This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
update docs for $`, $&, $' changes
[perl5.git] / pod / perlre.pod
CommitLineData
a0d0e21e 1=head1 NAME
d74e8afc 2X<regular expression> X<regex> X<regexp>
a0d0e21e
LW
3
4perlre - Perl regular expressions
5
6=head1 DESCRIPTION
7
5d458dd8 8This page describes the syntax of regular expressions in Perl.
91e0c79e 9
cc46d5f2 10If you haven't used regular expressions before, a quick-start
91e0c79e
MJD
11introduction is available in L<perlrequick>, and a longer tutorial
12introduction is available in L<perlretut>.
13
14For reference on how regular expressions are used in matching
15operations, plus various examples of the same, see discussions of
16C<m//>, C<s///>, C<qr//> and C<??> in L<perlop/"Regexp Quote-Like
17Operators">.
cb1a09d0 18
0d017f4d
WL
19
20=head2 Modifiers
21
19799a22 22Matching operations can have various modifiers. Modifiers
5a964f20 23that relate to the interpretation of the regular expression inside
19799a22 24are listed below. Modifiers that alter the way a regular expression
5d458dd8 25is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
1e66bd83 26L<perlop/"Gory details of parsing quoted constructs">.
a0d0e21e 27
55497cff
PP
28=over 4
29
54310121 30=item m
d74e8afc 31X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff
PP
32
33Treat string as multiple lines. That is, change "^" and "$" from matching
1e4aa4a5
AP
34the start or end of line only at the left and right ends of the string to
35matching them anywhere within the string.
55497cff 36
54310121 37=item s
d74e8afc
ITB
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
55497cff
PP
40
41Treat string as single line. That is, change "." to match any character
19799a22 42whatsoever, even a newline, which normally it would not match.
55497cff 43
34d67d80 44Used together, as C</ms>, they let the "." match any character whatsoever,
fb55449c 45while still allowing "^" and "$" to match, respectively, just after
19799a22 46and just before newlines within the string.
7b8d334a 47
87e95b7f
YO
48=item i
49X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
50X<regular expression, case-insensitive>
51
52Do case-insensitive pattern matching.
53
5027a30b
KW
54If locale matching rules are in effect, the case map is taken from the
55current
17580e7a 56locale for code points less than 255, and from Unicode rules for larger
ed7efc79
KW
57code points. However, matches that would cross the Unicode
58rules/non-Unicode rules boundary (ords 255/256) will not succeed. See
59L<perllocale>.
60
61There are a number of Unicode characters that match multiple characters
62under C</i>. For example, C<LATIN SMALL LIGATURE FI>
63should match the sequence C<fi>. Perl is not
64currently able to do this when the multiple characters are in the pattern and
65are split between groupings, or when one or more are quantified. Thus
66
67 "\N{LATIN SMALL LIGATURE FI}" =~ /fi/i; # Matches
68 "\N{LATIN SMALL LIGATURE FI}" =~ /[fi][fi]/i; # Doesn't match!
69 "\N{LATIN SMALL LIGATURE FI}" =~ /fi*/i; # Doesn't match!
70
71 # The below doesn't match, and it isn't clear what $1 and $2 would
72 # be even if it did!!
73 "\N{LATIN SMALL LIGATURE FI}" =~ /(f)(i)/i; # Doesn't match!
74
1f59b283
KW
75Perl doesn't match multiple characters in an inverted bracketed
76character class, which otherwise could be highly confusing. See
77L<perlrecharclass/Negation>.
78
25275596
KW
79Another bug involves character classes that match both a sequence of
80multiple characters, and an initial sub-string of that sequence. For
81example,
82
83 /[s\xDF]/i
84
85should match both a single and a double "s", since C<\xDF> (on ASCII
86platforms) matches "ss". However, this bug
87(L<[perl #89774]|https://rt.perl.org/rt3/Ticket/Display.html?id=89774>)
88causes it to only match a single "s", even if the final larger match
89fails, and matching the double "ss" would have succeeded.
90
1f59b283 91Also, Perl matching doesn't fully conform to the current Unicode C</i>
ed7efc79
KW
92recommendations, which ask that the matching be made upon the NFD
93(Normalization Form Decomposed) of the text. However, Unicode is
94in the process of reconsidering and revising their recommendations.
87e95b7f 95
54310121 96=item x
d74e8afc 97X</x>
55497cff
PP
98
99Extend your pattern's legibility by permitting whitespace and comments.
ed7efc79 100Details in L</"/x">
55497cff 101
87e95b7f
YO
102=item p
103X</p> X<regex, preserve> X<regexp, preserve>
104
632a1772 105Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
87e95b7f
YO
106${^POSTMATCH} are available for use after matching.
107
e2e6bec7
DN
108=item g and c
109X</g> X</c>
110
111Global matching, and keep the Current position after failed matching.
112Unlike i, m, s and x, these two flags affect the way the regex is used
113rather than the regex itself. See
114L<perlretut/"Using regular expressions in Perl"> for further explanation
115of the g and c modifiers.
116
b6fa137b
FC
117=item a, d, l and u
118X</a> X</d> X</l> X</u>
119
516074bb
KW
120These modifiers, all new in 5.14, affect which character-set semantics
121(Unicode, etc.) are used, as described below in
ed7efc79 122L</Character set modifiers>.
b6fa137b 123
55497cff 124=back
a0d0e21e 125
516074bb
KW
126Regular expression modifiers are usually written in documentation
127as e.g., "the C</x> modifier", even though the delimiter
b6fa137b 128in question might not really be a slash. The modifiers C</imsxadlup>
ab7bb42d 129may also be embedded within the regular expression itself using
ed7efc79 130the C<(?...)> construct, see L</Extended Patterns> below.
b6fa137b 131
ed7efc79
KW
132=head3 /x
133
b6fa137b 134C</x> tells
7b059540 135the regular expression parser to ignore most whitespace that is neither
55497cff 136backslashed nor within a character class. You can use this to break up
4633a7c4 137your regular expression into (slightly) more readable parts. The C<#>
54310121 138character is also treated as a metacharacter introducing a comment,
55497cff 139just as in ordinary Perl code. This also means that if you want real
14218588 140whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a 141class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
142escape them (using backslashes or C<\Q...\E>) or encode them using octal,
143hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
144making Perl's regular expressions more readable. Note that you have to
145be careful not to include the pattern delimiter in the comment--perl has
146no way of knowing you did not intend to close the pattern early. See
147the C-comment deletion code in L<perlop>. Also note that anything inside
7651b971 148a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
0b928c2f 149space interpretation within a single multi-character construct. For
7651b971 150example in C<\x{...}>, regardless of the C</x> modifier, there can be no
9bb1f947 151spaces. Same for a L<quantifier|/Quantifiers> such as C<{3}> or
f9e949fd
KW
152C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<?> and C<:>,
153but can between the C<(> and C<?>. Within any delimiters for such a
154construct, allowed spaces are not affected by C</x>, and depend on the
155construct. For example, C<\x{...}> can't have spaces because hexadecimal
156numbers don't have spaces in them. But, Unicode properties can have spaces, so
0b928c2f 157in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
9bb1f947 158L<perluniprops/Properties accessible through \p{} and \P{}>.
d74e8afc 159X</x>
a0d0e21e 160
ed7efc79
KW
161=head3 Character set modifiers
162
163C</d>, C</u>, C</a>, and C</l>, available starting in 5.14, are called
164the character set modifiers; they affect the character set semantics
165used for the regular expression.
166
808432af
KW
167The C</d>, C</u>, and C</l> modifiers are not likely to be of much use
168to you, and so you need not worry about them very much. They exist for
169Perl's internal use, so that complex regular expression data structures
170can be automatically serialized and later exactly reconstituted,
171including all their nuances. But, since Perl can't keep a secret, and
172there may be rare instances where they are useful, they are documented
173here.
ed7efc79 174
808432af
KW
175The C</a> modifier, on the other hand, may be useful. Its purpose is to
176allow code that is to work mostly on ASCII data to not have to concern
177itself with Unicode.
ca9560b2 178
808432af
KW
179Briefly, C</l> sets the character set to that of whatever B<L>ocale is in
180effect at the time of the execution of the pattern match.
ca9560b2 181
808432af
KW
182C</u> sets the character set to B<U>nicode.
183
184C</a> also sets the character set to Unicode, BUT adds several
185restrictions for B<A>SCII-safe matching.
186
187C</d> is the old, problematic, pre-5.14 B<D>efault character set
188behavior. Its only use is to force that old behavior.
189
190At any given time, exactly one of these modifiers is in effect. Their
191existence allows Perl to keep the originally compiled behavior of a
192regular expression, regardless of what rules are in effect when it is
193actually executed. And if it is interpolated into a larger regex, the
194original's rules continue to apply to it, and only it.
195
196The C</l> and C</u> modifiers are automatically selected for
197regular expressions compiled within the scope of various pragmas,
198and we recommend that in general, you use those pragmas instead of
199specifying these modifiers explicitly. For one thing, the modifiers
200affect only pattern matching, and do not extend to even any replacement
201done, whereas using the pragmas give consistent results for all
202appropriate operations within their scopes. For example,
203
204 s/foo/\Ubar/il
205
206will match "foo" using the locale's rules for case-insensitive matching,
207but the C</l> does not affect how the C<\U> operates. Most likely you
208want both of them to use locale rules. To do this, instead compile the
209regular expression within the scope of C<use locale>. This both
210implicitly adds the C</l> and applies locale rules to the C<\U>. The
211lesson is to C<use locale> and not C</l> explicitly.
212
213Similarly, it would be better to use C<use feature 'unicode_strings'>
214instead of,
215
216 s/foo/\Lbar/iu
217
218to get Unicode rules, as the C<\L> in the former (but not necessarily
219the latter) would also use Unicode rules.
220
221More detail on each of the modifiers follows. Most likely you don't
222need to know this detail for C</l>, C</u>, and C</d>, and can skip ahead
223to L<E<sol>a|/E<sol>a (and E<sol>aa)>.
ca9560b2 224
ed7efc79
KW
225=head4 /l
226
227means to use the current locale's rules (see L<perllocale>) when pattern
228matching. For example, C<\w> will match the "word" characters of that
229locale, and C<"/i"> case-insensitive matching will match according to
230the locale's case folding rules. The locale used will be the one in
231effect at the time of execution of the pattern match. This may not be
232the same as the compilation-time locale, and can differ from one match
233to another if there is an intervening call of the
b6fa137b 234L<setlocale() function|perllocale/The setlocale function>.
ed7efc79
KW
235
236Perl only supports single-byte locales. This means that code points
237above 255 are treated as Unicode no matter what locale is in effect.
238Under Unicode rules, there are a few case-insensitive matches that cross
239the 255/256 boundary. These are disallowed under C</l>. For example,
516074bb
KW
2400xFF (on ASCII platforms) does not caselessly match the character at
2410x178, C<LATIN CAPITAL LETTER Y WITH DIAERESIS>, because 0xFF may not be
242C<LATIN SMALL LETTER Y WITH DIAERESIS> in the current locale, and Perl
243has no way of knowing if that character even exists in the locale, much
244less what code point it is.
ed7efc79
KW
245
246This modifier may be specified to be the default by C<use locale>, but
247see L</Which character set modifier is in effect?>.
b6fa137b
FC
248X</l>
249
ed7efc79
KW
250=head4 /u
251
252means to use Unicode rules when pattern matching. On ASCII platforms,
253this means that the code points between 128 and 255 take on their
808432af
KW
254Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's).
