At this moment there was a loud ring at the bell, and I could hear Mrs. Hudson, our landlady, raising her voice in a wail of expostulation and dismay."By heaven, Holmes," I said, half rising, "I believe that they are really after us."
"No, it's not quite so bad as that. It is the unofficial force, -- the Baker Street irregulars."
(load "irregex.scm")
in your favorite Scheme implementation and you're good to go!
There is a global variable *all-chars* which is used for
generating character set complements. This defaults to the full
Unicode range 0..#x10FFFF, but if your implementation can't handle
characters that large you'll need to adjust it (a suitable ASCII
definition is commented out in the source).
If you are using an R6RS Scheme, you can instead
(load "irregex-r6rs.scm")
There are also a handful of utility procedures described below you may wish to use in irregex-utils.scm.
If you are using Chicken Scheme IrRegex is built in as a core unit, so
no need to install it. To use it, you just need to (use irregex).
(rx ...) syntax - just use normal Scheme lists, with
quasiquote if you like.
Technically a string by itself could be considered a valid (though
rather silly) SRE, so if you want to just match a literal string you
should use something like (irregex `(: ,str)), or use the explicit
(sre->irregex str).
The options are a list of any of the following symbols:
'i, 'case-insensitive - match case-insensitively
'm, 'multi-line - treat string as multiple lines (effects ^ and $)
's, 'single-line - treat string as a single line (. can match newline)
'utf8 - utf8-mode (assumes strings are byte-strings)
'fast - try to optimize the regular expression
'small - try to compile a smaller regular expression
'backtrack - enforce a backtracking implementation
The 'fast and 'small options are heuristic guidelines and will
not necessarily make the compiled expression faster or smaller.
maybe-string->sre does the same thing, but only if the argument is
a string, otherwise it assumes <obj> is an SRE and returns it
as-is. This is useful when you want to provide an API that allows
either a POSIX string or SRE (like irregex or irregex-search
below) - it ensures the result is an SRE.
#t iff the object is a regular expression.
#f.
Match objects can be used to query the original range of the string or
its submatches using the irregex-match-* procedures below.
Examples:
(irregex-search "foobar" "abcFOOBARdef") => #f
(irregex-search (irregex "foobar" 'i) "abcFOOBARdef") => #<match>
(irregex-search '(w/nocase "foobar") "abcFOOBARdef") => #<match>
Note, the actual match result is represented by a vector in the
default implementation. Throughout this document, we'll just write
<match> to show that a successful match was returned when the
details are not important.
Matching follows the POSIX leftmost, longest semantics, when
searching. That is, of all possible matches in the string,
irregex-search will return the match at the first position
(leftmost). If multiple matches are possible from that same first
position, the longest match is returned.
irregex-search, but performs an anchored match against the
beginning and end of the substring specified by <start> and <end>,
without searching.
Examples:
(irregex-match '(w/nocase "foobar") "abcFOOBARdef") => #f
(irregex-match '(w/nocase "foobar") "FOOBAR") => #<match>
#t iff the object is a successful match result from
irregex-search or irregex-match.
car of each item in this list is
the name of a submatch, the cdr of each item is the numerical
submatch corresponding to this name. If a named submatch occurs
multiple times in the irregex, it will also occur multiple times in
this list.
#t iff the index-or-name named submatch or index is
defined in the match object.
<replacement> can be either a string
literal, a numeric index, a symbol (as a named submatch), or a
procedure which takes one argument (the match object) and returns a
string.
Examples:
(irregex-replace "[aeiou]" "hello world" "*") => "h*llo world"
(irregex-replace/all "[aeiou]" "hello world" "*") => "h*ll* w*rld"
irregex-split splits the string <str> into substrings divided
by the pattern in <irx>. irregex-extract does the opposite,
returning a list of each instance of the pattern matched disregarding
the substrings in between.
<irx> occurs in the string str.
The <kons> procedure takes the following signature:
(<kons> <from-index> <match> <seed>)
where <from-index> is the index from where we started searching
(initially <start> and thereafter the end index of the last
match), <match> is the resulting match-data object, and <seed>
is the accumulated fold result starting with <knil>.
