This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
[perl #112522] Mildly incorrect wording in "perldoc perlre"
[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 "$" will match before any newline. At the
a0d0e21e
LW
492cost of a little more overhead, you can do this by using the /m modifier
493on the pattern match operator. (Older programs did this by setting C<$*>,
0b928c2f 494but this option was removed in perl 5.9.)
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<$`>
897if you can, but if you can't (and some algorithms really appreciate
898them), once you've used them once, use them at will, because you've
899already paid the price. As of 5.005, C<$&> is not so costly as the
900other two.
d74e8afc 901X<$&> X<$`> X<$'>
68dc0745 902
99d59c4d 903As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
904C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
905and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 906successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
907The use of these variables incurs no global performance penalty, unlike
908their punctuation char equivalents, however at the trade-off that you
909have to tell perl when you want to use them.
87e95b7f 910X</p> X<p modifier>
cde0cee5 911
9d727203
KW
912=head2 Quoting metacharacters
913
19799a22
GS
914Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
915C<\w>, C<\n>. Unlike some other regular expression languages, there
916are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 917that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
918interpreted as a literal character, not a metacharacter. This was
919once used in a common idiom to disable or quote the special meanings
920of regular expression metacharacters in a string that you want to
36bbe248 921use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
922
923 $pattern =~ s/(\W)/\\$1/g;
924
f1cbbd6e 925(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
926Today it is more common to use the quotemeta() function or the C<\Q>
927metaquoting escape sequence to disable all metacharacters' special
928meanings like this:
a0d0e21e
LW
929
930 /$unquoted\Q$quoted\E$unquoted/
931
9da458fc
IZ
932Beware that if you put literal backslashes (those not inside
933interpolated variables) between C<\Q> and C<\E>, double-quotish
934backslash interpolation may lead to confusing results. If you
935I<need> to use literal backslashes within C<\Q...\E>,
936consult L<perlop/"Gory details of parsing quoted constructs">.
937
736fe711
KW
938C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
939
19799a22
GS
940=head2 Extended Patterns
941
14218588 942Perl also defines a consistent extension syntax for features not
0b928c2f
FC
943found in standard tools like B<awk> and
944B<lex>. The syntax for most of these is a
14218588
GS
945pair of parentheses with a question mark as the first thing within
946the parentheses. The character after the question mark indicates
947the extension.
19799a22 948
14218588
GS
949The stability of these extensions varies widely. Some have been
950part of the core language for many years. Others are experimental
951and may change without warning or be completely removed. Check
952the documentation on an individual feature to verify its current
953status.
19799a22 954
14218588
GS
955A question mark was chosen for this and for the minimal-matching
956construct because 1) question marks are rare in older regular
957expressions, and 2) whenever you see one, you should stop and
0b928c2f 958"question" exactly what is going on. That's psychology....
a0d0e21e 959
70ca8714 960=over 4
a0d0e21e 961
cc6b7395 962=item C<(?#text)>
d74e8afc 963X<(?#)>
a0d0e21e 964
14218588 965A comment. The text is ignored. If the C</x> modifier enables
19799a22 966whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
967the comment as soon as it sees a C<)>, so there is no way to put a literal
968C<)> in the comment.
a0d0e21e 969
cfaf538b 970=item C<(?adlupimsx-imsx)>
fb85c044 971
cfaf538b 972=item C<(?^alupimsx)>
fb85c044 973X<(?)> X<(?^)>
19799a22 974
0b6d1084
JH
975One or more embedded pattern-match modifiers, to be turned on (or
976turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
977the remainder of the enclosing pattern group (if any).
978
fb85c044 979This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 980configuration file, taken from an argument, or specified in a table
0b928c2f
FC
981somewhere. Consider the case where some patterns want to be
982case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 983include C<(?i)> at the front of the pattern. For example:
19799a22
GS
984
985 $pattern = "foobar";
5d458dd8 986 if ( /$pattern/i ) { }
19799a22
GS
987
988 # more flexible:
989
990 $pattern = "(?i)foobar";
5d458dd8 991 if ( /$pattern/ ) { }
19799a22 992
0b6d1084 993These modifiers are restored at the end of the enclosing group. For example,
19799a22 994
d8b950dc 995 ( (?i) blah ) \s+ \g1
19799a22 996
0d017f4d
WL
997will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
998repetition of the previous word, assuming the C</x> modifier, and no C</i>
999modifier outside this group.
19799a22 1000
8eb5594e 1001These modifiers do not carry over into named subpatterns called in the
dd72e27b 1002enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
8eb5594e
DR
1003change the case-sensitivity of the "NAME" pattern.
1004
dc925305
KW
1005Any of these modifiers can be set to apply globally to all regular
1006expressions compiled within the scope of a C<use re>. See
a0bbd6ff 1007L<re/"'/flags' mode">.
dc925305 1008
9de15fec
KW
1009Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1010after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
1011C<"d">) may follow the caret to override it.
1012But a minus sign is not legal with it.
1013
dc925305 1014Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 1015that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 1016C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
1017others, and a maximum of one (or two C<a>'s) may appear in the
1018construct. Thus, for
0b928c2f 1019example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 1020C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1021
1022Note also that the C<p> modifier is special in that its presence
1023anywhere in a pattern has a global effect.
cde0cee5 1024
5a964f20 1025=item C<(?:pattern)>
d74e8afc 1026X<(?:)>
a0d0e21e 1027
cfaf538b 1028=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1029
cfaf538b 1030=item C<(?^aluimsx:pattern)>
fb85c044
KW
1031X<(?^:)>
1032
5a964f20
TC
1033This is for clustering, not capturing; it groups subexpressions like
1034"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1035
5a964f20 1036 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1037
1038is like
1039
5a964f20 1040 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1041
19799a22
GS
1042but doesn't spit out extra fields. It's also cheaper not to capture
1043characters if you don't need to.
a0d0e21e 1044
19799a22 1045Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1046C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1047
1048 /(?s-i:more.*than).*million/i
1049
14218588 1050is equivalent to the more verbose
ca9dfc88
IZ
1051
1052 /(?:(?s-i)more.*than).*million/i
1053
fb85c044 1054Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1055after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1056flags (except C<"d">) may follow the caret, so
fb85c044
KW
1057
1058 (?^x:foo)
1059
1060is equivalent to
1061
1062 (?x-ims:foo)
1063
1064The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1065surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1066modified by any flags specified.
1067
1068The caret allows for simpler stringification of compiled regular
1069expressions. These look like
1070
1071 (?^:pattern)
1072
1073with any non-default flags appearing between the caret and the colon.
1074A test that looks at such stringification thus doesn't need to have the
1075system default flags hard-coded in it, just the caret. If new flags are
1076added to Perl, the meaning of the caret's expansion will change to include
1077the default for those flags, so the test will still work, unchanged.
1078
1079Specifying a negative flag after the caret is an error, as the flag is
1080redundant.
1081
1082Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1083to match at the beginning.
1084
594d7033
YO
1085=item C<(?|pattern)>
1086X<(?|)> X<Branch reset>
1087
1088This is the "branch reset" pattern, which has the special property
c27a5cfe 1089that the capture groups are numbered from the same starting point
99d59c4d 1090in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1091
c27a5cfe 1092Capture groups are numbered from left to right, but inside this
693596a8 1093construct the numbering is restarted for each branch.