255(Otherwise Perl considers their meanings to be undefined.) Thus,
256under this modifier, the ASCII platform effectively becomes a Unicode
257platform; and hence, for example, C<\w> will match any of the more than
258100_000 word characters in Unicode.
ed7efc79
KW
259
260Unlike most locales, which are specific to a language and country pair,
516074bb
KW
261Unicode classifies all the characters that are letters I<somewhere> in
262the world as
ed7efc79
KW
263C<\w>. For example, your locale might not think that C<LATIN SMALL
264LETTER ETH> is a letter (unless you happen to speak Icelandic), but
265Unicode does. Similarly, all the characters that are decimal digits
266somewhere in the world will match C<\d>; this is hundreds, not 10,
267possible matches. And some of those digits look like some of the 10
268ASCII digits, but mean a different number, so a human could easily think
269a number is a different quantity than it really is. For example,
270C<BENGALI DIGIT FOUR> (U+09EA) looks very much like an
271C<ASCII DIGIT EIGHT> (U+0038). And, C<\d+>, may match strings of digits
272that are a mixture from different writing systems, creating a security
67592e11 273issue. L<Unicode::UCD/num()> can be used to sort
516074bb
KW
274this out. Or the C</a> modifier can be used to force C<\d> to match
275just the ASCII 0 through 9.
ed7efc79 276
516074bb
KW
277Also, under this modifier, case-insensitive matching works on the full
278set of Unicode
ed7efc79
KW
279characters. The C<KELVIN SIGN>, for example matches the letters "k" and
280"K"; and C<LATIN SMALL LIGATURE FF> matches the sequence "ff", which,
281if you're not prepared, might make it look like a hexadecimal constant,
282presenting another potential security issue. See
283L<http://unicode.org/reports/tr36> for a detailed discussion of Unicode
284security issues.
285
6368643f
KW
286On the EBCDIC platforms that Perl handles, the native character set is
287equivalent to Latin-1. Thus this modifier changes behavior only when
ed7efc79
KW
288the C<"/i"> modifier is also specified, and it turns out it affects only
289two characters, giving them full Unicode semantics: the C<MICRO SIGN>
6368643f 290will match the Greek capital and small letters C<MU>, otherwise not; and
ed7efc79
KW
291the C<LATIN CAPITAL LETTER SHARP S> will match any of C<SS>, C<Ss>,
292C<sS>, and C<ss>, otherwise not.
293
294This modifier may be specified to be the default by C<use feature
66cbab2c
KW
295'unicode_strings>, C<use locale ':not_characters'>, or
296C<L<use 5.012|perlfunc/use VERSION>> (or higher),
808432af 297but see L</Which character set modifier is in effect?>.
b6fa137b
FC
298X</u>
299
ed7efc79
KW
300=head4 /d
301
302This modifier means to use the "Default" native rules of the platform
303except when there is cause to use Unicode rules instead, as follows:
304
305=over 4
306
307=item 1
308
309the target string is encoded in UTF-8; or
310
311=item 2
312
313the pattern is encoded in UTF-8; or
314
315=item 3
316
317the pattern explicitly mentions a code point that is above 255 (say by
318C<\x{100}>); or
319
320=item 4
b6fa137b 321
ed7efc79
KW
322the pattern uses a Unicode name (C<\N{...}>); or
323
324=item 5
325
326the pattern uses a Unicode property (C<\p{...}>)
327
328=back
329
330Another mnemonic for this modifier is "Depends", as the rules actually
331used depend on various things, and as a result you can get unexpected
808432af
KW
332results. See L<perlunicode/The "Unicode Bug">. The Unicode Bug has
333become rather infamous, leading to yet another (printable) name for this
334modifier, "Dodgy".
ed7efc79
KW
335
336On ASCII platforms, the native rules are ASCII, and on EBCDIC platforms
337(at least the ones that Perl handles), they are Latin-1.
338
339Here are some examples of how that works on an ASCII platform:
340
341 $str = "\xDF"; # $str is not in UTF-8 format.
342 $str =~ /^\w/; # No match, as $str isn't in UTF-8 format.
343 $str .= "\x{0e0b}"; # Now $str is in UTF-8 format.
344 $str =~ /^\w/; # Match! $str is now in UTF-8 format.
345 chop $str;
346 $str =~ /^\w/; # Still a match! $str remains in UTF-8 format.
347
808432af
KW
348This modifier is automatically selected by default when none of the
349others are, so yet another name for it is "Default".
350
351Because of the unexpected behaviors associated with this modifier, you
352probably should only use it to maintain weird backward compatibilities.
353
354=head4 /a (and /aa)
355
356This modifier stands for ASCII-restrict (or ASCII-safe). This modifier,
357unlike the others, may be doubled-up to increase its effect.
358
359When it appears singly, it causes the sequences C<\d>, C<\s>, C<\w>, and
360the Posix character classes to match only in the ASCII range. They thus
361revert to their pre-5.6, pre-Unicode meanings. Under C</a>, C<\d>
362always means precisely the digits C<"0"> to C<"9">; C<\s> means the five
363characters C<[ \f\n\r\t]>; C<\w> means the 63 characters
364C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
365C<[[:print:]]> match only the appropriate ASCII-range characters.
366
367This modifier is useful for people who only incidentally use Unicode,
368and who do not wish to be burdened with its complexities and security
369concerns.
370
371With C</a>, one can write C<\d> with confidence that it will only match
372ASCII characters, and should the need arise to match beyond ASCII, you
373can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>). There are
374similar C<\p{...}> constructs that can match beyond ASCII both white
375space (see L<perlrecharclass/Whitespace>), and Posix classes (see
376L<perlrecharclass/POSIX Character Classes>). Thus, this modifier
377doesn't mean you can't use Unicode, it means that to get Unicode
378matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
379signals Unicode.
380
381As you would expect, this modifier causes, for example, C<\D> to mean
382the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
383C<\D>, C<\S>, and C<\W>. C<\b> still means to match at the boundary
384between C<\w> and C<\W>, using the C</a> definitions of them (similarly
385for C<\B>).
386
387Otherwise, C</a> behaves like the C</u> modifier, in that
388case-insensitive matching uses Unicode semantics; for example, "k" will
389match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
390points in the Latin1 range, above ASCII will have Unicode rules when it
391comes to case-insensitive matching.
392
393To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
394specify the "a" twice, for example C</aai> or C</aia>. (The first
395occurrence of "a" restricts the C<\d>, etc., and the second occurrence
396adds the C</i> restrictions.) But, note that code points outside the
397ASCII range will use Unicode rules for C</i> matching, so the modifier
398doesn't really restrict things to just ASCII; it just forbids the
399intermixing of ASCII and non-ASCII.
400
401To summarize, this modifier provides protection for applications that
402don't wish to be exposed to all of Unicode. Specifying it twice
403gives added protection.
404
405This modifier may be specified to be the default by C<use re '/a'>
406or C<use re '/aa'>. If you do so, you may actually have occasion to use
407the C</u> modifier explictly if there are a few regular expressions
408where you do want full Unicode rules (but even here, it's best if
409everything were under feature C<"unicode_strings">, along with the
410C<use re '/aa'>). Also see L</Which character set modifier is in
411effect?>.
412X</a>
413X</aa>
414
ed7efc79
KW
415=head4 Which character set modifier is in effect?
416
417Which of these modifiers is in effect at any given point in a regular
808432af
KW
418expression depends on a fairly complex set of interactions. These have
419been designed so that in general you don't have to worry about it, but
420this section gives the gory details. As
ed7efc79
KW
421explained below in L</Extended Patterns> it is possible to explicitly
422specify modifiers that apply only to portions of a regular expression.
423The innermost always has priority over any outer ones, and one applying
6368643f
KW
424to the whole expression has priority over any of the default settings that are
425described in the remainder of this section.
ed7efc79 426
916cec3f 427The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
ed7efc79
KW
428default modifiers (including these) for regular expressions compiled
429within its scope. This pragma has precedence over the other pragmas
516074bb 430listed below that also change the defaults.
ed7efc79
KW
431
432Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
66cbab2c 433and C<L<use feature 'unicode_strings|feature>>, or
ed7efc79
KW
434C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
435C</u> when not in the same scope as either C<L<use locale|perllocale>>
66cbab2c
KW
436or C<L<use bytes|bytes>>.
437(C<L<use locale ':not_characters'|perllocale/Unicode and UTF-8>> also
438sets the default to C</u>, overriding any plain C<use locale>.)
439Unlike the mechanisms mentioned above, these
6368643f
KW
440affect operations besides regular expressions pattern matching, and so
441give more consistent results with other operators, including using
442C<\U>, C<\l>, etc. in substitution replacements.
ed7efc79
KW
443
444If none of the above apply, for backwards compatibility reasons, the
445C</d> modifier is the one in effect by default. As this can lead to
446unexpected results, it is best to specify which other rule set should be
447used.
448
449=head4 Character set modifier behavior prior to Perl 5.14
450
451Prior to 5.14, there were no explicit modifiers, but C</l> was implied
452for regexes compiled within the scope of C<use locale>, and C</d> was
453implied otherwise. However, interpolating a regex into a larger regex
454would ignore the original compilation in favor of whatever was in effect
455at the time of the second compilation. There were a number of
456inconsistencies (bugs) with the C</d> modifier, where Unicode rules
457would be used when inappropriate, and vice versa. C<\p{}> did not imply
458Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
b6fa137b 459
a0d0e21e
LW
460=head2 Regular Expressions
461
04838cea
RGS
462=head3 Metacharacters
463
384f06ae 464The patterns used in Perl pattern matching evolved from those supplied in
14218588 465the Version 8 regex routines. (The routines are derived
19799a22
GS
466(distantly) from Henry Spencer's freely redistributable reimplementation
467of the V8 routines.) See L<Version 8 Regular Expressions> for
468details.
a0d0e21e
LW
469
470In particular the following metacharacters have their standard I<egrep>-ish
471meanings:
d74e8afc
ITB
472X<metacharacter>
473X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
474
a0d0e21e 475
f793d64a
KW
476 \ Quote the next metacharacter
477 ^ Match the beginning of the line
478 . Match any character (except newline)
479 $ Match the end of the line (or before newline at the end)
480 | Alternation
481 () Grouping
482 [] Bracketed Character class
a0d0e21e 483
14218588
GS
484By default, the "^" character is guaranteed to match only the
485beginning of the string, the "$" character only the end (or before the
486newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
487assumption that the string contains only one line. Embedded newlines
488will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 489string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
490newline within the string (except if the newline is the last character in
491the string), and "$" will match before any newline. At the
a0d0e21e
LW
492cost of a little more overhead, you can do this by using the /m modifier
493on the pattern match operator. (Older programs did this by setting C<$*>,
db7cd43a 494but this option was removed in perl 5.10.)
d74e8afc 495X<^> X<$> X</m>
a0d0e21e 496
14218588 497To simplify multi-line substitutions, the "." character never matches a
55497cff 498newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 499the string is a single line--even if it isn't.
d74e8afc 500X<.> X</s>
a0d0e21e 501
04838cea
RGS
502=head3 Quantifiers
503
a0d0e21e 504The following standard quantifiers are recognized:
d74e8afc 505X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 506
f793d64a
KW
507 * Match 0 or more times
508 + Match 1 or more times
509 ? Match 1 or 0 times
510 {n} Match exactly n times
511 {n,} Match at least n times
512 {n,m} Match at least n but not more than m times
a0d0e21e 513
0b928c2f
FC
514(If a curly bracket occurs in any other context and does not form part of
515a backslashed sequence like C<\x{...}>, it is treated
9af81bfe
KW
516as a regular character. In particular, the lower quantifier bound
517is not optional. However, in Perl v5.18, it is planned to issue a
518deprecation warning for all such occurrences, and in Perl v5.20 to
519require literal uses of a curly bracket to be escaped, say by preceding
520them with a backslash or enclosing them within square brackets, (C<"\{">
521or C<"[{]">). This change will allow for future syntax extensions (like
522making the lower bound of a quantifier optional), and better error
523checking of quantifiers. Now, a typo in a quantifier silently causes
524it to be treated as the literal characters. For example,
525
526 /o{4,3}/
527
528looks like a quantifier that matches 0 times, since 4 is greater than 3,
529but it really means to match the sequence of six characters
530S<C<"o { 4 , 3 }">>.)
