structure is expected:
/path/to/installation/txr
.../stdlib/cadr.tl
.../stdlib/cadr.tlo
.../stdlib/except.tl
...
.../lib/...
The "lib/" directory shown above is for third-party libraries. This is the directory indicated in the default value of the *load-search-dirs* special variable. The directory is not required to exist.
Note that this structure had changed starting in TXR 264. Older versions of TXR, when the executable is not in a directory named "bin", expect the following structure:
/path/to/installation/txr
.../share/txr/stdlib/cadr.tl
.../share/txr/stdlib/cadr.tlo
.../share/txr/stdlib/except.tl
...
When a custom application is deployed using a possibly renamed txr executable, one of the above structures should be observed: either the sysroot with a bin subdirectory where the executable is located, on the same level with the share directory, or else the second structure in which the stdlib directory is a direct subdirectory of the executable directory. If one of these structures is not observed, the application may fail due to the failure of a library file to load.
If the executable discovers that its name ends in the suffix "lisp" (or else "lisp.exe" on the MS Windows platform) then the behavior is as if the --lisp command line option had been given. Similarly, if the executable finds that its name ends in "vm" (or "vm.exe" on MS Windows) it behaves as if the --compiled option had been given.
(save-exe path arg-string)
The save-exe function produces an edited copy of the TXR executable at the specified path, inserting arg-string as the internal argument string.
In order for the copied executable to be useful, the required installation directory structure must be provided around it, as described in the previous section, Deployment Directory Structure.
The return value of save-exe is unspecified.
The arg-string should encode to 127 bytes of UTF-8 or less, or else it will be abruptly truncated, possibly in the middle of a UTF-8 sequence.
Create a copy of TXR called myapp which will load a file called main.tl that is located in the same directory.
(save-exe
"myapp"
"--args|-e|(load (path-cat (dir-name txr-exe-path) \
\ \"main.tl\"))")
TXR had a simple, crude, built-in debugger, which was removed.
New TXR versions are usually intended to be backward-compatible with prior releases in the sense that documented features will continue to work in the same way. Due to new features, new versions of TXR will supply new behaviors where old versions of TXR would have produced an error, such as a syntax error. Though, strictly speaking, this means that something is working differently in a new version, replacing an error situation with functionality is usually not considered a deviation from backward-compatibility.
There is one notable deviation from this general requirement for backwards compatibility: the handling of compiled files. For pragmatic reasons, from time to time TXR may break backward compatibility, such that a newer version of TXR will not load compiled files produced by an older version. The files will have to be recompiled with the new TXR. More details are given in the section Compiled File Compatibility under the major section LISP COMPILATION. The rationale for not requiring backward compatibility support for older compiled files is that older files require the older implementation of the virtual machine which they target. In some cases the differences between the older virtual machine and new is so great that TXR would have to carry a whole separate virtual-machine implementation for the sake of the older files, which is a significant burden.
When a change is introduced which is not backward compatible, TXR's -C option can be used to request emulation of old behavior.
The option was introduced in TXR 98, and so the oldest TXR version which can be emulated is TXR 97.
Side effects occur in the processing of the option. If the option is specified multiple times, the behavior is unspecified.
If the TXR_COMPAT environment variable exists, and its value is not en empty string, it must contain a decimal integer. Its value is taken by TXR as a request to emulate old behaviors, just like the value of the -C option.
If the variable has incorrect contents or an out-of-range value, TXR will print an error diagnostic and exit.
If both -C and the TXR_COMPAT environment variable are supplied, the behavior is unspecified.
The following version values which have a special meaning as arguments to the -C option, along with a description of what behaviors are affected. For each of these version values, the described behaviors are provided if -C is given an argument which is equal or lower. For instance -C 103 selects the behaviors described below for version 105, but not those for 102.
./txr: (file.txr:123): syntax error
The new format omits the program name prefix and parentheses.
Also, the kill function returned an integer, obtained from the return value of the underlying C function, rather than converting that value to a Boolean. The old behavior was not documented, and 114 compatibility restores it.