4deaaa80 1094
c27a5cfe 1095The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1096following this construct will be numbered as though the construct
1097contained only one branch, that being the one with the most capture
c27a5cfe 1098groups in it.
4deaaa80 1099
0b928c2f 1100This construct is useful when you want to capture one of a
4deaaa80
PJ
1101number of alternative matches.
1102
1103Consider the following pattern. The numbers underneath show in
c27a5cfe 1104which group the captured content will be stored.
594d7033
YO
1105
1106
1107 # before ---------------branch-reset----------- after
1108 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1109 # 1 2 2 3 2 3 4
1110
ab106183
A
1111Be careful when using the branch reset pattern in combination with
1112named captures. Named captures are implemented as being aliases to
c27a5cfe 1113numbered groups holding the captures, and that interferes with the
ab106183
A
1114implementation of the branch reset pattern. If you are using named
1115captures in a branch reset pattern, it's best to use the same names,
1116in the same order, in each of the alternations:
1117
1118 /(?| (?<a> x ) (?<b> y )
1119 | (?<a> z ) (?<b> w )) /x
1120
1121Not doing so may lead to surprises:
1122
1123 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1124 say $+ {a}; # Prints '12'
1125 say $+ {b}; # *Also* prints '12'.
1126
c27a5cfe
KW
1127The problem here is that both the group named C<< a >> and the group
1128named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1129
ee9b8eae
YO
1130=item Look-Around Assertions
1131X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1132
0b928c2f 1133Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1134pattern without including it in C<$&>. Positive assertions match when
1135their subpattern matches, negative assertions match when their subpattern
1136fails. Look-behind matches text up to the current match position,
1137look-ahead matches text following the current match position.
1138
1139=over 4
1140
5a964f20 1141=item C<(?=pattern)>
d74e8afc 1142X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1143
19799a22 1144A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1145matches a word followed by a tab, without including the tab in C<$&>.
1146
5a964f20 1147=item C<(?!pattern)>
d74e8afc 1148X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1149
19799a22 1150A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1151matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1152however that look-ahead and look-behind are NOT the same thing. You cannot
1153use this for look-behind.
7b8d334a 1154
5a964f20 1155If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1156will not do what you want. That's because the C<(?!foo)> is just saying that
1157the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1158match. Use look-behind instead (see below).
c277df42 1159
ee9b8eae
YO
1160=item C<(?<=pattern)> C<\K>
1161X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1162
c47ff5f1 1163A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1164matches a word that follows a tab, without including the tab in C<$&>.
1165Works only for fixed-width look-behind.
c277df42 1166
ee9b8eae
YO
1167There is a special form of this construct, called C<\K>, which causes the
1168regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1169not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1170look-behind. The use of C<\K> inside of another look-around assertion
1171is allowed, but the behaviour is currently not well defined.
1172
c62285ac 1173For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1174equivalent C<< (?<=...) >> construct, and it is especially useful in
1175situations where you want to efficiently remove something following
1176something else in a string. For instance
1177
1178 s/(foo)bar/$1/g;
1179
1180can be rewritten as the much more efficient
1181
1182 s/foo\Kbar//g;
1183
5a964f20 1184=item C<(?<!pattern)>
d74e8afc 1185X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1186
19799a22
GS
1187A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1188matches any occurrence of "foo" that does not follow "bar". Works
1189only for fixed-width look-behind.
c277df42 1190
ee9b8eae
YO
1191=back
1192
81714fb9
YO
1193=item C<(?'NAME'pattern)>
1194
1195=item C<< (?<NAME>pattern) >>
1196X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1197
c27a5cfe 1198A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1199parentheses C<()> but for the additional fact that the group
1200can be referred to by name in various regular expression
1201constructs (like C<\g{NAME}>) and can be accessed by name
1202after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1203for more details on the C<%+> and C<%-> hashes.
81714fb9 1204
c27a5cfe
KW
1205If multiple distinct capture groups have the same name then the
1206$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1207
0d017f4d 1208The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1209
1210B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1211function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1212numbered sequentially regardless of being named or not. Thus in the
1213pattern
1214
1215 /(x)(?<foo>y)(z)/
1216
1217$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1218the opposite which is what a .NET regex hacker might expect.
1219
1f1031fe
YO
1220Currently NAME is restricted to simple identifiers only.
1221In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1222its Unicode extension (see L<utf8>),
1223though it isn't extended by the locale (see L<perllocale>).
81714fb9 1224
1f1031fe 1225B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1226with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1227may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1228support the use of single quotes as a delimiter for the name.
81714fb9 1229
1f1031fe
YO
1230=item C<< \k<NAME> >>
1231
1232=item C<< \k'NAME' >>
81714fb9
YO
1233
1234Named backreference. Similar to numeric backreferences, except that
1235the group is designated by name and not number. If multiple groups
1236have the same name then it refers to the leftmost defined group in
1237the current match.
1238
0d017f4d 1239It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1240earlier in the pattern.
1241
1242Both forms are equivalent.
1243
1f1031fe 1244B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1245with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1246may be used instead of C<< \k<NAME> >>.
1f1031fe 1247
cc6b7395 1248=item C<(?{ code })>
d74e8afc 1249X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1250
19799a22 1251B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1252experimental, and may be changed without notice. Code executed that
1253has side effects may not perform identically from version to version
1254due to the effect of future optimisations in the regex engine.
c277df42 1255
cc46d5f2 1256This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
1257always succeeds, and its C<code> is not interpolated. Currently,
1258the rules to determine where the C<code> ends are somewhat convoluted.
1259
77ea4f6d
JV
1260This feature can be used together with the special variable C<$^N> to
1261capture the results of submatches in variables without having to keep
1262track of the number of nested parentheses. For example:
1263
1264 $_ = "The brown fox jumps over the lazy dog";
1265 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1266 print "color = $color, animal = $animal\n";
1267
754091cb
RGS
1268Inside the C<(?{...})> block, C<$_> refers to the string the regular
1269expression is matching against. You can also use C<pos()> to know what is
fa11829f 1270the current position of matching within this string.
754091cb 1271
19799a22
GS
1272The C<code> is properly scoped in the following sense: If the assertion
1273is backtracked (compare L<"Backtracking">), all changes introduced after
1274C<local>ization are undone, so that
b9ac3b5b
GS
1275
1276 $_ = 'a' x 8;
5d458dd8 1277 m<
d1fbf752 1278 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1279 (
5d458dd8 1280 a
b9ac3b5b 1281 (?{
d1fbf752
KW
1282 local $cnt = $cnt + 1; # Update $cnt,
1283 # backtracking-safe.
b9ac3b5b 1284 })
5d458dd8 1285 )*
b9ac3b5b 1286 aaaa
d1fbf752
KW
1287 (?{ $res = $cnt }) # On success copy to
1288 # non-localized location.
b9ac3b5b
GS
1289 >x;
1290
0d017f4d 1291will set C<$res = 4>. Note that after the match, C<$cnt> returns to the globally
14218588 1292introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
1293are unwound.