531
532The "*" quantifier is equivalent to C<{0,}>, the "+"
527e91da 533quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 534to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
535This is usually 32766 on the most common platforms. The actual limit can
536be seen in the error message generated by code such as this:
537
820475bd 538 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 539
54310121
PP
540By default, a quantified subpattern is "greedy", that is, it will match as
541many times as possible (given a particular starting location) while still
542allowing the rest of the pattern to match. If you want it to match the
543minimum number of times possible, follow the quantifier with a "?". Note
544that the meanings don't change, just the "greediness":
0d017f4d 545X<metacharacter> X<greedy> X<greediness>
d74e8afc 546X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 547
f793d64a
KW
548 *? Match 0 or more times, not greedily
549 +? Match 1 or more times, not greedily
550 ?? Match 0 or 1 time, not greedily
0b928c2f 551 {n}? Match exactly n times, not greedily (redundant)
f793d64a
KW
552 {n,}? Match at least n times, not greedily
553 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 554
b9b4dddf
YO
555By default, when a quantified subpattern does not allow the rest of the
556overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 557sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
558as well.
559
f793d64a
KW
560 *+ Match 0 or more times and give nothing back
561 ++ Match 1 or more times and give nothing back
562 ?+ Match 0 or 1 time and give nothing back
563 {n}+ Match exactly n times and give nothing back (redundant)
564 {n,}+ Match at least n times and give nothing back
565 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
566
567For instance,
568
569 'aaaa' =~ /a++a/
570
571will never match, as the C<a++> will gobble up all the C<a>'s in the
572string and won't leave any for the remaining part of the pattern. This
573feature can be extremely useful to give perl hints about where it
574shouldn't backtrack. For instance, the typical "match a double-quoted
575string" problem can be most efficiently performed when written as:
576
577 /"(?:[^"\\]++|\\.)*+"/
578
0d017f4d 579as we know that if the final quote does not match, backtracking will not
0b928c2f
FC
580help. See the independent subexpression
581L</C<< (?>pattern) >>> for more details;
b9b4dddf
YO
582possessive quantifiers are just syntactic sugar for that construct. For
583instance the above example could also be written as follows:
584
585 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
586
04838cea
RGS
587=head3 Escape sequences
588
0b928c2f 589Because patterns are processed as double-quoted strings, the following
a0d0e21e
LW
590also work:
591
f793d64a
KW
592 \t tab (HT, TAB)
593 \n newline (LF, NL)
594 \r return (CR)
595 \f form feed (FF)
596 \a alarm (bell) (BEL)
597 \e escape (think troff) (ESC)
f793d64a 598 \cK control char (example: VT)
dc0d9c48 599 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 600 \N{name} named Unicode character or character sequence
f793d64a 601 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 602 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
603 \l lowercase next char (think vi)
604 \u uppercase next char (think vi)
605 \L lowercase till \E (think vi)
606 \U uppercase till \E (think vi)
607 \Q quote (disable) pattern metacharacters till \E
608 \E end either case modification or quoted section, think vi
a0d0e21e 609
9bb1f947 610Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 611
e1d1eefb 612=head3 Character Classes and other Special Escapes
04838cea 613
a0d0e21e 614In addition, Perl defines the following:
d0b16107 615X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 616
f793d64a
KW
617 Sequence Note Description
618 [...] [1] Match a character according to the rules of the
619 bracketed character class defined by the "...".
620 Example: [a-z] matches "a" or "b" or "c" ... or "z"
621 [[:...:]] [2] Match a character according to the rules of the POSIX
622 character class "..." within the outer bracketed
623 character class. Example: [[:upper:]] matches any
624 uppercase character.
d35dd6c6
KW
625 \w [3] Match a "word" character (alphanumeric plus "_", plus
626 other connector punctuation chars plus Unicode
0b928c2f 627 marks)
f793d64a
KW
628 \W [3] Match a non-"word" character
629 \s [3] Match a whitespace character
630 \S [3] Match a non-whitespace character
631 \d [3] Match a decimal digit character
632 \D [3] Match a non-digit character
633 \pP [3] Match P, named property. Use \p{Prop} for longer names
634 \PP [3] Match non-P
635 \X [4] Match Unicode "eXtended grapheme cluster"
636 \C Match a single C-language char (octet) even if that is
637 part of a larger UTF-8 character. Thus it breaks up
638 characters into their UTF-8 bytes, so you may end up
639 with malformed pieces of UTF-8. Unsupported in
640 lookbehind.
c27a5cfe 641 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
642 '1' may actually be any positive integer.
643 \g1 [5] Backreference to a specific or previous group,
644 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 645 previous group and may optionally be wrapped in
f793d64a
KW
646 curly brackets for safer parsing.
647 \g{name} [5] Named backreference
648 \k<name> [5] Named backreference
649 \K [6] Keep the stuff left of the \K, don't include it in $&
650 \N [7] Any character but \n (experimental). Not affected by
651 /s modifier
652 \v [3] Vertical whitespace
653 \V [3] Not vertical whitespace
654 \h [3] Horizontal whitespace
655 \H [3] Not horizontal whitespace
656 \R [4] Linebreak
e1d1eefb 657
9bb1f947
KW
658=over 4
659
660=item [1]
661
662See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 663
9bb1f947 664=item [2]
b8c5462f 665
9bb1f947 666See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 667
9bb1f947 668=item [3]
5496314a 669
9bb1f947 670See L<perlrecharclass/Backslash sequences> for details.
5496314a 671
9bb1f947 672=item [4]
5496314a 673
9bb1f947 674See L<perlrebackslash/Misc> for details.
d0b16107 675
9bb1f947 676=item [5]
b8c5462f 677
c27a5cfe 678See L</Capture groups> below for details.
93733859 679
9bb1f947 680=item [6]
b8c5462f 681
9bb1f947
KW
682See L</Extended Patterns> below for details.
683
684=item [7]
685
686Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
687character or character sequence whose name is C<NAME>; and similarly
688when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
689code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947
KW
690
691=back
d0b16107 692
04838cea
RGS
693=head3 Assertions
694
a0d0e21e 695Perl defines the following zero-width assertions:
d74e8afc
ITB
696X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
697X<regexp, zero-width assertion>
698X<regular expression, zero-width assertion>
699X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 700
9bb1f947
KW
701 \b Match a word boundary
702 \B Match except at a word boundary
703 \A Match only at beginning of string
704 \Z Match only at end of string, or before newline at the end
705 \z Match only at end of string
706 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 707 of prior m//g)
a0d0e21e 708
14218588 709A word boundary (C<\b>) is a spot between two characters
19799a22
GS
710that has a C<\w> on one side of it and a C<\W> on the other side
711of it (in either order), counting the imaginary characters off the
712beginning and end of the string as matching a C<\W>. (Within
713character classes C<\b> represents backspace rather than a word
714boundary, just as it normally does in any double-quoted string.)
715The C<\A> and C<\Z> are just like "^" and "$", except that they
716won't match multiple times when the C</m> modifier is used, while
717"^" and "$" will match at every internal line boundary. To match
718the actual end of the string and not ignore an optional trailing
719newline, use C<\z>.
d74e8afc 720X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
721
722The C<\G> assertion can be used to chain global matches (using
723C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
724It is also useful when writing C<lex>-like scanners, when you have
725several patterns that you want to match against consequent substrings
0b928c2f 726of your string; see the previous reference. The actual location
19799a22 727where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d 728an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
0b928c2f
FC
729matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
730is modified somewhat, in that contents to the left of C<\G> are
58e23c8d
YO
731not counted when determining the length of the match. Thus the following
732will not match forever:
d74e8afc 733X<\G>
c47ff5f1 734
e761bb84
CO
735 my $string = 'ABC';
736 pos($string) = 1;
737 while ($string =~ /(.\G)/g) {
738 print $1;
739 }
58e23c8d
YO
740
741It will print 'A' and then terminate, as it considers the match to
742be zero-width, and thus will not match at the same position twice in a
743row.
744
745It is worth noting that C<\G> improperly used can result in an infinite
746loop. Take care when using patterns that include C<\G> in an alternation.
747
c27a5cfe 748=head3 Capture groups
04838cea 749
c27a5cfe
KW
750The bracketing construct C<( ... )> creates capture groups (also referred to as
751capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
752the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
753for the second, and so on.
754This is called a I<backreference>.
d74e8afc 755X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 756X<regex, capture group> X<regexp, capture group>
d74e8afc 757X<regular expression, capture buffer> X<backreference>
c27a5cfe 758X<regular expression, capture group> X<backreference>
1f1031fe 759X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
760X<named capture buffer> X<regular expression, named capture buffer>
761X<named capture group> X<regular expression, named capture group>
762X<%+> X<$+{name}> X<< \k<name> >>
763There is no limit to the number of captured substrings that you may use.
764Groups are numbered with the leftmost open parenthesis being number 1, etc. If
765a group did not match, the associated backreference won't match either. (This
766can happen if the group is optional, or in a different branch of an
767alternation.)
768You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
769this form, described below.
770
771You can also refer to capture groups relatively, by using a negative number, so
772that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
773group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
774example:
5624f11d
YO
775
776 /
c27a5cfe
KW
777 (Y) # group 1
778 ( # group 2
779 (X) # group 3
780 \g{-1} # backref to group 3
781 \g{-3} # backref to group 1
5624f11d
YO
782 )
783 /x
784
d8b950dc
KW
785would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
786interpolate regexes into larger regexes and not have to worry about the
787capture groups being renumbered.
788
789You can dispense with numbers altogether and create named capture groups.
790The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
791reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
792also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
793I<name> must not begin with a number, nor contain hyphens.
794When different groups within the same pattern have the same name, any reference
795to that name assumes the leftmost defined group. Named groups count in
796absolute and relative numbering, and so can also be referred to by those
797numbers.
798(It's possible to do things with named capture groups that would otherwise
799require C<(??{})>.)
800
801Capture group contents are dynamically scoped and available to you outside the
802pattern until the end of the enclosing block or until the next successful
803match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
804You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
805etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
806
807Braces are required in referring to named capture groups, but are optional for
808absolute or relative numbered ones. Braces are safer when creating a regex by
809concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
810contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
811is probably not what you intended.
812
813The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
814there were no named nor relative numbered capture groups. Absolute numbered
0b928c2f
FC
815groups were referred to using C<\1>,
816C<\2>, etc., and this notation is still
d8b950dc
KW
817accepted (and likely always will be). But it leads to some ambiguities if
818there are more than 9 capture groups, as C<\10> could mean either the tenth
819capture group, or the character whose ordinal in octal is 010 (a backspace in
820ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
821only if at least 10 left parentheses have opened before it. Likewise C<\11> is
822a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
823And so on. C<\1> through C<\9> are always interpreted as backreferences.
824There are several examples below that illustrate these perils. You can avoid
825the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
826and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
827digits padded with leading zeros, since a leading zero implies an octal
828constant.
d8b950dc
KW
829
830The C<\I<digit>> notation also works in certain circumstances outside
ed7efc79 831the pattern. See L</Warning on \1 Instead of $1> below for details.
81714fb9 832
14218588 833Examples:
a0d0e21e
LW
834
835 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
836
d8b950dc 837 /(.)\g1/ # find first doubled char
81714fb9
YO
838 and print "'$1' is the first doubled character\n";
839
840 /(?<char>.)\k<char>/ # ... a different way
841 and print "'$+{char}' is the first doubled character\n";
842
d8b950dc 843 /(?'char'.)\g1/ # ... mix and match
81714fb9 844 and print "'$1' is the first doubled character\n";
c47ff5f1 845
14218588 846 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
847 $hours = $1;
848 $minutes = $2;
849 $seconds = $3;
a0d0e21e 850 }
c47ff5f1 851
9d860678
KW
852 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
853 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
854 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
855 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
856
857 $a = '(.)\1'; # Creates problems when concatenated.
858 $b = '(.)\g{1}'; # Avoids the problems.
859 "aa" =~ /${a}/; # True
860 "aa" =~ /${b}/; # True
861 "aa0" =~ /${a}0/; # False!
862 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
863 "aa\x08" =~ /${a}0/; # True!
864 "aa\x08" =~ /${b}0/; # False
9d860678 865
14218588
GS
866Several special variables also refer back to portions of the previous
867match. C<$+> returns whatever the last bracket match matched.