The rationale for providing the <from-index> (which is not
provided in the SCSH regexp-fold utility), is because this
information is useful (e.g. for extracting the unmatched portion of
the string before the current match, as needed in
irregex-replace), and not otherwise directly accessible.
The optional <finish> takes two arguments:
(<finish> <from-index> <seed>)
which simiarly allows you to pick up the unmatched tail of the string,
and defaults to just returning the <seed>.
<start> and <end> are numeric indices letting you specify the
boundaries of the string on which you want to fold.
To extract all instances of a match out of a string, you can use
(map irregex-match-substring (irregex-fold <irx> (lambda (i m s) (cons m s)) '() <str> (lambda (i s) (reverse s))))
The following table summarizes the SRE syntax, with detailed explanations following.
;; basic patterns <string> ; literal string (seq <sre> ...) ; sequence (: <sre> ...) (or <sre> ...) ; alternation ;; optional/multiple patterns (? <sre> ...) ; 0 or 1 matches (* <sre> ...) ; 0 or more matches (+ <sre> ...) ; 1 or more matches (= <n> <sre> ...) ; exactly <n> matches (>= <n> <sre> ...) ; <n> or more matches (** <from> <to> <sre> ...) ; <n> to <m> matches (?? <sre> ...) ; non-greedy (non-greedy) pattern: (0 or 1) (*? <sre> ...) ; non-greedy kleene star (**? <from> <to> <sre> ...) ; non-greedy range ;; submatch patterns (submatch <sre> ...) ; numbered submatch ($ <sre> ...) (submatch-named <name> <sre> ...) ; named submatch (=> <name> <sre> ...) (backref <n-or-name>) ; match a previous submatch ;; toggling case-sensitivity (w/case <sre> ...) ; enclosed <sre>s are case-sensitive (w/nocase <sre> ...) ; enclosed <sre>s are case-insensitive ;; character sets <char> ; singleton char set (<string>) ; set of chars (or <cset-sre> ...) ; set union (~ <cset-sre> ...) ; set complement (i.e. [^...]) (- <cset-sre> ...) ; set difference (& <cset-sre> ...) ; set intersection (/ <range-spec> ...) ; pairs of chars as ranges ;; named character sets any nonl ascii lower-case lower upper-case upper alphabetic alpha numeric num alphanumeric alphanum alnum punctuation punct graphic graph whitespace white space printing print control cntrl hex-digit xdigit ;; assertions and conditionals bos eos ; beginning/end of string bol eol ; beginning/end of line bow eow ; beginning/end of word nwb ; non-word-boundary (look-ahead <sre> ...) ; zero-width look-ahead assertion (look-behind <sre> ...) ; zero-width look-behind assertion (neg-look-ahead <sre> ...) ; zero-width negative look-ahead assertion (neg-look-behind <sre> ...) ; zero-width negative look-behind assertion (atomic <sre> ...) ; for (?>...) independent patterns (if <test> <pass> [<fail>]) ; conditional patterns commit ; don't backtrack beyond this (i.e. cut) ;; backwards compatibility (posix-string <string>) ; embed a POSIX string literal
(irregex-search "needle" "hayneedlehay") => #<match>
By default the match is case-sensitive, though you can control this either with the compiler flags or local overrides:
(irregex-search "needle" "haynEEdlehay") => #f
(irregex-search (irregex "needle" 'i) "haynEEdlehay") => #<match>
(irregex-search '(w/nocase "needle") "haynEEdlehay") => #<match>
You can use w/case to switch back to case-sensitivity inside a
w/nocase or when the SRE was compiled with 'i:
(irregex-search '(w/nocase "SMALL" (w/case "BIG")) "smallBIGsmall") => #<match>
(irregex-search '(w/nocase "small" (w/case "big")) "smallBIGsmall") => #f
Important: characters outside the ASCII range are only matched case insensitively if the host Scheme system natively supports UTF8 in strings.