1294
19799a22
GS
1295This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
1296switch. If I<not> used in this way, the result of evaluation of
1297C<code> is put into the special variable C<$^R>. This happens
1298immediately, so C<$^R> can be used from other C<(?{ code })> assertions
1299inside the same regular expression.
b9ac3b5b 1300
19799a22
GS
1301The assignment to C<$^R> above is properly localized, so the old
1302value of C<$^R> is restored if the assertion is backtracked; compare
1303L<"Backtracking">.
b9ac3b5b 1304
19799a22
GS
1305For reasons of security, this construct is forbidden if the regular
1306expression involves run-time interpolation of variables, unless the
1307perilous C<use re 'eval'> pragma has been used (see L<re>), or the
0b928c2f 1308variables contain results of the C<qr//> operator (see
b6fa137b 1309L<perlop/"qr/STRINGE<sol>msixpodual">).
871b0233 1310
0d017f4d 1311This restriction is due to the wide-spread and remarkably convenient
19799a22 1312custom of using run-time determined strings as patterns. For example:
871b0233
IZ
1313
1314 $re = <>;
1315 chomp $re;
1316 $string =~ /$re/;
1317
14218588
GS
1318Before Perl knew how to execute interpolated code within a pattern,
1319this operation was completely safe from a security point of view,
1320although it could raise an exception from an illegal pattern. If
1321you turn on the C<use re 'eval'>, though, it is no longer secure,
1322so you should only do so if you are also using taint checking.
1323Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 1324compartment. See L<perlsec> for details about both these mechanisms.
871b0233 1325
e95d7314
GG
1326B<WARNING>: Use of lexical (C<my>) variables in these blocks is
1327broken. The result is unpredictable and will make perl unstable. The
1328workaround is to use global (C<our>) variables.
1329
8525cfae
FC
1330B<WARNING>: In perl 5.12.x and earlier, the regex engine
1331was not re-entrant, so interpolated code could not
1332safely invoke the regex engine either directly with
e95d7314 1333C<m//> or C<s///>), or indirectly with functions such as
8525cfae 1334C<split>. Invoking the regex engine in these blocks would make perl
e95d7314 1335unstable.
8988a1bb 1336
14455d6c 1337=item C<(??{ code })>
d74e8afc
ITB
1338X<(??{})>
1339X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1340
19799a22 1341B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1342experimental, and may be changed without notice. Code executed that
1343has side effects may not perform identically from version to version
1344due to the effect of future optimisations in the regex engine.
0f5d15d6 1345
19799a22
GS
1346This is a "postponed" regular subexpression. The C<code> is evaluated
1347at run time, at the moment this subexpression may match. The result
0b928c2f 1348of evaluation is considered a regular expression and matched as
61528107 1349if it were inserted instead of this construct. Note that this means
c27a5cfe 1350that the contents of capture groups defined inside an eval'ed pattern
6bda09f9 1351are not available outside of the pattern, and vice versa, there is no
b2711834
FC
1352way for the inner pattern returned from the code block to refer to a
1353capture group defined outside. (The code block itself can use C<$1>, etc.,
6cf90346 1354to refer to the enclosing pattern's capture groups.) Thus,
6bda09f9
YO
1355
1356 ('a' x 100)=~/(??{'(.)' x 100})/
1357
81714fb9 1358B<will> match, it will B<not> set $1.
0f5d15d6 1359
428594d9 1360The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
1361where the C<code> ends are currently somewhat convoluted.
1362
1363The following pattern matches a parenthesized group:
0f5d15d6 1364
d1fbf752
KW
1365 $re = qr{
1366 \(
1367 (?:
1368 (?> [^()]+ ) # Non-parens without backtracking
1369 |
1370 (??{ $re }) # Group with matching parens
1371 )*
1372 \)
1373 }x;
0f5d15d6 1374
6bda09f9
YO
1375See also C<(?PARNO)> for a different, more efficient way to accomplish
1376the same task.
1377
0b370c0a
A
1378For reasons of security, this construct is forbidden if the regular
1379expression involves run-time interpolation of variables, unless the
1380perilous C<use re 'eval'> pragma has been used (see L<re>), or the
0b928c2f 1381variables contain results of the C<qr//> operator (see
b6fa137b 1382L<perlop/"qrE<sol>STRINGE<sol>msixpodual">).
0b370c0a 1383
8525cfae
FC
1384In perl 5.12.x and earlier, because the regex engine was not re-entrant,
1385delayed code could not safely invoke the regex engine either directly with
1386C<m//> or C<s///>), or indirectly with functions such as C<split>.
8988a1bb 1387
5d458dd8
YO
1388Recursing deeper than 50 times without consuming any input string will
1389result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
1390changing it requires a custom build.
1391
542fa716
YO
1392=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1393X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1394X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1395X<regex, relative recursion>
6bda09f9 1396
81714fb9 1397Similar to C<(??{ code })> except it does not involve compiling any code,
c27a5cfe
KW
1398instead it treats the contents of a capture group as an independent
1399pattern that must match at the current position. Capture groups
81714fb9 1400contained by the pattern will have the value as determined by the
6bda09f9
YO
1401outermost recursion.
1402
894be9b7 1403PARNO is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1404the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1405the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716 1406C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
c27a5cfe 1407to be relative, with negative numbers indicating preceding capture groups
542fa716 1408and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1409declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1410Note that the counting for relative recursion differs from that of
c27a5cfe 1411relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1412included.
6bda09f9 1413
81714fb9 1414The following pattern matches a function foo() which may contain
f145b7e9 1415balanced parentheses as the argument.
6bda09f9 1416
d1fbf752 1417 $re = qr{ ( # paren group 1 (full function)
81714fb9 1418 foo
d1fbf752 1419 ( # paren group 2 (parens)
6bda09f9 1420 \(
d1fbf752 1421 ( # paren group 3 (contents of parens)
6bda09f9 1422 (?:
d1fbf752 1423 (?> [^()]+ ) # Non-parens without backtracking
6bda09f9 1424 |
d1fbf752 1425 (?2) # Recurse to start of paren group 2
6bda09f9
YO
1426 )*
1427 )
1428 \)
1429 )
1430 )
1431 }x;
1432
1433If the pattern was used as follows
1434
1435 'foo(bar(baz)+baz(bop))'=~/$re/
1436 and print "\$1 = $1\n",
1437 "\$2 = $2\n",
1438 "\$3 = $3\n";
1439
1440the output produced should be the following:
1441
1442 $1 = foo(bar(baz)+baz(bop))
1443 $2 = (bar(baz)+baz(bop))
81714fb9 1444 $3 = bar(baz)+baz(bop)
6bda09f9 1445
c27a5cfe 1446If there is no corresponding capture group defined, then it is a
61528107 1447fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1448string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1449into perl, so changing it requires a custom build.
1450
542fa716
YO
1451The following shows how using negative indexing can make it
1452easier to embed recursive patterns inside of a C<qr//> construct
1453for later use:
1454
1455 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1456 if (/foo $parens \s+ + \s+ bar $parens/x) {
1457 # do something here...
1458 }
1459
81714fb9 1460B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1461PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1462a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1463as atomic. Also, modifiers are resolved at compile time, so constructs
1464like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1465be processed.
6bda09f9 1466
894be9b7
YO
1467=item C<(?&NAME)>
1468X<(?&NAME)>
1469
0d017f4d
WL
1470Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1471parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1472the same name, then it recurses to the leftmost.
1473
1474It is an error to refer to a name that is not declared somewhere in the
1475pattern.
1476
1f1031fe
YO
1477B<NOTE:> In order to make things easier for programmers with experience
1478with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1479may be used instead of C<< (?&NAME) >>.