868C<$&> returns the entire matched string. (At one point C<$0> did
869also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
870everything before the matched string. C<$'> returns everything
871after the matched string. And C<$^N> contains whatever was matched by
872the most-recently closed group (submatch). C<$^N> can be used in
873extended patterns (see below), for example to assign a submatch to a
81714fb9 874variable.
d74e8afc 875X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 876
d8b950dc
KW
877These special variables, like the C<%+> hash and the numbered match variables
878(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
879until the end of the enclosing block or until the next successful
880match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
881X<$+> X<$^N> X<$&> X<$`> X<$'>
882X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
883
0d017f4d 884B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 885which makes it easier to write code that tests for a series of more
665e98b9
JH
886specific cases and remembers the best match.
887
14218588
GS
888B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
889C<$'> anywhere in the program, it has to provide them for every
890pattern match. This may substantially slow your program. Perl
d8b950dc 891uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
892price for each pattern that contains capturing parentheses. (To
893avoid this cost while retaining the grouping behaviour, use the
894extended regular expression C<(?: ... )> instead.) But if you never
895use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
896parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
897if you can, but if you can't (and some algorithms really appreciate
898them), once you've used them once, use them at will, because you've
c149d39e
DM
899already paid the price. As of 5.17.4, the presence of each of the three
900variables in a program is recorded separately, and depending on
901circumstances, perl may be able be more efficient knowing that only C<$&>
902rather than all three have been seen, for example.
d74e8afc 903X<$&> X<$`> X<$'>
68dc0745 904
99d59c4d 905As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
906C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
907and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 908successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
909The use of these variables incurs no global performance penalty, unlike
910their punctuation char equivalents, however at the trade-off that you
911have to tell perl when you want to use them.
87e95b7f 912X</p> X<p modifier>
cde0cee5 913
9d727203
KW
914=head2 Quoting metacharacters
915
19799a22
GS
916Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
917C<\w>, C<\n>. Unlike some other regular expression languages, there
918are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 919that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
920interpreted as a literal character, not a metacharacter. This was
921once used in a common idiom to disable or quote the special meanings
922of regular expression metacharacters in a string that you want to
36bbe248 923use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
924
925 $pattern =~ s/(\W)/\\$1/g;
926
f1cbbd6e 927(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
928Today it is more common to use the quotemeta() function or the C<\Q>
929metaquoting escape sequence to disable all metacharacters' special
930meanings like this:
a0d0e21e
LW
931
932 /$unquoted\Q$quoted\E$unquoted/
933
9da458fc
IZ
934Beware that if you put literal backslashes (those not inside
935interpolated variables) between C<\Q> and C<\E>, double-quotish
936backslash interpolation may lead to confusing results. If you
937I<need> to use literal backslashes within C<\Q...\E>,
938consult L<perlop/"Gory details of parsing quoted constructs">.
939
736fe711
KW
940C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
941
19799a22
GS
942=head2 Extended Patterns
943
14218588 944Perl also defines a consistent extension syntax for features not
0b928c2f
FC
945found in standard tools like B<awk> and
946B<lex>. The syntax for most of these is a
14218588
GS
947pair of parentheses with a question mark as the first thing within
948the parentheses. The character after the question mark indicates
949the extension.
19799a22 950
14218588
GS
951The stability of these extensions varies widely. Some have been
952part of the core language for many years. Others are experimental
953and may change without warning or be completely removed. Check
954the documentation on an individual feature to verify its current
955status.
19799a22 956
14218588
GS
957A question mark was chosen for this and for the minimal-matching
958construct because 1) question marks are rare in older regular
959expressions, and 2) whenever you see one, you should stop and
0b928c2f 960"question" exactly what is going on. That's psychology....
a0d0e21e 961
70ca8714 962=over 4
a0d0e21e 963
cc6b7395 964=item C<(?#text)>
d74e8afc 965X<(?#)>
a0d0e21e 966
14218588 967A comment. The text is ignored. If the C</x> modifier enables
19799a22 968whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
969the comment as soon as it sees a C<)>, so there is no way to put a literal
970C<)> in the comment.
a0d0e21e 971
cfaf538b 972=item C<(?adlupimsx-imsx)>
fb85c044 973
cfaf538b 974=item C<(?^alupimsx)>
fb85c044 975X<(?)> X<(?^)>
19799a22 976
0b6d1084
JH
977One or more embedded pattern-match modifiers, to be turned on (or
978turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
979the remainder of the enclosing pattern group (if any).
980
fb85c044 981This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 982configuration file, taken from an argument, or specified in a table
0b928c2f
FC
983somewhere. Consider the case where some patterns want to be
984case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 985include C<(?i)> at the front of the pattern. For example:
19799a22
GS
986
987 $pattern = "foobar";
5d458dd8 988 if ( /$pattern/i ) { }
19799a22
GS
989
990 # more flexible:
991
992 $pattern = "(?i)foobar";
5d458dd8 993 if ( /$pattern/ ) { }
19799a22 994
0b6d1084 995These modifiers are restored at the end of the enclosing group. For example,
19799a22 996
d8b950dc 997 ( (?i) blah ) \s+ \g1
19799a22 998
0d017f4d
WL
999will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
1000repetition of the previous word, assuming the C</x> modifier, and no C</i>
1001modifier outside this group.
19799a22 1002
8eb5594e 1003These modifiers do not carry over into named subpatterns called in the
dd72e27b 1004enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
8eb5594e
DR
1005change the case-sensitivity of the "NAME" pattern.
1006
dc925305
KW
1007Any of these modifiers can be set to apply globally to all regular
1008expressions compiled within the scope of a C<use re>. See
a0bbd6ff 1009L<re/"'/flags' mode">.
dc925305 1010
9de15fec
KW
1011Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1012after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
1013C<"d">) may follow the caret to override it.
1014But a minus sign is not legal with it.
1015
dc925305 1016Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 1017that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 1018C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
1019others, and a maximum of one (or two C<a>'s) may appear in the
1020construct. Thus, for
0b928c2f 1021example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 1022C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1023
1024Note also that the C<p> modifier is special in that its presence
1025anywhere in a pattern has a global effect.
cde0cee5 1026
5a964f20 1027=item C<(?:pattern)>
d74e8afc 1028X<(?:)>
a0d0e21e 1029
cfaf538b 1030=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1031
cfaf538b 1032=item C<(?^aluimsx:pattern)>
fb85c044
KW
1033X<(?^:)>
1034
5a964f20
TC
1035This is for clustering, not capturing; it groups subexpressions like
1036"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1037
5a964f20 1038 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1039
1040is like
1041
5a964f20 1042 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1043
19799a22
GS
1044but doesn't spit out extra fields. It's also cheaper not to capture
1045characters if you don't need to.
a0d0e21e 1046
19799a22 1047Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1048C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1049
1050 /(?s-i:more.*than).*million/i
1051
14218588 1052is equivalent to the more verbose
ca9dfc88
IZ
1053
1054 /(?:(?s-i)more.*than).*million/i
1055
fb85c044 1056Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1057after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1058flags (except C<"d">) may follow the caret, so
fb85c044
KW
1059
1060 (?^x:foo)
1061
1062is equivalent to
1063
1064 (?x-ims:foo)
1065
1066The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1067surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1068modified by any flags specified.
1069
1070The caret allows for simpler stringification of compiled regular
1071expressions. These look like
1072
1073 (?^:pattern)
1074
1075with any non-default flags appearing between the caret and the colon.
1076A test that looks at such stringification thus doesn't need to have the
1077system default flags hard-coded in it, just the caret. If new flags are
1078added to Perl, the meaning of the caret's expansion will change to include
1079the default for those flags, so the test will still work, unchanged.
1080
1081Specifying a negative flag after the caret is an error, as the flag is
1082redundant.
1083
1084Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1085to match at the beginning.
1086
594d7033
YO
1087=item C<(?|pattern)>
1088X<(?|)> X<Branch reset>
1089
1090This is the "branch reset" pattern, which has the special property
c27a5cfe 1091that the capture groups are numbered from the same starting point
99d59c4d 1092in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1093
c27a5cfe 1094Capture groups are numbered from left to right, but inside this
693596a8 1095construct the numbering is restarted for each branch.
4deaaa80 1096
c27a5cfe 1097The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1098following this construct will be numbered as though the construct
1099contained only one branch, that being the one with the most capture
c27a5cfe 1100groups in it.
4deaaa80 1101
0b928c2f 1102This construct is useful when you want to capture one of a
4deaaa80
PJ
1103number of alternative matches.
1104
1105Consider the following pattern. The numbers underneath show in
c27a5cfe 1106which group the captured content will be stored.
594d7033
YO
1107
1108
1109 # before ---------------branch-reset----------- after
1110 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1111 # 1 2 2 3 2 3 4
1112
ab106183
A
1113Be careful when using the branch reset pattern in combination with
1114named captures. Named captures are implemented as being aliases to
c27a5cfe 1115numbered groups holding the captures, and that interferes with the
ab106183
A
1116implementation of the branch reset pattern. If you are using named
1117captures in a branch reset pattern, it's best to use the same names,
1118in the same order, in each of the alternations:
1119
1120 /(?| (?<a> x ) (?<b> y )
1121 | (?<a> z ) (?<b> w )) /x
1122
1123Not doing so may lead to surprises:
1124
1125 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1126 say $+ {a}; # Prints '12'
1127 say $+ {b}; # *Also* prints '12'.
1128
c27a5cfe
KW
1129The problem here is that both the group named C<< a >> and the group
1130named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1131
ee9b8eae
YO
1132=item Look-Around Assertions
1133X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1134
0b928c2f 1135Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1136pattern without including it in C<$&>. Positive assertions match when
1137their subpattern matches, negative assertions match when their subpattern
1138fails. Look-behind matches text up to the current match position,
1139look-ahead matches text following the current match position.
1140
1141=over 4
1142
5a964f20 1143=item C<(?=pattern)>
d74e8afc 1144X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1145
19799a22 1146A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1147matches a word followed by a tab, without including the tab in C<$&>.
1148
5a964f20 1149=item C<(?!pattern)>
d74e8afc 1150X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1151
19799a22 1152A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1153matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1154however that look-ahead and look-behind are NOT the same thing. You cannot
1155use this for look-behind.
7b8d334a 1156
5a964f20 1157If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1158will not do what you want. That's because the C<(?!foo)> is just saying that
1159the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1160match. Use look-behind instead (see below).
c277df42 1161
ee9b8eae
YO
1162=item C<(?<=pattern)> C<\K>
1163X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1164
c47ff5f1 1165A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1166matches a word that follows a tab, without including the tab in C<$&>.
1167Works only for fixed-width look-behind.
c277df42 1168
ee9b8eae
YO
1169There is a special form of this construct, called C<\K>, which causes the
1170regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1171not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1172look-behind. The use of C<\K> inside of another look-around assertion
1173is allowed, but the behaviour is currently not well defined.
1174
c62285ac 1175For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1176equivalent C<< (?<=...) >> construct, and it is especially useful in
1177situations where you want to efficiently remove something following
1178something else in a string. For instance
1179
1180 s/(foo)bar/$1/g;
1181
1182can be rewritten as the much more efficient
1183
1184 s/foo\Kbar//g;
1185
5a964f20 1186=item C<(?<!pattern)>
d74e8afc 1187X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1188
19799a22
GS
1189A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1190matches any occurrence of "foo" that does not follow "bar". Works
1191only for fixed-width look-behind.
c277df42 1192
ee9b8eae
YO
1193=back
1194
81714fb9
YO
1195=item C<(?'NAME'pattern)>
1196
1197=item C<< (?<NAME>pattern) >>
1198X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1199
c27a5cfe 1200A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1201parentheses C<()> but for the additional fact that the group
1202can be referred to by name in various regular expression
1203constructs (like C<\g{NAME}>) and can be accessed by name
1204after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1205for more details on the C<%+> and C<%-> hashes.