Of course, literal strings by themselves aren't very interesting
regular expressions, so we want to be able to compose them. The most
basic way to do this is with the seq operator (or its abbreviation
:), which matches one or more patterns consecutively:
(irregex-search '(: "one" space "two" space "three") "one two three") => #<match>
As you may have noticed above, the w/case and w/nocase
operators allowed multiple SREs in a sequence - other operators that
take any number of arguments (e.g. the repetition operators below)
allow such implicit sequences.
To match any one of a set of patterns use the or alternation
operator:
(irregex-search '(or "eeney" "meeney" "miney") "meeney") => #<match>
(irregex-search '(or "eeney" "meeney" "miney") "moe") => #f
? which just optionally
matches the pattern:
(irregex-search '(: "match" (? "es") "!") "matches!") => #<match>
(irregex-search '(: "match" (? "es") "!") "match!") => #<match>
(irregex-search '(: "match" (? "es") "!") "matche!") => #f
To optionally match any number of times, use *, the Kleene star:
(irregex-search '(: "<" (* (~ #\>)) ">") "<html>") => #<match>
(irregex-search '(: "<" (* (~ #\>)) ">") "<>") => #<match>
(irregex-search '(: "<" (* (~ #\>)) ">") "<html") => #f
Often you want to match any number of times, but at least one time is
required, and for that you use +:
(irregex-search '(: "<" (+ (~ #\>)) ">") "<html>") => #<match>
(irregex-search '(: "<" (+ (~ #\>)) ">") "<a>") => #<match>
(irregex-search '(: "<" (+ (~ #\>)) ">") "<>") => #f
More generally, to match at least a given number of times, use >=:
(irregex-search '(: "<" (>= 3 (~ #\>)) ">") "<table>") => #<match>
(irregex-search '(: "<" (>= 3 (~ #\>)) ">") "<pre>") => #<match>
(irregex-search '(: "<" (>= 3 (~ #\>)) ">") "<tr>") => #f
To match a specific number of times exactly, use =:
(irregex-search '(: "<" (= 4 (~ #\>)) ">") "<html>") => #<match>
(irregex-search '(: "<" (= 4 (~ #\>)) ">") "<table>") => #f
And finally, the most general form is ** which specifies a range
of times to match. All of the earlier forms are special cases of this.
(irregex-search '(: (= 3 (** 1 3 numeric) ".") (** 1 3 numeric)) "192.168.1.10") => #<match>
(irregex-search '(: (= 3 (** 1 3 numeric) ".") (** 1 3 numeric)) "192.0168.1.10") => #f
There are also so-called "non-greedy" variants of these repetition
operators, by convention suffixed with an additional ?. Since the
normal repetition patterns can match any of the allotted repetition
range, these operators will match a string if and only if the normal
versions matched. However, when the endpoints of which submatch
matched where are taken into account (specifically, all matches when
using irregex-search since the endpoints of the match itself matter),
the use of a non-greedy repetition can change the result.
So, whereas ? can be thought to mean "match or don't match,"
?? means "don't match or match." * typically consumes as much
as possible, but *? tries first to match zero times, and only
consumes one at a time if that fails. If you have a greedy operator
followed by a non-greedy operator in the same pattern, they can
produce surprisins results as they compete to make the match longer or
shorter. If this seems confusing, that's because it is. Non-greedy
repetitions are defined only in terms of the specific backtracking
algorithm used to implement them, which for compatibility purposes
always means the Perl algorithm. Thus, when using these patterns you
force IrRegex to use a backtracking engine, and can't rely on
efficient execution.
or alternation pattern on a
list of single-character strings to simulate a character set, but this
is too clumsy for everyday use so SRE syntax allows a number of
shortcuts.
A single character matches that character literally, a trivial character class. More conveniently, a list holding a single element which is a string refers to the character set composed of every character in the string.
(irregex-match '(* #\-) "---") => #<match>
(irregex-match '(* #\-) "-_-") => #f
(irregex-match '(* ("aeiou")) "oui") => #<match>
(irregex-match '(* ("aeiou")) "ouais") => #f
Ranges are introduced with the / operator. Any strings or
characters in the / are flattened and then taken in pairs to
represent the start and end points, inclusive, of character ranges.