1f1031fe 1480
e2e6a0f1
YO
1481=item C<(?(condition)yes-pattern|no-pattern)>
1482X<(?()>
286f584a 1483
e2e6a0f1 1484=item C<(?(condition)yes-pattern)>
286f584a 1485
41ef34de
ML
1486Conditional expression. Matches C<yes-pattern> if C<condition> yields
1487a true value, matches C<no-pattern> otherwise. A missing pattern always
1488matches.
1489
1490C<(condition)> should be either an integer in
e2e6a0f1
YO
1491parentheses (which is valid if the corresponding pair of parentheses
1492matched), a look-ahead/look-behind/evaluate zero-width assertion, a
c27a5cfe 1493name in angle brackets or single quotes (which is valid if a group
e2e6a0f1
YO
1494with the given name matched), or the special symbol (R) (true when
1495evaluated inside of recursion or eval). Additionally the R may be
1496followed by a number, (which will be true when evaluated when recursing
1497inside of the appropriate group), or by C<&NAME>, in which case it will
1498be true only when evaluated during recursion in the named group.
1499
1500Here's a summary of the possible predicates:
1501
1502=over 4
1503
1504=item (1) (2) ...
1505
c27a5cfe 1506Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1507
1508=item (<NAME>) ('NAME')
1509
c27a5cfe 1510Checks if a group with the given name has matched something.
e2e6a0f1 1511
f01cd190
FC
1512=item (?=...) (?!...) (?<=...) (?<!...)
1513
1514Checks whether the pattern matches (or does not match, for the '!'
1515variants).
1516
e2e6a0f1
YO
1517=item (?{ CODE })
1518
f01cd190 1519Treats the return value of the code block as the condition.
e2e6a0f1
YO
1520
1521=item (R)
1522
1523Checks if the expression has been evaluated inside of recursion.
1524
1525=item (R1) (R2) ...
1526
1527Checks if the expression has been evaluated while executing directly
1528inside of the n-th capture group. This check is the regex equivalent of
1529
1530 if ((caller(0))[3] eq 'subname') { ... }
1531
1532In other words, it does not check the full recursion stack.
1533
1534=item (R&NAME)
1535
1536Similar to C<(R1)>, this predicate checks to see if we're executing
1537directly inside of the leftmost group with a given name (this is the same
1538logic used by C<(?&NAME)> to disambiguate). It does not check the full
1539stack, but only the name of the innermost active recursion.
1540
1541=item (DEFINE)
1542
1543In this case, the yes-pattern is never directly executed, and no
1544no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1545See below for details.
1546
1547=back
1548
1549For example:
1550
1551 m{ ( \( )?
1552 [^()]+
1553 (?(1) \) )
1554 }x
1555
1556matches a chunk of non-parentheses, possibly included in parentheses
1557themselves.
1558
0b928c2f
FC
1559A special form is the C<(DEFINE)> predicate, which never executes its
1560yes-pattern directly, and does not allow a no-pattern. This allows one to
1561define subpatterns which will be executed only by the recursion mechanism.
e2e6a0f1
YO
1562This way, you can define a set of regular expression rules that can be
1563bundled into any pattern you choose.
1564
1565It is recommended that for this usage you put the DEFINE block at the
1566end of the pattern, and that you name any subpatterns defined within it.
1567
1568Also, it's worth noting that patterns defined this way probably will
1569not be as efficient, as the optimiser is not very clever about
1570handling them.
1571
1572An example of how this might be used is as follows:
1573
2bf803e2 1574 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1575 (?(DEFINE)
2bf803e2
YO
1576 (?<NAME_PAT>....)
1577 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1578 )/x
1579
c27a5cfe
KW
1580Note that capture groups matched inside of recursion are not accessible
1581after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1582necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1583C<$+{NAME}> would be.
286f584a 1584
51a1303c
BF
1585Finally, keep in mind that subpatterns created inside a DEFINE block
1586count towards the absolute and relative number of captures, so this:
1587
1588 my @captures = "a" =~ /(.) # First capture
1589 (?(DEFINE)
1590 (?<EXAMPLE> 1 ) # Second capture
1591 )/x;
1592 say scalar @captures;
1593
1594Will output 2, not 1. This is particularly important if you intend to
1595compile the definitions with the C<qr//> operator, and later
1596interpolate them in another pattern.
1597
c47ff5f1 1598=item C<< (?>pattern) >>
6bda09f9 1599X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1600
19799a22
GS
1601An "independent" subexpression, one which matches the substring
1602that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1603position, and it matches I<nothing other than this substring>. This
19799a22
GS
1604construct is useful for optimizations of what would otherwise be
1605"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1606It may also be useful in places where the "grab all you can, and do not
1607give anything back" semantic is desirable.
19799a22 1608
c47ff5f1 1609For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1610(anchored at the beginning of string, as above) will match I<all>
1611characters C<a> at the beginning of string, leaving no C<a> for
1612C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1613since the match of the subgroup C<a*> is influenced by the following
1614group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1615C<a*ab> will match fewer characters than a standalone C<a*>, since
1616this makes the tail match.
1617
0b928c2f
FC
1618C<< (?>pattern) >> does not disable backtracking altogether once it has
1619matched. It is still possible to backtrack past the construct, but not
1620into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1621
c47ff5f1 1622An effect similar to C<< (?>pattern) >> may be achieved by writing
0b928c2f
FC
1623C<(?=(pattern))\g{-1}>. This matches the same substring as a standalone
1624C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
c47ff5f1 1625makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1626(The difference between these two constructs is that the second one
1627uses a capturing group, thus shifting ordinals of backreferences
1628in the rest of a regular expression.)
1629
1630Consider this pattern:
c277df42 1631
871b0233 1632 m{ \(
e2e6a0f1 1633 (
f793d64a 1634 [^()]+ # x+
e2e6a0f1 1635 |
871b0233
IZ
1636 \( [^()]* \)
1637 )+
e2e6a0f1 1638 \)
871b0233 1639 }x
5a964f20 1640
19799a22
GS
1641That will efficiently match a nonempty group with matching parentheses
1642two levels deep or less. However, if there is no such group, it
1643will take virtually forever on a long string. That's because there
1644are so many different ways to split a long string into several
1645substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1646to a subpattern of the above pattern. Consider how the pattern
1647above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1648seconds, but that each extra letter doubles this time. This
1649exponential performance will make it appear that your program has
14218588 1650hung. However, a tiny change to this pattern
5a964f20 1651
e2e6a0f1
YO
1652 m{ \(
1653 (
f793d64a 1654 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1655 |
871b0233
IZ
1656 \( [^()]* \)
1657 )+
e2e6a0f1 1658 \)
871b0233 1659 }x
c277df42 1660
c47ff5f1 1661which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1662this yourself would be a productive exercise), but finishes in a fourth
1663the time when used on a similar string with 1000000 C<a>s. Be aware,
0b928c2f
FC
1664however, that, when this construct is followed by a
1665quantifier, it currently triggers a warning message under
9f1b1f2d 1666the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1667C<"matches null string many times in regex">.
c277df42 1668
c47ff5f1 1669On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1670effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1671This was only 4 times slower on a string with 1000000 C<a>s.