81714fb9 1206
c27a5cfe
KW
1207If multiple distinct capture groups have the same name then the
1208$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1209
0d017f4d 1210The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1211
1212B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1213function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1214numbered sequentially regardless of being named or not. Thus in the
1215pattern
1216
1217 /(x)(?<foo>y)(z)/
1218
1219$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1220the opposite which is what a .NET regex hacker might expect.
1221
1f1031fe
YO
1222Currently NAME is restricted to simple identifiers only.
1223In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1224its Unicode extension (see L<utf8>),
1225though it isn't extended by the locale (see L<perllocale>).
81714fb9 1226
1f1031fe 1227B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1228with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1229may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1230support the use of single quotes as a delimiter for the name.
81714fb9 1231
1f1031fe
YO
1232=item C<< \k<NAME> >>
1233
1234=item C<< \k'NAME' >>
81714fb9
YO
1235
1236Named backreference. Similar to numeric backreferences, except that
1237the group is designated by name and not number. If multiple groups
1238have the same name then it refers to the leftmost defined group in
1239the current match.
1240
0d017f4d 1241It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1242earlier in the pattern.
1243
1244Both forms are equivalent.
1245
1f1031fe 1246B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1247with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1248may be used instead of C<< \k<NAME> >>.
1f1031fe 1249
cc6b7395 1250=item C<(?{ code })>
d74e8afc 1251X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1252
19799a22 1253B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1254experimental, and may be changed without notice. Code executed that
1255has side effects may not perform identically from version to version
e128ab2c
DM
1256due to the effect of future optimisations in the regex engine. The
1257implementation of this feature was radically overhauled for the 5.18.0
1258release, and its behaviour in earlier versions of perl was much buggier,
1259especially in relation to parsing, lexical vars, scoping, recursion and
1260reentrancy.
c277df42 1261
e128ab2c
DM
1262This zero-width assertion executes any embedded Perl code. It always
1263succeeds, and its return value is set as C<$^R>.
19799a22 1264
e128ab2c
DM
1265In literal patterns, the code is parsed at the same time as the
1266surrounding code. While within the pattern, control is passed temporarily
1267back to the perl parser, until the logically-balancing closing brace is
1268encountered. This is similar to the way that an array index expression in
1269a literal string is handled, for example
77ea4f6d 1270
e128ab2c
DM
1271 "abc$array[ 1 + f('[') + g()]def"
1272
1273In particular, braces do not need to be balanced:
1274
1275 /abc(?{ f('{'); })/def/
1276
1277Even in a pattern that is interpolated and compiled at run-time, literal
1278code blocks will be compiled once, at perl compile time; the following
1279prints "ABCD":
1280
1281 print "D";
1282 my $qr = qr/(?{ BEGIN { print "A" } })/;
1283 my $foo = "foo";
1284 /$foo$qr(?{ BEGIN { print "B" } })/;
1285 BEGIN { print "C" }
1286
1287In patterns where the text of the code is derived from run-time
1288information rather than appearing literally in a source code /pattern/,
1289the code is compiled at the same time that the pattern is compiled, and
1290fro reasons of security, C<use re 'eval'> must be in scope. This is to
1291stop user-supplied patterns containing code snippets from being
1292executable.
1293
1294In situations where you need enable this with C<use re 'eval'>, you should
1295also have taint checking enabled. Better yet, use the carefully
1296constrained evaluation within a Safe compartment. See L<perlsec> for
1297details about both these mechanisms.
1298
1299From the viewpoint of parsing, lexical variable scope and closures,
1300
1301 /AAA(?{ BBB })CCC/
1302
1303behaves approximately like
1304
1305 /AAA/ && do { BBB } && /CCC/
1306
1307Similarly,
1308
1309 qr/AAA(?{ BBB })CCC/
1310
1311behaves approximately like
77ea4f6d 1312
e128ab2c
DM
1313 sub { /AAA/ && do { BBB } && /CCC/ }
1314
1315In particular:
1316
1317 { my $i = 1; $r = qr/(?{ print $i })/ }
1318 my $i = 2;
1319 /$r/; # prints "1"
1320
1321Inside a C<(?{...})> block, C<$_> refers to the string the regular
754091cb 1322expression is matching against. You can also use C<pos()> to know what is
fa11829f 1323the current position of matching within this string.
754091cb 1324
e128ab2c
DM
1325The code block introduces a new scope from the perspective of lexical
1326variable declarations, but B<not> from the perspective of C<local> and
1327similar localizing behaviours. So later code blocks within the same
1328pattern will still see the values which were localized in earlier blocks.
1329These accumulated localizations are undone either at the end of a
1330successful match, or if the assertion is backtracked (compare
1331L<"Backtracking">). For example,
b9ac3b5b
GS
1332
1333 $_ = 'a' x 8;
5d458dd8 1334 m<
d1fbf752 1335 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1336 (
5d458dd8 1337 a
b9ac3b5b 1338 (?{
d1fbf752
KW
1339 local $cnt = $cnt + 1; # Update $cnt,
1340 # backtracking-safe.
b9ac3b5b 1341 })
5d458dd8 1342 )*
b9ac3b5b 1343 aaaa
d1fbf752
KW
1344 (?{ $res = $cnt }) # On success copy to
1345 # non-localized location.
b9ac3b5b
GS
1346 >x;
1347
e128ab2c
DM
1348will initially increment C<$cnt> up to 8; then during backtracking, its
1349value will be unwound back to 4, which is the value assigned to C<$res>.
1350At the end of the regex execution, $cnt will be wound back to its initial
1351value of 0.
1352
1353This assertion may be used as the condition in a
b9ac3b5b 1354
e128ab2c
DM
1355 (?(condition)yes-pattern|no-pattern)
1356
1357switch. If I<not> used in this way, the result of evaluation of C<code>
1358is put into the special variable C<$^R>. This happens immediately, so
1359C<$^R> can be used from other C<(?{ code })> assertions inside the same
1360regular expression.
b9ac3b5b 1361
19799a22
GS
1362The assignment to C<$^R> above is properly localized, so the old
1363value of C<$^R> is restored if the assertion is backtracked; compare
1364L<"Backtracking">.
b9ac3b5b 1365
e128ab2c
DM
1366Note that the special variable C<$^N> is particularly useful with code
1367blocks to capture the results of submatches in variables without having to
1368keep track of the number of nested parentheses. For example:
1369
1370 $_ = "The brown fox jumps over the lazy dog";
1371 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1372 print "color = $color, animal = $animal\n";
1373
8988a1bb 1374
14455d6c 1375=item C<(??{ code })>
d74e8afc
ITB
1376X<(??{})>
1377X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1378
19799a22 1379B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1380experimental, and may be changed without notice. Code executed that
1381has side effects may not perform identically from version to version
1382due to the effect of future optimisations in the regex engine.
0f5d15d6 1383
e128ab2c
DM
1384This is a "postponed" regular subexpression. It behaves in I<exactly> the
1385same way as a C<(?{ code })> code block as described above, except that
1386its return value, rather than being assigned to C<$^R>, is treated as a
1387pattern, compiled if it's a string (or used as-is if its a qr// object),
1388then matched as if it were inserted instead of this construct.
6bda09f9 1389
e128ab2c
DM
1390During the matching of this sub-pattern, it has its own set of
1391captures which are valid during the sub-match, but are discarded once
1392control returns to the main pattern. For example, the following matches,
1393with the inner pattern capturing "B" and matching "BB", while the outer
1394pattern captures "A";
1395
1396 my $inner = '(.)\1';
1397 "ABBA" =~ /^(.)(??{ $inner })\1/;
1398 print $1; # prints "A";
6bda09f9 1399
e128ab2c
DM
1400Note that this means that there is no way for the inner pattern to refer
1401to a capture group defined outside. (The code block itself can use C<$1>,
1402etc., to refer to the enclosing pattern's capture groups.) Thus, although
0f5d15d6 1403
e128ab2c
DM
1404 ('a' x 100)=~/(??{'(.)' x 100})/
1405
1406I<will> match, it will I<not> set $1 on exit.
19799a22
GS
1407
1408The following pattern matches a parenthesized group:
0f5d15d6 1409
d1fbf752
KW
1410 $re = qr{
1411 \(
1412 (?:
1413 (?> [^()]+ ) # Non-parens without backtracking
1414 |
1415 (??{ $re }) # Group with matching parens
1416 )*
1417 \)
1418 }x;
0f5d15d6 1419
6bda09f9
YO
1420See also C<(?PARNO)> for a different, more efficient way to accomplish
1421the same task.
1422
e128ab2c
DM
1423Executing a postponed regular expression 50 times without consuming any
1424input string will result in a fatal error. The maximum depth is compiled
1425into perl, so changing it requires a custom build.
6bda09f9 1426
542fa716
YO
1427=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1428X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1429X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1430X<regex, relative recursion>
6bda09f9 1431
e128ab2c
DM
1432Similar to C<(??{ code })> except that it does not involve executing any
1433code or potentially compiling a returned pattern string; instead it treats
1434the part of the current pattern contained within a specified capture group
1435as an independent pattern that must match at the current position.
1436Capture groups contained by the pattern will have the value as determined
1437by the outermost recursion.
6bda09f9 1438
894be9b7 1439PARNO is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1440the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1441the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716 1442C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
c27a5cfe 1443to be relative, with negative numbers indicating preceding capture groups
542fa716 1444and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1445declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1446Note that the counting for relative recursion differs from that of
c27a5cfe 1447relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1448included.
6bda09f9 1449
81714fb9 1450The following pattern matches a function foo() which may contain
f145b7e9 1451balanced parentheses as the argument.
6bda09f9 1452
d1fbf752 1453 $re = qr{ ( # paren group 1 (full function)
81714fb9 1454 foo
d1fbf752 1455 ( # paren group 2 (parens)
6bda09f9 1456 \(
d1fbf752 1457 ( # paren group 3 (contents of parens)
6bda09f9 1458 (?:
d1fbf752 1459 (?> [^()]+ ) # Non-parens without backtracking
6bda09f9 1460 |
d1fbf752 1461 (?2) # Recurse to start of paren group 2
6bda09f9
YO
1462 )*
1463 )
1464 \)
1465 )
1466 )
1467 }x;
1468
1469If the pattern was used as follows
1470
1471 'foo(bar(baz)+baz(bop))'=~/$re/
1472 and print "\$1 = $1\n",
1473 "\$2 = $2\n",
1474 "\$3 = $3\n";
1475
1476the output produced should be the following:
1477
1478 $1 = foo(bar(baz)+baz(bop))
1479 $2 = (bar(baz)+baz(bop))
81714fb9 1480 $3 = bar(baz)+baz(bop)
6bda09f9 1481
c27a5cfe 1482If there is no corresponding capture group defined, then it is a
61528107 1483fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1484string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1485into perl, so changing it requires a custom build.
1486
542fa716
YO
1487The following shows how using negative indexing can make it
1488easier to embed recursive patterns inside of a C<qr//> construct
1489for later use:
1490
1491 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
c77257ed 1492 if (/foo $parens \s+ \+ \s+ bar $parens/x) {
542fa716
YO
1493 # do something here...
1494 }
1495
81714fb9 1496B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1497PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1498a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1499as atomic. Also, modifiers are resolved at compile time, so constructs
1500like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1501be processed.
6bda09f9 1502
894be9b7
YO
1503=item C<(?&NAME)>
1504X<(?&NAME)>
1505
0d017f4d
WL
1506Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1507parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1508the same name, then it recurses to the leftmost.
1509
1510It is an error to refer to a name that is not declared somewhere in the
1511pattern.
1512
1f1031fe
YO
1513B<NOTE:> In order to make things easier for programmers with experience
1514with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1515may be used instead of C<< (?&NAME) >>.
1f1031fe 1516
e2e6a0f1
YO
1517=item C<(?(condition)yes-pattern|no-pattern)>
1518X<(?()>
286f584a 1519
e2e6a0f1 1520=item C<(?(condition)yes-pattern)>
286f584a 1521
41ef34de
ML
1522Conditional expression. Matches C<yes-pattern> if C<condition> yields
1523a true value, matches C<no-pattern> otherwise. A missing pattern always
1524matches.