(irregex-match '(* (/ "AZ09")) "R2D2") => #<match>
(irregex-match '(* (/ "AZ09")) "C-3PO") => #f
In addition, a number of set algebra operations are provided. or,
of course, has the same meaning, but when all the options are
character sets it can be thought of as the set union operator. This
is further extended by the & set intersection, - set
difference, and ~ set complement operators.
(irregex-match '(* (& (/ "az") (~ ("aeiou")))) "xyzzy") => #<match>
(irregex-match '(* (& (/ "az") (~ ("aeiou")))) "vowels") => #f
(irregex-match '(* (- (/ "az") ("aeiou"))) "xyzzy") => #<match>
(irregex-match '(* (- (/ "az") ("aeiou"))) "vowels") => #f
(irregex-search '(: "foo" eow) "foo") => #<match>
(irregex-search '(: "foo" eow) "foo!") => #<match>
(irregex-search '(: "foo" eow) "foof") => #f
The bow, bol, eol, bos and eos work similarly.
nwb asserts that you are not in a word-boundary - if replaced for
eow in the above examples it would reverse all the results.
There is no wb, since you tend to know from context whether it
would be the beginning or end of a word, but if you need it you can
always use (or bow eow).
Somewhat more generally, Perl introduced positive and negative look-ahead and look-behind patterns. Perl look-behind patterns are limited to a fixed length, however the IrRegex versions have no such limit.
(irregex-search '(: "regular" (look-ahead " expression"))
"regular expression")
=> #<match>
The most general case, of course, would be an and pattern to
complement the or pattern - all the patterns must match or the
whole pattern fails. This may be provided in a future release,
although it (and look-ahead and look-behind assertions) are unlikely
to be compiled efficiently.
In some cases, but not always, these will overlap. When they are
different, irregex-search will naturally always want the searching
version, so IrRegex provides that version.
As an example where these might be different, consider a URL. If you want to match all the URLs in some arbitrary text, you probably want to exclude a period or comma at the tail end of a URL, since it's more likely being used as punctuation rather than part of the URL, despite the fact that it would be valid URL syntax.
Another problem with the RFC definitions is the standard itself may have become irrelevant. For example, the pattern IrRegex provides for email addresses doesn't match quoted local parts (e.g. "first last"@domain.com) because these are increasingly rare, and unsupported by enough software that it's better to discourage their use. Conversely, technically consecutive periods (e.g. first..last@domain.com) are not allowed in email addresses, but most email software does allow this, and in fact such addresses are quite common in Japan.
The current patterns provided are:
newline ; general newline pattern (crlf, cr, lf) integer ; an integer real ; a real number (including scientific) string ; a "quoted" string symbol ; an R5RS Scheme symbol ipv4-address ; a numeric decimal ipv4 address ipv6-address ; a numeric hexadecimal ipv6 address domain ; a domain name email ; an email address http-url ; a URL beginning with https?://
Because of these issues the exact definitions of these patterns are subject to be changed, but will be documented clearly when they are finalized. More common patterns are also planned, but as what you want increases in complexity it's probably better to use a real parser.
Unicode character classes (\P) are not supported, but will be in an upcoming release. \C named characters are not supported.
Callbacks, subroutine patterns and recursive patterns are not supported. (*FOO) patterns are not supported and may never be.
\G and \K are not supported.
Octal character escapes are not supported because they are ambiguous with back-references - just use hex character escapes.
Other than that everything should work, including named submatches, zero-width assertions, conditional patterns, etc.
In addition, \< and \> act as beginning-of-word and end-of-word marks, respectively, as in Emacs regular expressions.
Also, two escapes are provided to embed SRE patterns inside PCRE strings, "\'<sre>" and "(*'<sre>)". For example, to match a comma-delimited list of integers you could use
"\\'integer(,\\'integer)*"
and to match a URL in angle brackets you could use
"<('*http-url)>"
Note in the second example the enclosing "('*...)" syntax is needed because the Scheme reader would consider the closing ">" as part of the SRE symbol.