1672
9da458fc
IZ
1673The "grab all you can, and do not give anything back" semantic is desirable
1674in many situations where on the first sight a simple C<()*> looks like
1675the correct solution. Suppose we parse text with comments being delimited
1676by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1677its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1678the comment delimiter, because it may "give up" some whitespace if
1679the remainder of the pattern can be made to match that way. The correct
1680answer is either one of these:
1681
1682 (?>#[ \t]*)
1683 #[ \t]*(?![ \t])
1684
1685For example, to grab non-empty comments into $1, one should use either
1686one of these:
1687
1688 / (?> \# [ \t]* ) ( .+ ) /x;
1689 / \# [ \t]* ( [^ \t] .* ) /x;
1690
1691Which one you pick depends on which of these expressions better reflects
1692the above specification of comments.
1693
6bda09f9
YO
1694In some literature this construct is called "atomic matching" or
1695"possessive matching".
1696
b9b4dddf
YO
1697Possessive quantifiers are equivalent to putting the item they are applied
1698to inside of one of these constructs. The following equivalences apply:
1699
1700 Quantifier Form Bracketing Form
1701 --------------- ---------------
1702 PAT*+ (?>PAT*)
1703 PAT++ (?>PAT+)
1704 PAT?+ (?>PAT?)
1705 PAT{min,max}+ (?>PAT{min,max})
1706
e2e6a0f1
YO
1707=back
1708
1709=head2 Special Backtracking Control Verbs
1710
1711B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1712removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1713be noted to avoid problems during upgrades.
1714
1715These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1716otherwise stated the ARG argument is optional; in some cases, it is
1717forbidden.
1718
1719Any pattern containing a special backtracking verb that allows an argument
e1020413 1720has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1721C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1722rules apply:
e2e6a0f1 1723
5d458dd8
YO
1724On failure, the C<$REGERROR> variable will be set to the ARG value of the
1725verb pattern, if the verb was involved in the failure of the match. If the
1726ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1727name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1728none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1729
5d458dd8
YO
1730On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1731the C<$REGMARK> variable will be set to the name of the last
1732C<(*MARK:NAME)> pattern executed. See the explanation for the
1733C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1734
5d458dd8 1735B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1736and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1737readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1738Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1739
1740If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1741argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1742
70ca8714 1743=over 3
e2e6a0f1
YO
1744
1745=item Verbs that take an argument
1746
1747=over 4
1748
5d458dd8 1749=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1750X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1751
5d458dd8
YO
1752This zero-width pattern prunes the backtracking tree at the current point
1753when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1754where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1755A may backtrack as necessary to match. Once it is reached, matching
1756continues in B, which may also backtrack as necessary; however, should B
1757not match, then no further backtracking will take place, and the pattern
1758will fail outright at the current starting position.
54612592
YO
1759
1760The following example counts all the possible matching strings in a
1761pattern (without actually matching any of them).
1762
e2e6a0f1 1763 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1764 print "Count=$count\n";
1765
1766which produces:
1767
1768 aaab
1769 aaa
1770 aa
1771 a
1772 aab
1773 aa
1774 a
1775 ab
1776 a
1777 Count=9
1778
5d458dd8 1779If we add a C<(*PRUNE)> before the count like the following
54612592 1780
5d458dd8 1781 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1782 print "Count=$count\n";
1783
0b928c2f 1784we prevent backtracking and find the count of the longest matching string
353c6505 1785at each matching starting point like so:
54612592
YO
1786
1787 aaab
1788 aab
1789 ab
1790 Count=3
1791
5d458dd8 1792Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1793
5d458dd8
YO
1794See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1795control backtracking. In some cases, the use of C<(*PRUNE)> can be
1796replaced with a C<< (?>pattern) >> with no functional difference; however,
1797C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1798C<< (?>pattern) >> alone.
54612592 1799
5d458dd8
YO
1800=item C<(*SKIP)> C<(*SKIP:NAME)>
1801X<(*SKIP)>
e2e6a0f1 1802
5d458dd8 1803This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1804failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1805to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1806of this pattern. This effectively means that the regex engine "skips" forward
1807to this position on failure and tries to match again, (assuming that
1808there is sufficient room to match).
1809
1810The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1811C<(*MARK:NAME)> was encountered while matching, then it is that position
1812which is used as the "skip point". If no C<(*MARK)> of that name was
1813encountered, then the C<(*SKIP)> operator has no effect. When used
1814without a name the "skip point" is where the match point was when
1815executing the (*SKIP) pattern.
1816
0b928c2f 1817Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1818is twice as long:
1819
d1fbf752
KW
1820 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1821 print "Count=$count\n";
24b23f37
YO
1822
1823outputs
1824
1825 aaab
1826 aaab
1827 Count=2
1828
5d458dd8 1829Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1830executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1831C<(*SKIP)> was executed.
1832
5d458dd8 1833=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1834X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1835
1836This zero-width pattern can be used to mark the point reached in a string
1837when a certain part of the pattern has been successfully matched. This
1838mark may be given a name. A later C<(*SKIP)> pattern will then skip
1839forward to that point if backtracked into on failure. Any number of
b4222fa9 1840C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1841
1842In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1843can be used to "label" a pattern branch, so that after matching, the
1844program can determine which branches of the pattern were involved in the
1845match.
1846
1847When a match is successful, the C<$REGMARK> variable will be set to the
1848name of the most recently executed C<(*MARK:NAME)> that was involved
1849in the match.
1850
1851This can be used to determine which branch of a pattern was matched
c27a5cfe 1852without using a separate capture group for each branch, which in turn
5d458dd8
YO
1853can result in a performance improvement, as perl cannot optimize
1854C</(?:(x)|(y)|(z))/> as efficiently as something like
1855C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1856
1857When a match has failed, and unless another verb has been involved in
1858failing the match and has provided its own name to use, the C<$REGERROR>
1859variable will be set to the name of the most recently executed
1860C<(*MARK:NAME)>.
1861
42ac7c82 1862See L</(*SKIP)> for more details.
5d458dd8 1863
b62d2d15
YO
1864As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1865
5d458dd8
YO
1866=item C<(*THEN)> C<(*THEN:NAME)>
1867
ac9d8485 1868This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1869C<(*PRUNE)>, this verb always matches, and when backtracked into on
1870failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
1871innermost enclosing group (capturing or otherwise) that has alternations.
1872The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1873count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
1874
1875Its name comes from the observation that this operation combined with the
1876alternation operator (C<|>) can be used to create what is essentially a
1877pattern-based if/then/else block:
1878
1879 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1880
1881Note that if this operator is used and NOT inside of an alternation then
1882it acts exactly like the C<(*PRUNE)> operator.
1883
1884 / A (*PRUNE) B /
1885
1886is the same as
1887
1888 / A (*THEN) B /
1889
1890but
1891
1892 / ( A (*THEN) B | C (*THEN) D ) /
1893
1894is not the same as
1895
1896 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1897
1898as after matching the A but failing on the B the C<(*THEN)> verb will
1899backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1900
cbeadc21
JV
1901=back
1902
1903=item Verbs without an argument
1904
1905=over 4
1906
e2e6a0f1
YO
1907=item C<(*COMMIT)>
1908X<(*COMMIT)>
24b23f37 1909
241e7389 1910This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1911zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1912into on failure it causes the match to fail outright. No further attempts
1913to find a valid match by advancing the start pointer will occur again.