1525
25e26d77 1526C<(condition)> should be one of: 1) an integer in
e2e6a0f1 1527parentheses (which is valid if the corresponding pair of parentheses
25e26d77 1528matched); 2) a look-ahead/look-behind/evaluate zero-width assertion; 3) a
c27a5cfe 1529name in angle brackets or single quotes (which is valid if a group
25e26d77 1530with the given name matched); or 4) the special symbol (R) (true when
e2e6a0f1
YO
1531evaluated inside of recursion or eval). Additionally the R may be
1532followed by a number, (which will be true when evaluated when recursing
1533inside of the appropriate group), or by C<&NAME>, in which case it will
1534be true only when evaluated during recursion in the named group.
1535
1536Here's a summary of the possible predicates:
1537
1538=over 4
1539
1540=item (1) (2) ...
1541
c27a5cfe 1542Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1543
1544=item (<NAME>) ('NAME')
1545
c27a5cfe 1546Checks if a group with the given name has matched something.
e2e6a0f1 1547
f01cd190
FC
1548=item (?=...) (?!...) (?<=...) (?<!...)
1549
1550Checks whether the pattern matches (or does not match, for the '!'
1551variants).
1552
e2e6a0f1
YO
1553=item (?{ CODE })
1554
f01cd190 1555Treats the return value of the code block as the condition.
e2e6a0f1
YO
1556
1557=item (R)
1558
1559Checks if the expression has been evaluated inside of recursion.
1560
1561=item (R1) (R2) ...
1562
1563Checks if the expression has been evaluated while executing directly
1564inside of the n-th capture group. This check is the regex equivalent of
1565
1566 if ((caller(0))[3] eq 'subname') { ... }
1567
1568In other words, it does not check the full recursion stack.
1569
1570=item (R&NAME)
1571
1572Similar to C<(R1)>, this predicate checks to see if we're executing
1573directly inside of the leftmost group with a given name (this is the same
1574logic used by C<(?&NAME)> to disambiguate). It does not check the full
1575stack, but only the name of the innermost active recursion.
1576
1577=item (DEFINE)
1578
1579In this case, the yes-pattern is never directly executed, and no
1580no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1581See below for details.
1582
1583=back
1584
1585For example:
1586
1587 m{ ( \( )?
1588 [^()]+
1589 (?(1) \) )
1590 }x
1591
1592matches a chunk of non-parentheses, possibly included in parentheses
1593themselves.
1594
0b928c2f
FC
1595A special form is the C<(DEFINE)> predicate, which never executes its
1596yes-pattern directly, and does not allow a no-pattern. This allows one to
1597define subpatterns which will be executed only by the recursion mechanism.
e2e6a0f1
YO
1598This way, you can define a set of regular expression rules that can be
1599bundled into any pattern you choose.
1600
1601It is recommended that for this usage you put the DEFINE block at the
1602end of the pattern, and that you name any subpatterns defined within it.
1603
1604Also, it's worth noting that patterns defined this way probably will
1605not be as efficient, as the optimiser is not very clever about
1606handling them.
1607
1608An example of how this might be used is as follows:
1609
2bf803e2 1610 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1611 (?(DEFINE)
2bf803e2
YO
1612 (?<NAME_PAT>....)
1613 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1614 )/x
1615
c27a5cfe
KW
1616Note that capture groups matched inside of recursion are not accessible
1617after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1618necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1619C<$+{NAME}> would be.
286f584a 1620
51a1303c
BF
1621Finally, keep in mind that subpatterns created inside a DEFINE block
1622count towards the absolute and relative number of captures, so this:
1623
1624 my @captures = "a" =~ /(.) # First capture
1625 (?(DEFINE)
1626 (?<EXAMPLE> 1 ) # Second capture
1627 )/x;
1628 say scalar @captures;
1629
1630Will output 2, not 1. This is particularly important if you intend to
1631compile the definitions with the C<qr//> operator, and later
1632interpolate them in another pattern.
1633
c47ff5f1 1634=item C<< (?>pattern) >>
6bda09f9 1635X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1636
19799a22
GS
1637An "independent" subexpression, one which matches the substring
1638that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1639position, and it matches I<nothing other than this substring>. This
19799a22
GS
1640construct is useful for optimizations of what would otherwise be
1641"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1642It may also be useful in places where the "grab all you can, and do not
1643give anything back" semantic is desirable.
19799a22 1644
c47ff5f1 1645For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1646(anchored at the beginning of string, as above) will match I<all>
1647characters C<a> at the beginning of string, leaving no C<a> for
1648C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1649since the match of the subgroup C<a*> is influenced by the following
1650group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1651C<a*ab> will match fewer characters than a standalone C<a*>, since
1652this makes the tail match.
1653
0b928c2f
FC
1654C<< (?>pattern) >> does not disable backtracking altogether once it has
1655matched. It is still possible to backtrack past the construct, but not
1656into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1657
c47ff5f1 1658An effect similar to C<< (?>pattern) >> may be achieved by writing
0b928c2f
FC
1659C<(?=(pattern))\g{-1}>. This matches the same substring as a standalone
1660C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
c47ff5f1 1661makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1662(The difference between these two constructs is that the second one
1663uses a capturing group, thus shifting ordinals of backreferences
1664in the rest of a regular expression.)
1665
1666Consider this pattern:
c277df42 1667
871b0233 1668 m{ \(
e2e6a0f1 1669 (
f793d64a 1670 [^()]+ # x+
e2e6a0f1 1671 |
871b0233
IZ
1672 \( [^()]* \)
1673 )+
e2e6a0f1 1674 \)
871b0233 1675 }x
5a964f20 1676
19799a22
GS
1677That will efficiently match a nonempty group with matching parentheses
1678two levels deep or less. However, if there is no such group, it
1679will take virtually forever on a long string. That's because there
1680are so many different ways to split a long string into several
1681substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1682to a subpattern of the above pattern. Consider how the pattern
1683above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1684seconds, but that each extra letter doubles this time. This
1685exponential performance will make it appear that your program has
14218588 1686hung. However, a tiny change to this pattern
5a964f20 1687
e2e6a0f1
YO
1688 m{ \(
1689 (
f793d64a 1690 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1691 |
871b0233
IZ
1692 \( [^()]* \)
1693 )+
e2e6a0f1 1694 \)
871b0233 1695 }x
c277df42 1696
c47ff5f1 1697which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1698this yourself would be a productive exercise), but finishes in a fourth
1699the time when used on a similar string with 1000000 C<a>s. Be aware,
0b928c2f
FC
1700however, that, when this construct is followed by a
1701quantifier, it currently triggers a warning message under
9f1b1f2d 1702the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1703C<"matches null string many times in regex">.
c277df42 1704
c47ff5f1 1705On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1706effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1707This was only 4 times slower on a string with 1000000 C<a>s.
1708
9da458fc
IZ
1709The "grab all you can, and do not give anything back" semantic is desirable
1710in many situations where on the first sight a simple C<()*> looks like
1711the correct solution. Suppose we parse text with comments being delimited
1712by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1713its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1714the comment delimiter, because it may "give up" some whitespace if
1715the remainder of the pattern can be made to match that way. The correct
1716answer is either one of these:
1717
1718 (?>#[ \t]*)
1719 #[ \t]*(?![ \t])
1720
1721For example, to grab non-empty comments into $1, one should use either
1722one of these:
1723
1724 / (?> \# [ \t]* ) ( .+ ) /x;
1725 / \# [ \t]* ( [^ \t] .* ) /x;
1726
1727Which one you pick depends on which of these expressions better reflects
1728the above specification of comments.
1729
6bda09f9
YO
1730In some literature this construct is called "atomic matching" or
1731"possessive matching".
1732
b9b4dddf
YO
1733Possessive quantifiers are equivalent to putting the item they are applied
1734to inside of one of these constructs. The following equivalences apply:
1735
1736 Quantifier Form Bracketing Form
1737 --------------- ---------------
1738 PAT*+ (?>PAT*)
1739 PAT++ (?>PAT+)
1740 PAT?+ (?>PAT?)
1741 PAT{min,max}+ (?>PAT{min,max})
1742
e2e6a0f1
YO
1743=back
1744
1745=head2 Special Backtracking Control Verbs
1746
1747B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1748removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1749be noted to avoid problems during upgrades.
1750
1751These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1752otherwise stated the ARG argument is optional; in some cases, it is
1753forbidden.
1754
1755Any pattern containing a special backtracking verb that allows an argument
e1020413 1756has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1757C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1758rules apply:
e2e6a0f1 1759
5d458dd8
YO
1760On failure, the C<$REGERROR> variable will be set to the ARG value of the
1761verb pattern, if the verb was involved in the failure of the match. If the
1762ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1763name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1764none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1765
5d458dd8
YO
1766On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1767the C<$REGMARK> variable will be set to the name of the last
1768C<(*MARK:NAME)> pattern executed. See the explanation for the
1769C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1770
5d458dd8 1771B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1772and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1773readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1774Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1775
1776If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1777argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1778
70ca8714 1779=over 3
e2e6a0f1
YO
1780
1781=item Verbs that take an argument
1782
1783=over 4
1784
5d458dd8 1785=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1786X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1787
5d458dd8
YO
1788This zero-width pattern prunes the backtracking tree at the current point
1789when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1790where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1791A may backtrack as necessary to match. Once it is reached, matching
1792continues in B, which may also backtrack as necessary; however, should B
1793not match, then no further backtracking will take place, and the pattern
1794will fail outright at the current starting position.
54612592
YO
1795
1796The following example counts all the possible matching strings in a
1797pattern (without actually matching any of them).
1798
e2e6a0f1 1799 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1800 print "Count=$count\n";
1801
1802which produces:
1803
1804 aaab
1805 aaa
1806 aa
1807 a
1808 aab
1809 aa
1810 a
1811 ab
1812 a
1813 Count=9
1814
5d458dd8 1815If we add a C<(*PRUNE)> before the count like the following
54612592 1816
5d458dd8 1817 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1818 print "Count=$count\n";
1819
0b928c2f 1820we prevent backtracking and find the count of the longest matching string
353c6505 1821at each matching starting point like so:
54612592
YO
1822
1823 aaab
1824 aab
1825 ab
1826 Count=3
1827
5d458dd8 1828Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1829
5d458dd8
YO
1830See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1831control backtracking. In some cases, the use of C<(*PRUNE)> can be
1832replaced with a C<< (?>pattern) >> with no functional difference; however,
1833C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1834C<< (?>pattern) >> alone.
54612592 1835
5d458dd8
YO
1836=item C<(*SKIP)> C<(*SKIP:NAME)>
1837X<(*SKIP)>
e2e6a0f1 1838
5d458dd8 1839This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1840failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1841to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1842of this pattern. This effectively means that the regex engine "skips" forward
1843to this position on failure and tries to match again, (assuming that
1844there is sufficient room to match).
1845
1846The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1847C<(*MARK:NAME)> was encountered while matching, then it is that position
1848which is used as the "skip point". If no C<(*MARK)> of that name was
1849encountered, then the C<(*SKIP)> operator has no effect. When used
1850without a name the "skip point" is where the match point was when
1851executing the (*SKIP) pattern.
1852
0b928c2f 1853Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1854is twice as long:
1855
d1fbf752
KW
1856 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1857 print "Count=$count\n";
24b23f37
YO
1858
1859outputs
1860
1861 aaab
1862 aaab
1863 Count=2
1864
5d458dd8 1865Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1866executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1867C<(*SKIP)> was executed.
1868
5d458dd8 1869=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1870X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1871
1872This zero-width pattern can be used to mark the point reached in a string
1873when a certain part of the pattern has been successfully matched. This
1874mark may be given a name. A later C<(*SKIP)> pattern will then skip
1875forward to that point if backtracked into on failure. Any number of
b4222fa9 1876C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1877
1878In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1879can be used to "label" a pattern branch, so that after matching, the
1880program can determine which branches of the pattern were involved in the
1881match.
1882
1883When a match is successful, the C<$REGMARK> variable will be set to the
1884name of the most recently executed C<(*MARK:NAME)> that was involved
1885in the match.