The following chart gives a quick reference from PCRE form to the SRE equivalent:
;; basic syntax "^" ;; bos (or eos inside (?m: ...)) "$" ;; eos (or eos inside (?m: ...)) "." ;; nonl "a?" ;; (? a) "a*" ;; (* a) "a+" ;; (+ a) "a??" ;; (?? a) "a*?" ;; (*? a) "a+?" ;; (+? a) "a{n,m}" ;; (** n m a) ;; grouping "(...)" ;; (submatch ...) "(?:...)" ;; (: ...) "(?i:...)" ;; (w/nocase ...) "(?-i:...)" ;; (w/case ...) "(?<name>...)" ;; (=> <name>...) ;; character classes "[aeiou]" ;; ("aeiou") "[^aeiou]" ;; (~ "aeiou") "[a-z]" ;; (/ "az") or (/ "a" "z") "[[:alpha:]]" ;; alpha ;; assertions "(?=...)" ;; (look-ahead ...) "(?!...)" ;; (neg-look-ahead ...) "(?<=...)" ;; (look-behind ...) "(?<!...)" ;; (neg-look-behind ...) "(?(test)pass|fail)" ;; (if test pass fail) "(*COMMIT)" ;; commit
IrRegex provides a chunked string API specifically for this purpose. You define a chunking API with
(<get-next> chunk) => returns the next chunk, or #f if there are no more chunks
(<get-string> chunk) => a string source for the chunk
(<get-start> chunk) => the start index of the result of <get-string> (defaults to always 0)
(<get-end> chunk) => the end (exclusive) of the string (defaults to string-length of the source string)
(<get-substring> cnk1 i cnk2 j) => a substring for the range between the chunk cnk1 starting at index i and ending at cnk2 at index j
(<get-subchunk> cnk1 i cnk2 j) => as above but returns a new chunked data type instead of a string (optional)
There are two important constraints on the <get-next> procedure.
It must return an eq? identical object when called multiple times
on the same chunk, and it must not return a chunk with an empty string
(start == end). This second constraint is for performance reasons -
we push the work of possibly filtering empty chunks to the chunker
since there are many chunk types for which empty strings aren't
possible, and this work is thus not needed. Note that the initial
chunk passed to match on is allowed to be empty.
<get-substring> is provided for possible performance improvements
- without it a default is used. <get-subchunk> is optional -
without it you may not use irregex-match-subchunk described above.
You can then match chunks of these types with the following procedures:
Example:
To match against a simple, flat list of strings use:
(define (rope->string rope1 start rope2 end) (if (eq? rope1 rope2) (substring (car rope1) start end) (let loop ((rope (cdr rope1)) (res (list (substring (car rope1) start)))) (if (eq? rope rope2) (string-concatenate-reverse ; from SRFI-13 (cons (substring (car rope) 0 end) res)) (loop (cdr rope) (cons (car rope) res)))))) (define rope-chunker (make-irregex-chunker (lambda (x) (and (pair? (cdr x)) (cdr x))) car (lambda (x) 0) (lambda (x) (string-length (car x))) rope->string)) (irregex-search/chunked <pat> rope-chunker <list-of-strings>)
Here we are just using the default start, end and substring behaviors, so the above chunker could simply be defined as:
(define rope-chunker (make-irregex-chunker (lambda (x) (and (pair? (cdr x)) (cdr x))) car))
irregex-fold.
regexp-opt. Note this optimization
doesn't help when irregex is able to build a DFA.
0.7 - chunked string API (DONE)
0.8 - utilities and API finalization (DONE)
0.9 - refactoring, implementation-specific performance enhancements (DONE)
1.0 - cleanup and better documentation
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. 3. The name of the author may not be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
[2] Russ Cox Implementing Regular Expressions
[3] Russ Cox Henry Spencer's Tcl Regex Library
[4] Olin Shivers Proposed SRE regular-expression notation
[5] Olin Shivers Pattern-matching strings with regular expressions
[6] Shiro Kawai Gauche Scheme - Regular Expressions
[7] Damian Conway Perl6 Exegesis 5 - Regular Expressions
[8] Philip Hazel PCRE - Perl Compatible Regular Expressions
[9] Ville Laurikari NFAs with Tagged Transitions, their Conversion to Deterministic Automata and Application to Regular Expressions
[10] Ville Laurikari Efficient submatch addressing for regular expressions