1914For example,
24b23f37 1915
d1fbf752
KW
1916 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
1917 print "Count=$count\n";
24b23f37
YO
1918
1919outputs
1920
1921 aaab
1922 Count=1
1923
e2e6a0f1
YO
1924In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1925does not match, the regex engine will not try any further matching on the
1926rest of the string.
c277df42 1927
e2e6a0f1
YO
1928=item C<(*FAIL)> C<(*F)>
1929X<(*FAIL)> X<(*F)>
9af228c6 1930
e2e6a0f1
YO
1931This pattern matches nothing and always fails. It can be used to force the
1932engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1933fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1934
e2e6a0f1 1935It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1936
e2e6a0f1
YO
1937=item C<(*ACCEPT)>
1938X<(*ACCEPT)>
9af228c6 1939
e2e6a0f1
YO
1940B<WARNING:> This feature is highly experimental. It is not recommended
1941for production code.
9af228c6 1942
e2e6a0f1
YO
1943This pattern matches nothing and causes the end of successful matching at
1944the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1945whether there is actually more to match in the string. When inside of a
0d017f4d 1946nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1947via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1948
c27a5cfe 1949If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1950marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1951For instance:
9af228c6 1952
e2e6a0f1 1953 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1954
e2e6a0f1 1955will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 1956be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 1957string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1958
1959=back
c277df42 1960
a0d0e21e
LW
1961=back
1962
c07a80fd 1963=head2 Backtracking
d74e8afc 1964X<backtrack> X<backtracking>
c07a80fd 1965
35a734be
IZ
1966NOTE: This section presents an abstract approximation of regular
1967expression behavior. For a more rigorous (and complicated) view of
1968the rules involved in selecting a match among possible alternatives,
0d017f4d 1969see L<Combining RE Pieces>.
35a734be 1970
c277df42 1971A fundamental feature of regular expression matching involves the
5a964f20 1972notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1973by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1974C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1975internally, but the general principle outlined here is valid.
c07a80fd
PP
1976
1977For a regular expression to match, the I<entire> regular expression must
1978match, not just part of it. So if the beginning of a pattern containing a
1979quantifier succeeds in a way that causes later parts in the pattern to
1980fail, the matching engine backs up and recalculates the beginning
1981part--that's why it's called backtracking.
1982
1983Here is an example of backtracking: Let's say you want to find the
1984word following "foo" in the string "Food is on the foo table.":
1985
1986 $_ = "Food is on the foo table.";
1987 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 1988 print "$2 follows $1.\n";
c07a80fd
PP
1989 }
1990
1991When the match runs, the first part of the regular expression (C<\b(foo)>)
1992finds a possible match right at the beginning of the string, and loads up
1993$1 with "Foo". However, as soon as the matching engine sees that there's
1994no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1995mistake and starts over again one character after where it had the
c07a80fd
PP
1996tentative match. This time it goes all the way until the next occurrence
1997of "foo". The complete regular expression matches this time, and you get
1998the expected output of "table follows foo."
1999
2000Sometimes minimal matching can help a lot. Imagine you'd like to match
2001everything between "foo" and "bar". Initially, you write something
2002like this:
2003
2004 $_ = "The food is under the bar in the barn.";
2005 if ( /foo(.*)bar/ ) {
f793d64a 2006 print "got <$1>\n";
c07a80fd
PP
2007 }
2008
2009Which perhaps unexpectedly yields:
2010
2011 got <d is under the bar in the >
2012
2013That's because C<.*> was greedy, so you get everything between the
14218588 2014I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
2015to use minimal matching to make sure you get the text between a "foo"
2016and the first "bar" thereafter.
2017
2018 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2019 got <d is under the >
2020
0d017f4d 2021Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2022of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2023So you write this:
2024
2025 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2026 if ( /(.*)(\d*)/ ) { # Wrong!
2027 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2028 }
2029
2030That won't work at all, because C<.*> was greedy and gobbled up the
2031whole string. As C<\d*> can match on an empty string the complete
2032regular expression matched successfully.
2033
8e1088bc 2034 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2035
2036Here are some variants, most of which don't work:
2037
2038 $_ = "I have 2 numbers: 53147";
2039 @pats = qw{
f793d64a
KW
2040 (.*)(\d*)
2041 (.*)(\d+)
2042 (.*?)(\d*)
2043 (.*?)(\d+)
2044 (.*)(\d+)$
2045 (.*?)(\d+)$
2046 (.*)\b(\d+)$
2047 (.*\D)(\d+)$
c07a80fd
PP
2048 };
2049
2050 for $pat (@pats) {
f793d64a
KW
2051 printf "%-12s ", $pat;
2052 if ( /$pat/ ) {
2053 print "<$1> <$2>\n";
2054 } else {
2055 print "FAIL\n";
2056 }
c07a80fd
PP
2057 }
2058
2059That will print out:
2060
2061 (.*)(\d*) <I have 2 numbers: 53147> <>
2062 (.*)(\d+) <I have 2 numbers: 5314> <7>
2063 (.*?)(\d*) <> <>
2064 (.*?)(\d+) <I have > <2>
2065 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2066 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2067 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2068 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2069
2070As you see, this can be a bit tricky. It's important to realize that a
2071regular expression is merely a set of assertions that gives a definition
2072of success. There may be 0, 1, or several different ways that the
2073definition might succeed against a particular string. And if there are
5a964f20
TC
2074multiple ways it might succeed, you need to understand backtracking to
2075know which variety of success you will achieve.
c07a80fd 2076
19799a22 2077When using look-ahead assertions and negations, this can all get even
8b19b778 2078trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2079followed by "123". You might try to write that as
2080
871b0233 2081 $_ = "ABC123";
f793d64a
KW
2082 if ( /^\D*(?!123)/ ) { # Wrong!
2083 print "Yup, no 123 in $_\n";
871b0233 2084 }
c07a80fd
PP
2085
2086But that isn't going to match; at least, not the way you're hoping. It
2087claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2088why that pattern matches, contrary to popular expectations:
c07a80fd 2089
4358a253
SS
2090 $x = 'ABC123';
2091 $y = 'ABC445';
c07a80fd 2092
4358a253
SS
2093 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2094 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2095
4358a253
SS
2096 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2097 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2098
2099This prints
2100
2101 2: got ABC
2102 3: got AB
2103 4: got ABC
2104
5f05dabc 2105You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2106general purpose version of test 1. The important difference between
2107them is that test 3 contains a quantifier (C<\D*>) and so can use
2108backtracking, whereas test 1 will not. What's happening is
2109that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2110non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2111let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2112fail.
14218588 2113
c07a80fd 2114The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2115try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2116a quantifier (C<\D*>) has been used in the regular expression, the
2117search engine can backtrack and retry the match differently
54310121 2118in the hope of matching the complete regular expression.
c07a80fd 2119
5a964f20
TC
2120The pattern really, I<really> wants to succeed, so it uses the
2121standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2122time. Now there's indeed something following "AB" that is not
14218588 2123"123". It's "C123", which suffices.
c07a80fd 2124
14218588
GS
2125We can deal with this by using both an assertion and a negation.