Lastly, prior to 115, random state objects were of type *random-state* (the same symbol as the special variable name) rather than of type random-state. This is a bug whose behavior is simulated by 114 compatibility.
Also allows unrecognized backslash escape sequences in regular expression syntax to simply denote the escaped character literally, as was historically the case prior to TXR 106, so that \z for instance denotes z. As of TXR 106, these are diagnosed as errors.
The txr-version variable gives the version of the TXR executable. Programs can express conditional variable based on detecting the version.
The lib-version variable gives the version of the installed library of TXR code accompanying the executable.
It is expected that these two variables have an identical value. Any discrepancy in their value indicates an installation whose library or TXR executable were upgraded independently. Should such a situation arise in any system and cause a problem, TXR programs can be defensively coded against it with the help of these variables.
Some features of the TXR library are built into the executable, whereas others are in the library directory. This aspect of library symbols isn't specified in this manual; knowing which of these two variables is relevant to a library feature requires familiarity with the implementation.
Users familiar with regular expressions may not be familiar with the complement and intersection operators, which are often absent from text processing tools that support regular expressions. The following remarks are offered in the hope that they may be of some use.
Regexp intersection is not essential; it may be obtained from complement and union as follows, since De Morgan's law applies to regular-expression algebra: (R1)&(R2) = ~(~(R1)|~(R2)). (The complement of the union of the complements of R1 and R2 constitutes the intersection.) This law works because the regular expression operators denote set operations in a straightforward way. A regular expression denotes a set of strings (a potentially infinite one) in a condensed way. The union of two regular expressions R1|R2 denotes the union of the set of texts denoted by R1 and that denoted by R2. Similarly R1&R2 denotes a set intersection, and ~R denotes a set complement. Thus algebraic laws that apply to set operations apply to regular expressions. It's useful to keep in mind this relationship between regular expressions and sets in understanding intersection and complement.
Given a finite set of strings, like the set { "abc", "def" } which corresponds to the regular expression (abc|def), the complement is the set which contains an infinite number of strings: it consists of all possible strings except "abc" and "def". It includes the empty string, all strings of length 1, all strings of length 2, all strings of length 3 other than "abc" and "def", all strings of length 4, etc. This means that a "harmless looking" expression like ~(abc|def) can actually match arbitrarily long inputs.
How about matching only three-character-long strings other than "abc" or "def"? To express this, regex intersection can be used: these strings are the intersection of the set of all three-character strings, and the set of all strings which are not "abc" or "def". The straightforward set-based reasoning leads us to this: ...&~(abc|def). This A&~B idiom is also called set difference, sometimes notated with a minus sign: A-B (which is not supported in TXR regular-expression syntax). Elements which are in the set A, but not B, are those elements which are in the intersection of A with the complement of B. This is similar to the arithmetic rule A - B = A + -B: subtraction is equivalent to addition of the additive inverse. Set difference is a useful tool: it enables us to write a positive match which captures a more general set than what is intended (but one whose regular expression is far simpler than a positive match for the exact set we want), then we can intersect this over-generalized set with the complemented set of another regular expression which matches the particulars that we wish excluded.
It turns out that regular expressions which do not make use of the complement or intersection operators are just as powerful as expressions that do. That is to say, with or without these operators, regular expressions can match the same sets of strings (all regular languages). This means that for a given regular expression which uses intersection and complement, it is possible to find a regular expression which doesn't use these operators, yet matches the same set of strings. But, though they exist, such equivalent regular expressions are often much more complicated, which makes them difficult to design. Such expressions do not necessarily express what it is they match; they merely capture the equivalent set. They perform a job, without making it obvious what it is they do. The use of complement and intersection leads to natural ways of expressing many kinds of matching sets, which not only demonstrate the power to carry out an operation, but also easily express the concept.