1886
1887This can be used to determine which branch of a pattern was matched
c27a5cfe 1888without using a separate capture group for each branch, which in turn
5d458dd8
YO
1889can result in a performance improvement, as perl cannot optimize
1890C</(?:(x)|(y)|(z))/> as efficiently as something like
1891C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1892
1893When a match has failed, and unless another verb has been involved in
1894failing the match and has provided its own name to use, the C<$REGERROR>
1895variable will be set to the name of the most recently executed
1896C<(*MARK:NAME)>.
1897
42ac7c82 1898See L</(*SKIP)> for more details.
5d458dd8 1899
b62d2d15
YO
1900As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1901
5d458dd8
YO
1902=item C<(*THEN)> C<(*THEN:NAME)>
1903
ac9d8485 1904This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1905C<(*PRUNE)>, this verb always matches, and when backtracked into on
1906failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
1907innermost enclosing group (capturing or otherwise) that has alternations.
1908The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1909count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
1910
1911Its name comes from the observation that this operation combined with the
1912alternation operator (C<|>) can be used to create what is essentially a
1913pattern-based if/then/else block:
1914
1915 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1916
1917Note that if this operator is used and NOT inside of an alternation then
1918it acts exactly like the C<(*PRUNE)> operator.
1919
1920 / A (*PRUNE) B /
1921
1922is the same as
1923
1924 / A (*THEN) B /
1925
1926but
1927
25e26d77 1928 / ( A (*THEN) B | C ) /
5d458dd8
YO
1929
1930is not the same as
1931
25e26d77 1932 / ( A (*PRUNE) B | C ) /
5d458dd8
YO
1933
1934as after matching the A but failing on the B the C<(*THEN)> verb will
1935backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1936
cbeadc21
JV
1937=back
1938
1939=item Verbs without an argument
1940
1941=over 4
1942
e2e6a0f1
YO
1943=item C<(*COMMIT)>
1944X<(*COMMIT)>
24b23f37 1945
241e7389 1946This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1947zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1948into on failure it causes the match to fail outright. No further attempts
1949to find a valid match by advancing the start pointer will occur again.
1950For example,
24b23f37 1951
d1fbf752
KW
1952 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
1953 print "Count=$count\n";
24b23f37
YO
1954
1955outputs
1956
1957 aaab
1958 Count=1
1959
e2e6a0f1
YO
1960In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1961does not match, the regex engine will not try any further matching on the
1962rest of the string.
c277df42 1963
e2e6a0f1
YO
1964=item C<(*FAIL)> C<(*F)>
1965X<(*FAIL)> X<(*F)>
9af228c6 1966
e2e6a0f1
YO
1967This pattern matches nothing and always fails. It can be used to force the
1968engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1969fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1970
e2e6a0f1 1971It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1972
e2e6a0f1
YO
1973=item C<(*ACCEPT)>
1974X<(*ACCEPT)>
9af228c6 1975
e2e6a0f1
YO
1976B<WARNING:> This feature is highly experimental. It is not recommended
1977for production code.
9af228c6 1978
e2e6a0f1
YO
1979This pattern matches nothing and causes the end of successful matching at
1980the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1981whether there is actually more to match in the string. When inside of a
0d017f4d 1982nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1983via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1984
c27a5cfe 1985If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1986marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1987For instance:
9af228c6 1988
e2e6a0f1 1989 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1990
e2e6a0f1 1991will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 1992be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 1993string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1994
1995=back
c277df42 1996
a0d0e21e
LW
1997=back
1998
c07a80fd 1999=head2 Backtracking
d74e8afc 2000X<backtrack> X<backtracking>
c07a80fd 2001
35a734be
IZ
2002NOTE: This section presents an abstract approximation of regular
2003expression behavior. For a more rigorous (and complicated) view of
2004the rules involved in selecting a match among possible alternatives,
0d017f4d 2005see L<Combining RE Pieces>.
35a734be 2006
c277df42 2007A fundamental feature of regular expression matching involves the
5a964f20 2008notion called I<backtracking>, which is currently used (when needed)
0d017f4d 2009by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
2010C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
2011internally, but the general principle outlined here is valid.
c07a80fd
PP
2012
2013For a regular expression to match, the I<entire> regular expression must
2014match, not just part of it. So if the beginning of a pattern containing a
2015quantifier succeeds in a way that causes later parts in the pattern to
2016fail, the matching engine backs up and recalculates the beginning
2017part--that's why it's called backtracking.
2018
2019Here is an example of backtracking: Let's say you want to find the
2020word following "foo" in the string "Food is on the foo table.":
2021
2022 $_ = "Food is on the foo table.";
2023 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 2024 print "$2 follows $1.\n";
c07a80fd
PP
2025 }
2026
2027When the match runs, the first part of the regular expression (C<\b(foo)>)
2028finds a possible match right at the beginning of the string, and loads up
2029$1 with "Foo". However, as soon as the matching engine sees that there's
2030no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 2031mistake and starts over again one character after where it had the
c07a80fd
PP
2032tentative match. This time it goes all the way until the next occurrence
2033of "foo". The complete regular expression matches this time, and you get
2034the expected output of "table follows foo."
2035
2036Sometimes minimal matching can help a lot. Imagine you'd like to match
2037everything between "foo" and "bar". Initially, you write something
2038like this:
2039
2040 $_ = "The food is under the bar in the barn.";
2041 if ( /foo(.*)bar/ ) {
f793d64a 2042 print "got <$1>\n";
c07a80fd
PP
2043 }
2044
2045Which perhaps unexpectedly yields:
2046
2047 got <d is under the bar in the >
2048
2049That's because C<.*> was greedy, so you get everything between the
14218588 2050I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
2051to use minimal matching to make sure you get the text between a "foo"
2052and the first "bar" thereafter.
2053
2054 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2055 got <d is under the >
2056
0d017f4d 2057Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2058of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2059So you write this:
2060
2061 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2062 if ( /(.*)(\d*)/ ) { # Wrong!
2063 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2064 }
2065
2066That won't work at all, because C<.*> was greedy and gobbled up the
2067whole string. As C<\d*> can match on an empty string the complete
2068regular expression matched successfully.
2069
8e1088bc 2070 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2071
2072Here are some variants, most of which don't work:
2073
2074 $_ = "I have 2 numbers: 53147";
2075 @pats = qw{
f793d64a
KW
2076 (.*)(\d*)
2077 (.*)(\d+)
2078 (.*?)(\d*)
2079 (.*?)(\d+)
2080 (.*)(\d+)$
2081 (.*?)(\d+)$
2082 (.*)\b(\d+)$
2083 (.*\D)(\d+)$
c07a80fd
PP
2084 };
2085
2086 for $pat (@pats) {
f793d64a
KW
2087 printf "%-12s ", $pat;
2088 if ( /$pat/ ) {
2089 print "<$1> <$2>\n";
2090 } else {
2091 print "FAIL\n";
2092 }
c07a80fd
PP
2093 }
2094
2095That will print out:
2096
2097 (.*)(\d*) <I have 2 numbers: 53147> <>
2098 (.*)(\d+) <I have 2 numbers: 5314> <7>
2099 (.*?)(\d*) <> <>
2100 (.*?)(\d+) <I have > <2>
2101 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2102 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2103 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2104 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2105
2106As you see, this can be a bit tricky. It's important to realize that a
2107regular expression is merely a set of assertions that gives a definition
2108of success. There may be 0, 1, or several different ways that the
2109definition might succeed against a particular string. And if there are
5a964f20
TC
2110multiple ways it might succeed, you need to understand backtracking to
2111know which variety of success you will achieve.
c07a80fd 2112
19799a22 2113When using look-ahead assertions and negations, this can all get even
8b19b778 2114trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2115followed by "123". You might try to write that as
2116
871b0233 2117 $_ = "ABC123";
f793d64a
KW
2118 if ( /^\D*(?!123)/ ) { # Wrong!
2119 print "Yup, no 123 in $_\n";
871b0233 2120 }
c07a80fd
PP
2121
2122But that isn't going to match; at least, not the way you're hoping. It
2123claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2124why that pattern matches, contrary to popular expectations:
c07a80fd 2125
4358a253
SS
2126 $x = 'ABC123';
2127 $y = 'ABC445';
c07a80fd 2128
4358a253
SS
2129 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2130 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2131
4358a253
SS
2132 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2133 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2134
2135This prints
2136
2137 2: got ABC
2138 3: got AB
2139 4: got ABC
2140
5f05dabc 2141You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2142general purpose version of test 1. The important difference between
2143them is that test 3 contains a quantifier (C<\D*>) and so can use
2144backtracking, whereas test 1 will not. What's happening is
2145that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2146non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2147let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2148fail.
14218588 2149
c07a80fd 2150The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2151try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2152a quantifier (C<\D*>) has been used in the regular expression, the
2153search engine can backtrack and retry the match differently
54310121 2154in the hope of matching the complete regular expression.
c07a80fd 2155
5a964f20
TC
2156The pattern really, I<really> wants to succeed, so it uses the
2157standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2158time. Now there's indeed something following "AB" that is not
14218588 2159"123". It's "C123", which suffices.
c07a80fd 2160
14218588
GS
2161We can deal with this by using both an assertion and a negation.
2162We'll say that the first part in $1 must be followed both by a digit
2163and by something that's not "123". Remember that the look-aheads
2164are zero-width expressions--they only look, but don't consume any
2165of the string in their match. So rewriting this way produces what
c07a80fd
PP
2166you'd expect; that is, case 5 will fail, but case 6 succeeds:
2167
4358a253
SS
2168 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2169 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2170
2171 6: got ABC
2172
5a964f20 2173In other words, the two zero-width assertions next to each other work as though
19799a22 2174they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2175matches only if you're at the beginning of the line AND the end of the
2176line simultaneously. The deeper underlying truth is that juxtaposition in
2177regular expressions always means AND, except when you write an explicit OR
2178using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2179although the attempted matches are made at different positions because "a"
2180is not a zero-width assertion, but a one-width assertion.
2181
0d017f4d 2182B<WARNING>: Particularly complicated regular expressions can take
14218588 2183exponential time to solve because of the immense number of possible
0d017f4d 2184ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2185internal optimizations done by the regular expression engine, this will
2186take a painfully long time to run:
c07a80fd 2187
e1901655
IZ
2188 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2189
2190And if you used C<*>'s in the internal groups instead of limiting them
2191to 0 through 5 matches, then it would take forever--or until you ran
2192out of stack space. Moreover, these internal optimizations are not
2193always applicable. For example, if you put C<{0,5}> instead of C<*>
2194on the external group, no current optimization is applicable, and the
2195match takes a long time to finish.
c07a80fd 2196
9da458fc
IZ
2197A powerful tool for optimizing such beasts is what is known as an
2198"independent group",
96090e4f 2199which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2200zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2201the tail match, since they are in "logical" context: only
14218588 2202whether they match is considered relevant. For an example
9da458fc 2203where side-effects of look-ahead I<might> have influenced the
96090e4f 2204following match, see L</C<< (?>pattern) >>>.
c277df42 2205
a0d0e21e 2206=head2 Version 8 Regular Expressions
d74e8afc 2207X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2208
5a964f20 2209In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2210routines, here are the pattern-matching rules not described above.
2211
54310121 2212Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2213with a special meaning described here or above. You can cause
5a964f20 2214characters that normally function as metacharacters to be interpreted
5f05dabc 2215literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2216character; "\\" matches a "\"). This escape mechanism is also required
2217for the character used as the pattern delimiter.
2218
2219A series of characters matches that series of characters in the target
0b928c2f 2220string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2221string.
a0d0e21e
LW
2222
2223You can specify a character class, by enclosing a list of characters
5d458dd8 2224in C<[]>, which will match any character from the list. If the
a0d0e21e 2225first character after the "[" is "^", the class matches any character not
14218588 2226in the list. Within a list, the "-" character specifies a
5a964f20 2227range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2228inclusive. If you want either "-" or "]" itself to be a member of a
2229class, put it at the start of the list (possibly after a "^"), or
2230escape it with a backslash. "-" is also taken literally when it is
2231at the end of the list, just before the closing "]". (The
84850974
DD
2232following all specify the same class of three characters: C<[-az]>,
2233C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2234specifies a class containing twenty-six characters, even on EBCDIC-based
2235character sets.) Also, if you try to use the character
2236classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2237a range, the "-" is understood literally.
a0d0e21e 2238
8ada0baa
JH
2239Note also that the whole range idea is rather unportable between
2240character sets--and even within character sets they may cause results
2241you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2242that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2243[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2244spell out the character sets in full.