2126We'll say that the first part in $1 must be followed both by a digit
2127and by something that's not "123". Remember that the look-aheads
2128are zero-width expressions--they only look, but don't consume any
2129of the string in their match. So rewriting this way produces what
c07a80fd
PP
2130you'd expect; that is, case 5 will fail, but case 6 succeeds:
2131
4358a253
SS
2132 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2133 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2134
2135 6: got ABC
2136
5a964f20 2137In other words, the two zero-width assertions next to each other work as though
19799a22 2138they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2139matches only if you're at the beginning of the line AND the end of the
2140line simultaneously. The deeper underlying truth is that juxtaposition in
2141regular expressions always means AND, except when you write an explicit OR
2142using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2143although the attempted matches are made at different positions because "a"
2144is not a zero-width assertion, but a one-width assertion.
2145
0d017f4d 2146B<WARNING>: Particularly complicated regular expressions can take
14218588 2147exponential time to solve because of the immense number of possible
0d017f4d 2148ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2149internal optimizations done by the regular expression engine, this will
2150take a painfully long time to run:
c07a80fd 2151
e1901655
IZ
2152 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2153
2154And if you used C<*>'s in the internal groups instead of limiting them
2155to 0 through 5 matches, then it would take forever--or until you ran
2156out of stack space. Moreover, these internal optimizations are not
2157always applicable. For example, if you put C<{0,5}> instead of C<*>
2158on the external group, no current optimization is applicable, and the
2159match takes a long time to finish.
c07a80fd 2160
9da458fc
IZ
2161A powerful tool for optimizing such beasts is what is known as an
2162"independent group",
96090e4f 2163which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2164zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2165the tail match, since they are in "logical" context: only
14218588 2166whether they match is considered relevant. For an example
9da458fc 2167where side-effects of look-ahead I<might> have influenced the
96090e4f 2168following match, see L</C<< (?>pattern) >>>.
c277df42 2169
a0d0e21e 2170=head2 Version 8 Regular Expressions
d74e8afc 2171X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2172
5a964f20 2173In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2174routines, here are the pattern-matching rules not described above.
2175
54310121 2176Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2177with a special meaning described here or above. You can cause
5a964f20 2178characters that normally function as metacharacters to be interpreted
5f05dabc 2179literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2180character; "\\" matches a "\"). This escape mechanism is also required
2181for the character used as the pattern delimiter.
2182
2183A series of characters matches that series of characters in the target
0b928c2f 2184string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2185string.
a0d0e21e
LW
2186
2187You can specify a character class, by enclosing a list of characters
5d458dd8 2188in C<[]>, which will match any character from the list. If the
a0d0e21e 2189first character after the "[" is "^", the class matches any character not
14218588 2190in the list. Within a list, the "-" character specifies a
5a964f20 2191range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2192inclusive. If you want either "-" or "]" itself to be a member of a
2193class, put it at the start of the list (possibly after a "^"), or
2194escape it with a backslash. "-" is also taken literally when it is
2195at the end of the list, just before the closing "]". (The
84850974
DD
2196following all specify the same class of three characters: C<[-az]>,
2197C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2198specifies a class containing twenty-six characters, even on EBCDIC-based
2199character sets.) Also, if you try to use the character
2200classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2201a range, the "-" is understood literally.
a0d0e21e 2202
8ada0baa
JH
2203Note also that the whole range idea is rather unportable between
2204character sets--and even within character sets they may cause results
2205you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2206that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2207[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2208spell out the character sets in full.
2209
54310121 2210Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2211used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2212"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2213of three octal digits, matches the character whose coded character set value
5d458dd8 2214is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2215matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2216matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2217matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2218
2219You can specify a series of alternatives for a pattern using "|" to
2220separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2221or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2222first alternative includes everything from the last pattern delimiter
0b928c2f 2223("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2224the last alternative contains everything from the last "|" to the next
0b928c2f 2225closing pattern delimiter. That's why it's common practice to include
14218588 2226alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2227start and end.
2228
5a964f20 2229Alternatives are tried from left to right, so the first
a3cb178b
GS
2230alternative found for which the entire expression matches, is the one that
2231is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2232example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2233part will match, as that is the first alternative tried, and it successfully
2234matches the target string. (This might not seem important, but it is
2235important when you are capturing matched text using parentheses.)
2236
5a964f20 2237Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2238so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2239
14218588
GS
2240Within a pattern, you may designate subpatterns for later reference
2241by enclosing them in parentheses, and you may refer back to the
2242I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2243\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2244of their opening parenthesis. A backreference matches whatever
2245actually matched the subpattern in the string being examined, not
d8b950dc 2246the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2247match "0x1234 0x4321", but not "0x1234 01234", because subpattern
22481 matched "0x", even though the rule C<0|0x> could potentially match
2249the leading 0 in the second number.
cb1a09d0 2250
0d017f4d 2251=head2 Warning on \1 Instead of $1
cb1a09d0 2252
5a964f20 2253Some people get too used to writing things like:
cb1a09d0
AD
2254
2255 $pattern =~ s/(\W)/\\\1/g;
2256
3ff1c45a
KW
2257This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2258shocking the
cb1a09d0 2259B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2260PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2261the usual double-quoted string means a control-A. The customary Unix
2262meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2263of doing that, you get yourself into trouble if you then add an C</e>
2264modifier.
2265
f793d64a 2266 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2267
2268Or if you try to do
2269
2270 s/(\d+)/\1000/;
2271
2272You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2273C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2274with the operation of matching a backreference. Certainly they mean two
2275different things on the I<left> side of the C<s///>.
9fa51da4 2276
0d017f4d 2277=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2278
19799a22 2279B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2280
2281Regular expressions provide a terse and powerful programming language. As
2282with most other power tools, power comes together with the ability
2283to wreak havoc.
2284
2285A common abuse of this power stems from the ability to make infinite
628afcb5 2286loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2287
2288 'foo' =~ m{ ( o? )* }x;
2289
0d017f4d 2290The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2291in the string is not moved by the match, C<o?> would match again and again
527e91da 2292because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2293is with the looping modifier C<//g>:
2294
2295 @matches = ( 'foo' =~ m{ o? }xg );
2296
2297or
2298
2299 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2300
2301or the loop implied by split().
2302
2303However, long experience has shown that many programming tasks may
14218588
GS
2304be significantly simplified by using repeated subexpressions that
2305may match zero-length substrings. Here's a simple example being:
c84d73f1 2306
d1fbf752 2307 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2308 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2309
9da458fc 2310Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2311the infinite loop>. The rules for this are different for lower-level
527e91da 2312loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2313ones like the C</g> modifier or split() operator.
2314
19799a22
GS
2315The lower-level loops are I<interrupted> (that is, the loop is
2316broken) when Perl detects that a repeated expression matched a
2317zero-length substring. Thus
c84d73f1
IZ
2318
2319 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2320
5d458dd8 2321is made equivalent to
c84d73f1 2322
0b928c2f
FC
2323 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2324
2325For example, this program
2326
2327 #!perl -l
2328 "aaaaab" =~ /
2329 (?:
2330 a # non-zero
2331 | # or
2332 (?{print "hello"}) # print hello whenever this
2333 # branch is tried
2334 (?=(b)) # zero-width assertion
2335 )* # any number of times
2336 /x;
2337 print $&;
2338 print $1;
c84d73f1 2339
0b928c2f
FC
2340prints
2341
2342 hello
2343 aaaaa
2344 b
2345
2346Notice that "hello" is only printed once, as when Perl sees that the sixth
2347iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2348the C<*>.
2349
2350The higher-level loops preserve an additional state between iterations:
5d458dd8 2351whether the last match was zero-length. To break the loop, the following
c84d73f1 2352match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2353This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2354and so the I<second best> match is chosen if the I<best> match is of
2355zero length.