For instance, using complement, we can write a straightforward regular expression which matches C language comments. A C language comment is the digraph /*, followed by any string which does not contain the closing sequence */, followed by that closing sequence. Examples of valid comments are /**/, /* abc */ or /***/. But C comments do not nest (cannot contain comments), so that /* /* nested */ */ actually consists of the comment /* /* nested */, which is followed by the trailing junk */. Our simple characterization of the interior part of a C comment as a string which does not contain the terminating digraph makes use of the complement, and can be expressed using the complemented regular expression like this: (~.*[*][/].*). That is to say, strings which contain */ are matched by the expression .*[*][/].*: zero or more arbitrary characters, followed by */, followed by zero or more arbitrary characters. Therefore, the complement of this expression matches all other strings: those which do not contain */. These strings make up the inside of a C comment between the /* and */.
The equivalent simple regex is quite a bit more complicated. Without complement, we must somehow write a positive match for all strings such that we avoid matching */. Obviously, sequences of characters other than * are included: [^*]*. Occurrences of * are also allowed, but only if followed by something other than a slash, so let's include this via union:
([^*]|[*][^/])*.
Alas, we already have a bug in this expression. The subexpression [*][^/] can match **, since a * is not a /. If the next character in the input is /, we missed a comment close. To fix the problem we revise to this:
([^*]|[*][^*/])*
(The interior of a C language comment is any mixture of zero or more non-asterisks, or digraphs consisting of an asterisk followed by something other than a slash or another asterisk). Oops, now we have a problem again. What if two asterisks occur in a comment? They are not matched by [^*], and they are not matched by [*][^*/]. Actually, our regex must not simply match asterisk-non-asterisk digraphs, but rather sequences of one or more asterisks followed by a non-asterisk:
([^*]|[*]*[^*/])*
This is still not right, because, for instance, it fails to match the interior of a comment which is terminated by asterisks, including the simple test cases where the comment interior is nothing but asterisks. We have no provision in our expression for this case; the expression requires all runs of asterisks to be followed by something which is not a slash or asterisk. The way to fix this is to add on a subexpression which optionally matches a run of zero or more interior asterisks before the comment close:
([^*]|[*]*[^*/])*[*]*
Thus the semi-final regular expression is
[/][*]([^*]|[*]*[^*/])*[*]*[*][/]
(Interpretation: a C comment is an interior string enclosed in /* */, where this interior part consists of a mixture of non-asterisk characters, as well as runs of asterisk characters which are terminated by a character other than a slash, except for possibly one rightmost run of asterisks which extends to the end of the interior, touching the comment close. Phew!) One final simplification is possible: the tail part [*]*[*][/] can be reduced to [*]+[/] such that the final run of asterisks is regarded as part of an extended comment terminator which consists of one or more asterisks followed by a slash. The regular expression works, but it's cryptic; to someone who has not developed it, it isn't obvious what it is intended to match. Working out complemented matching without complement support from the language is not impossible, but it may be difficult and error-prone, possibly requiring multiple iterations of trial-and-error development involving numerous test cases, resulting in an expression that doesn't have a straightforward relationship to the original idea.
The non-greedy operator % is actually defined in terms of a set difference, which is in turn based on intersection and complement. The uninteresting case (R%) where the right operand is empty reduces to (R*): if there is no trailing context, the non-greedy operator matches R as far as possible, possibly to the end of the input, exactly like the greedy operator. The interesting case (R%T) is defined as a "syntactic sugar" which expands to the expression ((R*)&(~.*(T&.+).*))T which means: match the longest string which is matched by R*, but which does not contain a non-empty match for T; then, match T. This is a useful and expressive notation. With it, we can write the regular expression for matching C language comments simply like this: [/][*].%[*][/] (match the opening sequence /*, then match a sequence of zero or more characters non-greedily, and then the closing sequence */. With the non-greedy operator, we don't have to think about the interior of the comment as set of strings which excludes */. Though the non-greedy operator appears expressive, its apparent simplicity may be deceptive. It looks as if it works "magically" by itself; "somehow" this .% part "knows" only to consume enough characters so that it doesn't swallow an occurrence of the trailing context. Care must be taken that the trailing context passed to the operator really is the correct text that should be excluded by the non-greedy match. For instance, take the expression .%abc. If you intend the trailing context to be merely a, you must be careful to write (.%a)bc. Otherwise, the trailing context is abc, and this means that the .% match will consume the longest string that does not contain abc, when in fact what was intended was to consume the longest string that does not contain a. The change in behavior of the % operator upon modifying the trailing context is not as intuitive as that of the * operator, because the trailing context is deeply involved in its logic.