2245
54310121 2246Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2247used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2248"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2249of three octal digits, matches the character whose coded character set value
5d458dd8 2250is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2251matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2252matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2253matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2254
2255You can specify a series of alternatives for a pattern using "|" to
2256separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2257or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2258first alternative includes everything from the last pattern delimiter
0b928c2f 2259("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2260the last alternative contains everything from the last "|" to the next
0b928c2f 2261closing pattern delimiter. That's why it's common practice to include
14218588 2262alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2263start and end.
2264
5a964f20 2265Alternatives are tried from left to right, so the first
a3cb178b
GS
2266alternative found for which the entire expression matches, is the one that
2267is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2268example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2269part will match, as that is the first alternative tried, and it successfully
2270matches the target string. (This might not seem important, but it is
2271important when you are capturing matched text using parentheses.)
2272
5a964f20 2273Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2274so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2275
14218588
GS
2276Within a pattern, you may designate subpatterns for later reference
2277by enclosing them in parentheses, and you may refer back to the
2278I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2279\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2280of their opening parenthesis. A backreference matches whatever
2281actually matched the subpattern in the string being examined, not
d8b950dc 2282the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2283match "0x1234 0x4321", but not "0x1234 01234", because subpattern
22841 matched "0x", even though the rule C<0|0x> could potentially match
2285the leading 0 in the second number.
cb1a09d0 2286
0d017f4d 2287=head2 Warning on \1 Instead of $1
cb1a09d0 2288
5a964f20 2289Some people get too used to writing things like:
cb1a09d0
AD
2290
2291 $pattern =~ s/(\W)/\\\1/g;
2292
3ff1c45a
KW
2293This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2294shocking the
cb1a09d0 2295B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2296PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2297the usual double-quoted string means a control-A. The customary Unix
2298meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2299of doing that, you get yourself into trouble if you then add an C</e>
2300modifier.
2301
f793d64a 2302 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2303
2304Or if you try to do
2305
2306 s/(\d+)/\1000/;
2307
2308You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2309C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2310with the operation of matching a backreference. Certainly they mean two
2311different things on the I<left> side of the C<s///>.
9fa51da4 2312
0d017f4d 2313=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2314
19799a22 2315B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2316
2317Regular expressions provide a terse and powerful programming language. As
2318with most other power tools, power comes together with the ability
2319to wreak havoc.
2320
2321A common abuse of this power stems from the ability to make infinite
628afcb5 2322loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2323
2324 'foo' =~ m{ ( o? )* }x;
2325
0d017f4d 2326The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2327in the string is not moved by the match, C<o?> would match again and again
527e91da 2328because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2329is with the looping modifier C<//g>:
2330
2331 @matches = ( 'foo' =~ m{ o? }xg );
2332
2333or
2334
2335 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2336
2337or the loop implied by split().
2338
2339However, long experience has shown that many programming tasks may
14218588
GS
2340be significantly simplified by using repeated subexpressions that
2341may match zero-length substrings. Here's a simple example being:
c84d73f1 2342
d1fbf752 2343 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2344 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2345
9da458fc 2346Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2347the infinite loop>. The rules for this are different for lower-level
527e91da 2348loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2349ones like the C</g> modifier or split() operator.
2350
19799a22
GS
2351The lower-level loops are I<interrupted> (that is, the loop is
2352broken) when Perl detects that a repeated expression matched a
2353zero-length substring. Thus
c84d73f1
IZ
2354
2355 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2356
5d458dd8 2357is made equivalent to
c84d73f1 2358
0b928c2f
FC
2359 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2360
2361For example, this program
2362
2363 #!perl -l
2364 "aaaaab" =~ /
2365 (?:
2366 a # non-zero
2367 | # or
2368 (?{print "hello"}) # print hello whenever this
2369 # branch is tried
2370 (?=(b)) # zero-width assertion
2371 )* # any number of times
2372 /x;
2373 print $&;
2374 print $1;
c84d73f1 2375
0b928c2f
FC
2376prints
2377
2378 hello
2379 aaaaa
2380 b
2381
2382Notice that "hello" is only printed once, as when Perl sees that the sixth
2383iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2384the C<*>.
2385
2386The higher-level loops preserve an additional state between iterations:
5d458dd8 2387whether the last match was zero-length. To break the loop, the following
c84d73f1 2388match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2389This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2390and so the I<second best> match is chosen if the I<best> match is of
2391zero length.
2392
19799a22 2393For example:
c84d73f1
IZ
2394
2395 $_ = 'bar';
2396 s/\w??/<$&>/g;
2397
20fb949f 2398results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2399match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2400best> match is what is matched by C<\w>. Thus zero-length matches
2401alternate with one-character-long matches.
2402
5d458dd8 2403Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2404position one notch further in the string.
2405
19799a22 2406The additional state of being I<matched with zero-length> is associated with
c84d73f1 2407the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2408Zero-length matches at the end of the previous match are ignored
2409during C<split>.
c84d73f1 2410
0d017f4d 2411=head2 Combining RE Pieces
35a734be
IZ
2412
2413Each of the elementary pieces of regular expressions which were described
2414before (such as C<ab> or C<\Z>) could match at most one substring
2415at the given position of the input string. However, in a typical regular
2416expression these elementary pieces are combined into more complicated
0b928c2f 2417patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2418(in these examples C<S> and C<T> are regular subexpressions).
2419
2420Such combinations can include alternatives, leading to a problem of choice:
2421if we match a regular expression C<a|ab> against C<"abc">, will it match
2422substring C<"a"> or C<"ab">? One way to describe which substring is
2423actually matched is the concept of backtracking (see L<"Backtracking">).
2424However, this description is too low-level and makes you think
2425in terms of a particular implementation.
2426
2427Another description starts with notions of "better"/"worse". All the
2428substrings which may be matched by the given regular expression can be
2429sorted from the "best" match to the "worst" match, and it is the "best"
2430match which is chosen. This substitutes the question of "what is chosen?"
2431by the question of "which matches are better, and which are worse?".
2432
2433Again, for elementary pieces there is no such question, since at most
2434one match at a given position is possible. This section describes the
2435notion of better/worse for combining operators. In the description
2436below C<S> and C<T> are regular subexpressions.
2437
13a2d996 2438=over 4
35a734be
IZ
2439
2440=item C<ST>
2441
2442Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2443substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2444which can be matched by C<T>.
35a734be 2445
0b928c2f 2446If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2447match than C<A'B'>.
2448
2449If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2450C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2451
2452=item C<S|T>
2453
2454When C<S> can match, it is a better match than when only C<T> can match.
2455
2456Ordering of two matches for C<S> is the same as for C<S>. Similar for
2457two matches for C<T>.
2458
2459=item C<S{REPEAT_COUNT}>
2460
2461Matches as C<SSS...S> (repeated as many times as necessary).
2462
2463=item C<S{min,max}>
2464
2465Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2466
2467=item C<S{min,max}?>
2468
2469Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2470
2471=item C<S?>, C<S*>, C<S+>
2472
2473Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2474
2475=item C<S??>, C<S*?>, C<S+?>
2476
2477Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2478
c47ff5f1 2479=item C<< (?>S) >>
35a734be
IZ
2480
2481Matches the best match for C<S> and only that.
2482
2483=item C<(?=S)>, C<(?<=S)>
2484
2485Only the best match for C<S> is considered. (This is important only if
2486C<S> has capturing parentheses, and backreferences are used somewhere
2487else in the whole regular expression.)
2488
2489=item C<(?!S)>, C<(?<!S)>
2490
2491For this grouping operator there is no need to describe the ordering, since
2492only whether or not C<S> can match is important.
2493
6bda09f9 2494=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2495
2496The ordering is the same as for the regular expression which is
c27a5cfe 2497the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
2498
2499=item C<(?(condition)yes-pattern|no-pattern)>
2500
2501Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2502already determined. The ordering of the matches is the same as for the
2503chosen subexpression.
2504
2505=back
2506
2507The above recipes describe the ordering of matches I<at a given position>.
2508One more rule is needed to understand how a match is determined for the
2509whole regular expression: a match at an earlier position is always better
2510than a match at a later position.
2511
0d017f4d 2512=head2 Creating Custom RE Engines
c84d73f1 2513
0b928c2f
FC
2514As of Perl 5.10.0, one can create custom regular expression engines. This
2515is not for the faint of heart, as they have to plug in at the C level. See
2516L<perlreapi> for more details.
2517
2518As an alternative, overloaded constants (see L<overload>) provide a simple
2519way to extend the functionality of the RE engine, by substituting one
2520pattern for another.
c84d73f1
IZ
2521
2522Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2523matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2524characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2525at these positions, so we want to have each C<\Y|> in the place of the
2526more complicated version. We can create a module C<customre> to do
2527this:
2528
2529 package customre;
2530 use overload;
2531
2532 sub import {
2533 shift;
2534 die "No argument to customre::import allowed" if @_;
2535 overload::constant 'qr' => \&convert;
2536 }
2537
2538 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2539
580a9fe1
RGS
2540 # We must also take care of not escaping the legitimate \\Y|
2541 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2542 my %rules = ( '\\' => '\\\\',
f793d64a 2543 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2544 sub convert {
2545 my $re = shift;
5d458dd8 2546 $re =~ s{
c84d73f1
IZ
2547 \\ ( \\ | Y . )
2548 }
5d458dd8 2549 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2550 return $re;
2551 }
2552
2553Now C<use customre> enables the new escape in constant regular
2554expressions, i.e., those without any runtime variable interpolations.
2555As documented in L<overload>, this conversion will work only over
2556literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2557part of this regular expression needs to be converted explicitly
2558(but only if the special meaning of C<\Y|> should be enabled inside $re):
2559
2560 use customre;
2561 $re = <>;
2562 chomp $re;
2563 $re = customre::convert $re;
2564 /\Y|$re\Y|/;
2565
0b928c2f 2566=head2 PCRE/Python Support
1f1031fe 2567
0b928c2f 2568As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2569to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2570Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2571
2572=over 4
2573
ae5648b3 2574=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2575
c27a5cfe 2576Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2577
2578=item C<< (?P=NAME) >>
2579
c27a5cfe 2580Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2581
2582=item C<< (?P>NAME) >>
2583
c27a5cfe 2584Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2585
ee9b8eae 2586=back
1f1031fe 2587
19799a22
GS
2588=head1 BUGS
2589
88c9975e
KW
2590Many regular expression constructs don't work on EBCDIC platforms.
2591
ed7efc79
KW
2592There are a number of issues with regard to case-insensitive matching
2593in Unicode rules. See C<i> under L</Modifiers> above.
2594
9da458fc
IZ
2595This document varies from difficult to understand to completely
2596and utterly opaque. The wandering prose riddled with jargon is
2597hard to fathom in several places.
2598
2599This document needs a rewrite that separates the tutorial content
2600from the reference content.
19799a22
GS
2601
2602=head1 SEE ALSO
9fa51da4 2603
91e0c79e
MJD
2604L<perlrequick>.
2605
2606L<perlretut>.
2607
9b599b2a
GS
2608L<perlop/"Regexp Quote-Like Operators">.
2609
1e66bd83
PP
2610L<perlop/"Gory details of parsing quoted constructs">.
2611
14218588
GS
2612L<perlfaq6>.
2613
9b599b2a
GS
2614L<perlfunc/pos>.
2615
2616L<perllocale>.
2617
fb55449c
JH
2618L<perlebcdic>.
2619
14218588
GS
2620I<Mastering Regular Expressions> by Jeffrey Friedl, published
2621by O'Reilly and Associates.