2356
19799a22 2357For example:
c84d73f1
IZ
2358
2359 $_ = 'bar';
2360 s/\w??/<$&>/g;
2361
20fb949f 2362results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2363match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2364best> match is what is matched by C<\w>. Thus zero-length matches
2365alternate with one-character-long matches.
2366
5d458dd8 2367Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2368position one notch further in the string.
2369
19799a22 2370The additional state of being I<matched with zero-length> is associated with
c84d73f1 2371the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2372Zero-length matches at the end of the previous match are ignored
2373during C<split>.
c84d73f1 2374
0d017f4d 2375=head2 Combining RE Pieces
35a734be
IZ
2376
2377Each of the elementary pieces of regular expressions which were described
2378before (such as C<ab> or C<\Z>) could match at most one substring
2379at the given position of the input string. However, in a typical regular
2380expression these elementary pieces are combined into more complicated
0b928c2f 2381patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2382(in these examples C<S> and C<T> are regular subexpressions).
2383
2384Such combinations can include alternatives, leading to a problem of choice:
2385if we match a regular expression C<a|ab> against C<"abc">, will it match
2386substring C<"a"> or C<"ab">? One way to describe which substring is
2387actually matched is the concept of backtracking (see L<"Backtracking">).
2388However, this description is too low-level and makes you think
2389in terms of a particular implementation.
2390
2391Another description starts with notions of "better"/"worse". All the
2392substrings which may be matched by the given regular expression can be
2393sorted from the "best" match to the "worst" match, and it is the "best"
2394match which is chosen. This substitutes the question of "what is chosen?"
2395by the question of "which matches are better, and which are worse?".
2396
2397Again, for elementary pieces there is no such question, since at most
2398one match at a given position is possible. This section describes the
2399notion of better/worse for combining operators. In the description
2400below C<S> and C<T> are regular subexpressions.
2401
13a2d996 2402=over 4
35a734be
IZ
2403
2404=item C<ST>
2405
2406Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2407substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2408which can be matched by C<T>.
35a734be 2409
0b928c2f 2410If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2411match than C<A'B'>.
2412
2413If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2414C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2415
2416=item C<S|T>
2417
2418When C<S> can match, it is a better match than when only C<T> can match.
2419
2420Ordering of two matches for C<S> is the same as for C<S>. Similar for
2421two matches for C<T>.
2422
2423=item C<S{REPEAT_COUNT}>
2424
2425Matches as C<SSS...S> (repeated as many times as necessary).
2426
2427=item C<S{min,max}>
2428
2429Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2430
2431=item C<S{min,max}?>
2432
2433Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2434
2435=item C<S?>, C<S*>, C<S+>
2436
2437Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2438
2439=item C<S??>, C<S*?>, C<S+?>
2440
2441Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2442
c47ff5f1 2443=item C<< (?>S) >>
35a734be
IZ
2444
2445Matches the best match for C<S> and only that.
2446
2447=item C<(?=S)>, C<(?<=S)>
2448
2449Only the best match for C<S> is considered. (This is important only if
2450C<S> has capturing parentheses, and backreferences are used somewhere
2451else in the whole regular expression.)
2452
2453=item C<(?!S)>, C<(?<!S)>
2454
2455For this grouping operator there is no need to describe the ordering, since
2456only whether or not C<S> can match is important.
2457
6bda09f9 2458=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2459
2460The ordering is the same as for the regular expression which is
c27a5cfe 2461the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
2462
2463=item C<(?(condition)yes-pattern|no-pattern)>
2464
2465Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2466already determined. The ordering of the matches is the same as for the
2467chosen subexpression.
2468
2469=back
2470
2471The above recipes describe the ordering of matches I<at a given position>.
2472One more rule is needed to understand how a match is determined for the
2473whole regular expression: a match at an earlier position is always better
2474than a match at a later position.
2475
0d017f4d 2476=head2 Creating Custom RE Engines
c84d73f1 2477
0b928c2f
FC
2478As of Perl 5.10.0, one can create custom regular expression engines. This
2479is not for the faint of heart, as they have to plug in at the C level. See
2480L<perlreapi> for more details.
2481
2482As an alternative, overloaded constants (see L<overload>) provide a simple
2483way to extend the functionality of the RE engine, by substituting one
2484pattern for another.
c84d73f1
IZ
2485
2486Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2487matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2488characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2489at these positions, so we want to have each C<\Y|> in the place of the
2490more complicated version. We can create a module C<customre> to do
2491this:
2492
2493 package customre;
2494 use overload;
2495
2496 sub import {
2497 shift;
2498 die "No argument to customre::import allowed" if @_;
2499 overload::constant 'qr' => \&convert;
2500 }
2501
2502 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2503
580a9fe1
RGS
2504 # We must also take care of not escaping the legitimate \\Y|
2505 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2506 my %rules = ( '\\' => '\\\\',
f793d64a 2507 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2508 sub convert {
2509 my $re = shift;
5d458dd8 2510 $re =~ s{
c84d73f1
IZ
2511 \\ ( \\ | Y . )
2512 }
5d458dd8 2513 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2514 return $re;
2515 }
2516
2517Now C<use customre> enables the new escape in constant regular
2518expressions, i.e., those without any runtime variable interpolations.
2519As documented in L<overload>, this conversion will work only over
2520literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2521part of this regular expression needs to be converted explicitly
2522(but only if the special meaning of C<\Y|> should be enabled inside $re):
2523
2524 use customre;
2525 $re = <>;
2526 chomp $re;
2527 $re = customre::convert $re;
2528 /\Y|$re\Y|/;
2529
0b928c2f 2530=head2 PCRE/Python Support
1f1031fe 2531
0b928c2f 2532As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2533to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2534Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2535
2536=over 4
2537
ae5648b3 2538=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2539
c27a5cfe 2540Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2541
2542=item C<< (?P=NAME) >>
2543
c27a5cfe 2544Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2545
2546=item C<< (?P>NAME) >>
2547
c27a5cfe 2548Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2549
ee9b8eae 2550=back
1f1031fe 2551
19799a22
GS
2552=head1 BUGS
2553
88c9975e
KW
2554Many regular expression constructs don't work on EBCDIC platforms.
2555
ed7efc79
KW
2556There are a number of issues with regard to case-insensitive matching
2557in Unicode rules. See C<i> under L</Modifiers> above.
2558
9da458fc
IZ
2559This document varies from difficult to understand to completely
2560and utterly opaque. The wandering prose riddled with jargon is
2561hard to fathom in several places.
2562
2563This document needs a rewrite that separates the tutorial content
2564from the reference content.
19799a22
GS
2565
2566=head1 SEE ALSO
9fa51da4 2567
91e0c79e
MJD
2568L<perlrequick>.
2569
2570L<perlretut>.
2571
9b599b2a
GS
2572L<perlop/"Regexp Quote-Like Operators">.
2573
1e66bd83
PP
2574L<perlop/"Gory details of parsing quoted constructs">.
2575
14218588
GS
2576L<perlfaq6>.
2577
9b599b2a
GS
2578L<perlfunc/pos>.
2579
2580L<perllocale>.
2581
fb55449c
JH
2582L<perlebcdic>.
2583
14218588
GS
2584I<Mastering Regular Expressions> by Jeffrey Friedl, published
2585by O'Reilly and Associates.