On a related note, for single-character trailing contexts, it may be a good idea to use a complemented character class instead. That is to say, rather than (.%a)bc, consider [^a]*abc. The set of strings which don't contain the character a is adequately expressed by [^a]*.
Topics matching >:
Topics matching exp2:
Topics matching put-buf:
Topics matching hash:
Topics matching require:
Topics matching log:
Topics matching width:
Topics matching load-time:
Topics matching logtrunc:
Topics matching put-byte:
Topics matching errno:
Topics matching -:
Topics matching exp10:
Topics matching *:
Topics matching abs:
Topics matching gather:
Topics matching iter-reset:
Topics matching bool:
Topics matching buf-d:
Topics matching in-package:
Topics matching fill-buf:
Topics matching erfc:
Topics matching print:
Topics matching trunc:
Topics matching clear-error:
Topics matching /:
Topics matching coll:
Topics matching if:
Topics matching time-parse:
Topics matching tgamma:
Topics matching round:
Topics matching length:
Topics matching nearbyint:
Topics matching sinh:
Topics matching struct:
Topics matching oddp:
Topics matching rplaca:
Topics matching >=:
Topics matching <:
Topics matching bind:
Topics matching put-string:
Topics matching tofloat:
Topics matching put-char:
Topics matching not:
Topics matching acos:
Topics matching all:
Topics matching log2:
Topics matching toint:
Topics matching get-error:
Topics matching nullify:
Topics matching y0:
Topics matching flatten:
Topics matching close:
Topics matching cbrt:
Topics matching defex:
Topics matching iter-step:
Topics matching acosh:
Topics matching accept:
Topics matching sign-extend:
Topics matching slot:
Topics matching get-char:
Topics matching get-line:
Topics matching collect:
Topics matching y1:
Topics matching length-<:
Topics matching set:
Topics matching log1p:
Topics matching signum:
Topics matching equal:
Topics matching tanh:
Topics matching cdr:
Topics matching or:
Topics matching asinh:
Topics matching =:
Topics matching qref:
Topics matching elemtype:
Topics matching static-slot:
Topics matching gamma:
Topics matching push:
Topics matching <=:
Topics matching ash:
Topics matching erf:
Topics matching file-put-jsons:
Topics matching exp:
Topics matching throw:
Topics matching lognot:
Topics matching load:
Topics matching rep:
Topics matching iter-begin:
Topics matching block:
Topics matching significand:
Topics matching bitset:
Topics matching atanh:
Topics matching square:
Topics matching do:
Topics matching cos:
Topics matching cptr:
Topics matching tan:
Topics matching none:
Topics matching some:
Topics matching trailer:
Topics matching iter-item:
Topics matching cosh:
Topics matching repeat:
Topics matching mod:
Topics matching get-byte:
Topics matching minusp:
Topics matching get-error-str:
Topics matching sqrt:
Topics matching atan:
Topics matching unget-char:
Topics matching unget-byte:
Topics matching sin:
Topics matching exptmod:
Topics matching bit:
Topics matching stat:
Topics matching rplacd:
Topics matching ceil:
Topics matching until:
Topics matching isqrt:
Topics matching catch:
Topics matching logb:
Topics matching logior:
Topics matching expt:
Topics matching atan2:
Topics matching +:
Topics matching lgamma:
Topics matching logand:
Topics matching awk:
Topics matching log10:
Topics matching finally:
Topics matching dwim:
Topics matching zerop:
Topics matching merge:
Topics matching del:
Topics matching next:
Topics matching lambda:
Topics matching assert:
Topics matching j0:
Topics matching floor:
Topics matching str:
Topics matching rint:
Topics matching time:
Topics matching and:
Topics matching evenp:
Topics matching buf:
Topics matching file-put-json:
Topics matching plusp:
Topics matching copy:
Topics matching static-slot-set:
Topics matching car:
Topics matching iter-more:
Topics matching expm1:
Topics matching logcount:
Topics matching last:
Topics matching j1:
Topics matching asin: