This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
typo fix for re pod change use of optimise to be consistent with other uses of optimize
[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
9d53c457
KW
75Perl doesn't match multiple characters in a bracketed
76character class unless the character that maps to them is explicitly
77mentioned, and it doesn't match them at all if the character class is
78inverted, which otherwise could be highly confusing. See
79L<perlrecharclass/Bracketed Character Classes>, and
1f59b283
KW
80L<perlrecharclass/Negation>.
81
54310121 82=item x
d74e8afc 83X</x>
55497cff
PP
84
85Extend your pattern's legibility by permitting whitespace and comments.
ed7efc79 86Details in L</"/x">
55497cff 87
87e95b7f
YO
88=item p
89X</p> X<regex, preserve> X<regexp, preserve>
90
632a1772 91Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
87e95b7f
YO
92${^POSTMATCH} are available for use after matching.
93
e2e6bec7
DN
94=item g and c
95X</g> X</c>
96
97Global matching, and keep the Current position after failed matching.
98Unlike i, m, s and x, these two flags affect the way the regex is used
99rather than the regex itself. See
100L<perlretut/"Using regular expressions in Perl"> for further explanation
101of the g and c modifiers.
102
b6fa137b
FC
103=item a, d, l and u
104X</a> X</d> X</l> X</u>
105
516074bb
KW
106These modifiers, all new in 5.14, affect which character-set semantics
107(Unicode, etc.) are used, as described below in
ed7efc79 108L</Character set modifiers>.
b6fa137b 109
55497cff 110=back
a0d0e21e 111
516074bb
KW
112Regular expression modifiers are usually written in documentation
113as e.g., "the C</x> modifier", even though the delimiter
b6fa137b 114in question might not really be a slash. The modifiers C</imsxadlup>
ab7bb42d 115may also be embedded within the regular expression itself using
ed7efc79 116the C<(?...)> construct, see L</Extended Patterns> below.
b6fa137b 117
ed7efc79
KW
118=head3 /x
119
b6fa137b 120C</x> tells
7b059540 121the regular expression parser to ignore most whitespace that is neither
55497cff 122backslashed nor within a character class. You can use this to break up
4633a7c4 123your regular expression into (slightly) more readable parts. The C<#>
54310121 124character is also treated as a metacharacter introducing a comment,
55497cff 125just as in ordinary Perl code. This also means that if you want real
14218588 126whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a 127class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
128escape them (using backslashes or C<\Q...\E>) or encode them using octal,
129hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
130making Perl's regular expressions more readable. Note that you have to
131be careful not to include the pattern delimiter in the comment--perl has
132no way of knowing you did not intend to close the pattern early. See
133the C-comment deletion code in L<perlop>. Also note that anything inside
7651b971 134a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
0b928c2f 135space interpretation within a single multi-character construct. For
7651b971 136example in C<\x{...}>, regardless of the C</x> modifier, there can be no
9bb1f947 137spaces. Same for a L<quantifier|/Quantifiers> such as C<{3}> or
000947ad
KW
138C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<(>,
139C<?>, and C<:>. Within any delimiters for such a
f9e949fd
KW
140construct, allowed spaces are not affected by C</x>, and depend on the
141construct. For example, C<\x{...}> can't have spaces because hexadecimal
142numbers don't have spaces in them. But, Unicode properties can have spaces, so
0b928c2f 143in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
9bb1f947 144L<perluniprops/Properties accessible through \p{} and \P{}>.
d74e8afc 145X</x>
a0d0e21e 146
ed7efc79
KW
147=head3 Character set modifiers
148
149C</d>, C</u>, C</a>, and C</l>, available starting in 5.14, are called
150the character set modifiers; they affect the character set semantics
151used for the regular expression.
152
808432af
KW
153The C</d>, C</u>, and C</l> modifiers are not likely to be of much use
154to you, and so you need not worry about them very much. They exist for
155Perl's internal use, so that complex regular expression data structures
156can be automatically serialized and later exactly reconstituted,
157including all their nuances. But, since Perl can't keep a secret, and
158there may be rare instances where they are useful, they are documented
159here.
ed7efc79 160
808432af
KW
161The C</a> modifier, on the other hand, may be useful. Its purpose is to
162allow code that is to work mostly on ASCII data to not have to concern
163itself with Unicode.
ca9560b2 164
808432af
KW
165Briefly, C</l> sets the character set to that of whatever B<L>ocale is in
166effect at the time of the execution of the pattern match.
ca9560b2 167
808432af
KW
168C</u> sets the character set to B<U>nicode.
169
170C</a> also sets the character set to Unicode, BUT adds several
171restrictions for B<A>SCII-safe matching.
172
173C</d> is the old, problematic, pre-5.14 B<D>efault character set
174behavior. Its only use is to force that old behavior.
175
176At any given time, exactly one of these modifiers is in effect. Their
177existence allows Perl to keep the originally compiled behavior of a
178regular expression, regardless of what rules are in effect when it is
179actually executed. And if it is interpolated into a larger regex, the
180original's rules continue to apply to it, and only it.
181
182The C</l> and C</u> modifiers are automatically selected for
183regular expressions compiled within the scope of various pragmas,
184and we recommend that in general, you use those pragmas instead of
185specifying these modifiers explicitly. For one thing, the modifiers
186affect only pattern matching, and do not extend to even any replacement
187done, whereas using the pragmas give consistent results for all
188appropriate operations within their scopes. For example,
189
190 s/foo/\Ubar/il
191
192will match "foo" using the locale's rules for case-insensitive matching,
193but the C</l> does not affect how the C<\U> operates. Most likely you
194want both of them to use locale rules. To do this, instead compile the
195regular expression within the scope of C<use locale>. This both
196implicitly adds the C</l> and applies locale rules to the C<\U>. The
197lesson is to C<use locale> and not C</l> explicitly.
198
199Similarly, it would be better to use C<use feature 'unicode_strings'>
200instead of,
201
202 s/foo/\Lbar/iu
203
204to get Unicode rules, as the C<\L> in the former (but not necessarily
205the latter) would also use Unicode rules.
206
207More detail on each of the modifiers follows. Most likely you don't
208need to know this detail for C</l>, C</u>, and C</d>, and can skip ahead
209to L<E<sol>a|/E<sol>a (and E<sol>aa)>.
ca9560b2 210
ed7efc79
KW
211=head4 /l
212
213means to use the current locale's rules (see L<perllocale>) when pattern
214matching. For example, C<\w> will match the "word" characters of that
215locale, and C<"/i"> case-insensitive matching will match according to
216the locale's case folding rules. The locale used will be the one in
217effect at the time of execution of the pattern match. This may not be
218the same as the compilation-time locale, and can differ from one match
219to another if there is an intervening call of the
b6fa137b 220L<setlocale() function|perllocale/The setlocale function>.
ed7efc79
KW
221
222Perl only supports single-byte locales. This means that code points
223above 255 are treated as Unicode no matter what locale is in effect.
224Under Unicode rules, there are a few case-insensitive matches that cross
225the 255/256 boundary. These are disallowed under C</l>. For example,
516074bb
KW
2260xFF (on ASCII platforms) does not caselessly match the character at
2270x178, C<LATIN CAPITAL LETTER Y WITH DIAERESIS>, because 0xFF may not be
228C<LATIN SMALL LETTER Y WITH DIAERESIS> in the current locale, and Perl
229has no way of knowing if that character even exists in the locale, much
230less what code point it is.
ed7efc79
KW
231
232This modifier may be specified to be the default by C<use locale>, but
233see L</Which character set modifier is in effect?>.
b6fa137b
FC
234X</l>
235
ed7efc79
KW
236=head4 /u
237
238means to use Unicode rules when pattern matching. On ASCII platforms,
239this means that the code points between 128 and 255 take on their
808432af
KW
240Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's).
241(Otherwise Perl considers their meanings to be undefined.) Thus,
242under this modifier, the ASCII platform effectively becomes a Unicode
243platform; and hence, for example, C<\w> will match any of the more than
244100_000 word characters in Unicode.
ed7efc79
KW
245
246Unlike most locales, which are specific to a language and country pair,
516074bb
KW
247Unicode classifies all the characters that are letters I<somewhere> in
248the world as
ed7efc79
KW
249C<\w>. For example, your locale might not think that C<LATIN SMALL
250LETTER ETH> is a letter (unless you happen to speak Icelandic), but
251Unicode does. Similarly, all the characters that are decimal digits
252somewhere in the world will match C<\d>; this is hundreds, not 10,
253possible matches. And some of those digits look like some of the 10
254ASCII digits, but mean a different number, so a human could easily think
255a number is a different quantity than it really is. For example,
256C<BENGALI DIGIT FOUR> (U+09EA) looks very much like an
257C<ASCII DIGIT EIGHT> (U+0038). And, C<\d+>, may match strings of digits
258that are a mixture from different writing systems, creating a security
67592e11 259issue. L<Unicode::UCD/num()> can be used to sort
516074bb
KW
260this out. Or the C</a> modifier can be used to force C<\d> to match
261just the ASCII 0 through 9.
ed7efc79 262
516074bb
KW
263Also, under this modifier, case-insensitive matching works on the full
264set of Unicode
ed7efc79
KW
265characters. The C<KELVIN SIGN>, for example matches the letters "k" and
266"K"; and C<LATIN SMALL LIGATURE FF> matches the sequence "ff", which,
267if you're not prepared, might make it look like a hexadecimal constant,
268presenting another potential security issue. See
269L<http://unicode.org/reports/tr36> for a detailed discussion of Unicode
270security issues.
271
ed7efc79 272This modifier may be specified to be the default by C<use feature
66cbab2c
KW
273'unicode_strings>, C<use locale ':not_characters'>, or
274C<L<use 5.012|perlfunc/use VERSION>> (or higher),
808432af 275but see L</Which character set modifier is in effect?>.
b6fa137b
FC
276X</u>
277
ed7efc79
KW
278=head4 /d
279
280This modifier means to use the "Default" native rules of the platform
281except when there is cause to use Unicode rules instead, as follows:
282
283=over 4
284
285=item 1
286
287the target string is encoded in UTF-8; or
288
289=item 2
290
291the pattern is encoded in UTF-8; or
292
293=item 3
294
295the pattern explicitly mentions a code point that is above 255 (say by
296C<\x{100}>); or
297
298=item 4
b6fa137b 299
ed7efc79
KW
300the pattern uses a Unicode name (C<\N{...}>); or
301
302=item 5
303
9d1a5160
KW
304the pattern uses a Unicode property (C<\p{...}>); or
305
306=item 6
307
308the pattern uses L</C<(?[ ])>>
ed7efc79
KW
309
310=back
311
312Another mnemonic for this modifier is "Depends", as the rules actually
313used depend on various things, and as a result you can get unexpected
808432af
KW
314results. See L<perlunicode/The "Unicode Bug">. The Unicode Bug has
315become rather infamous, leading to yet another (printable) name for this
316modifier, "Dodgy".
ed7efc79 317
4b9734bf
KW
318Unless the pattern or string are encoded in UTF-8, only ASCII characters
319can match positively.
ed7efc79
KW
320
321Here are some examples of how that works on an ASCII platform:
322
323 $str = "\xDF"; # $str is not in UTF-8 format.
324 $str =~ /^\w/; # No match, as $str isn't in UTF-8 format.
325 $str .= "\x{0e0b}"; # Now $str is in UTF-8 format.
326 $str =~ /^\w/; # Match! $str is now in UTF-8 format.
327 chop $str;
328 $str =~ /^\w/; # Still a match! $str remains in UTF-8 format.
329
808432af
KW
330This modifier is automatically selected by default when none of the
331others are, so yet another name for it is "Default".
332
333Because of the unexpected behaviors associated with this modifier, you
334probably should only use it to maintain weird backward compatibilities.
335
336=head4 /a (and /aa)
337
338This modifier stands for ASCII-restrict (or ASCII-safe). This modifier,
339unlike the others, may be doubled-up to increase its effect.
340
341When it appears singly, it causes the sequences C<\d>, C<\s>, C<\w>, and
342the Posix character classes to match only in the ASCII range. They thus
343revert to their pre-5.6, pre-Unicode meanings. Under C</a>, C<\d>
344always means precisely the digits C<"0"> to C<"9">; C<\s> means the five
d28d8023
KW
345characters C<[ \f\n\r\t]>, and starting in Perl v5.18, experimentally,
346the vertical tab; C<\w> means the 63 characters
808432af
KW
347C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
348C<[[:print:]]> match only the appropriate ASCII-range characters.
349
350This modifier is useful for people who only incidentally use Unicode,
351and who do not wish to be burdened with its complexities and security
352concerns.
353
354With C</a>, one can write C<\d> with confidence that it will only match
355ASCII characters, and should the need arise to match beyond ASCII, you
356can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>). There are
357similar C<\p{...}> constructs that can match beyond ASCII both white
358space (see L<perlrecharclass/Whitespace>), and Posix classes (see
359L<perlrecharclass/POSIX Character Classes>). Thus, this modifier
360doesn't mean you can't use Unicode, it means that to get Unicode
361matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
362signals Unicode.
363
364As you would expect, this modifier causes, for example, C<\D> to mean
365the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
366C<\D>, C<\S>, and C<\W>. C<\b> still means to match at the boundary
367between C<\w> and C<\W>, using the C</a> definitions of them (similarly
368for C<\B>).
369
370Otherwise, C</a> behaves like the C</u> modifier, in that
371case-insensitive matching uses Unicode semantics; for example, "k" will
372match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
373points in the Latin1 range, above ASCII will have Unicode rules when it
374comes to case-insensitive matching.
375
376To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
377specify the "a" twice, for example C</aai> or C</aia>. (The first
378occurrence of "a" restricts the C<\d>, etc., and the second occurrence
379adds the C</i> restrictions.) But, note that code points outside the
380ASCII range will use Unicode rules for C</i> matching, so the modifier
381doesn't really restrict things to just ASCII; it just forbids the
382intermixing of ASCII and non-ASCII.
383
384To summarize, this modifier provides protection for applications that
385don't wish to be exposed to all of Unicode. Specifying it twice
386gives added protection.
387
388This modifier may be specified to be the default by C<use re '/a'>
389or C<use re '/aa'>. If you do so, you may actually have occasion to use
31dc26d6 390the C</u> modifier explicitly if there are a few regular expressions
808432af
KW
391where you do want full Unicode rules (but even here, it's best if
392everything were under feature C<"unicode_strings">, along with the
393C<use re '/aa'>). Also see L</Which character set modifier is in
394effect?>.
395X</a>
396X</aa>
397
ed7efc79
KW
398=head4 Which character set modifier is in effect?
399
400Which of these modifiers is in effect at any given point in a regular
808432af
KW
401expression depends on a fairly complex set of interactions. These have
402been designed so that in general you don't have to worry about it, but
403this section gives the gory details. As
ed7efc79
KW
404explained below in L</Extended Patterns> it is possible to explicitly
405specify modifiers that apply only to portions of a regular expression.
406The innermost always has priority over any outer ones, and one applying
6368643f
KW
407to the whole expression has priority over any of the default settings that are
408described in the remainder of this section.
ed7efc79 409
916cec3f 410The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
ed7efc79
KW
411default modifiers (including these) for regular expressions compiled
412within its scope. This pragma has precedence over the other pragmas
516074bb 413listed below that also change the defaults.
ed7efc79
KW
414
415Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
66cbab2c 416and C<L<use feature 'unicode_strings|feature>>, or
ed7efc79
KW
417C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
418C</u> when not in the same scope as either C<L<use locale|perllocale>>
66cbab2c
KW
419or C<L<use bytes|bytes>>.
420(C<L<use locale ':not_characters'|perllocale/Unicode and UTF-8>> also
421sets the default to C</u>, overriding any plain C<use locale>.)
422Unlike the mechanisms mentioned above, these
6368643f
KW
423affect operations besides regular expressions pattern matching, and so
424give more consistent results with other operators, including using
425C<\U>, C<\l>, etc. in substitution replacements.
ed7efc79
KW
426
427If none of the above apply, for backwards compatibility reasons, the
428C</d> modifier is the one in effect by default. As this can lead to
429unexpected results, it is best to specify which other rule set should be
430used.
431
432=head4 Character set modifier behavior prior to Perl 5.14
433
434Prior to 5.14, there were no explicit modifiers, but C</l> was implied
435for regexes compiled within the scope of C<use locale>, and C</d> was
436implied otherwise. However, interpolating a regex into a larger regex
437would ignore the original compilation in favor of whatever was in effect
438at the time of the second compilation. There were a number of
439inconsistencies (bugs) with the C</d> modifier, where Unicode rules
440would be used when inappropriate, and vice versa. C<\p{}> did not imply
441Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
b6fa137b 442
a0d0e21e
LW
443=head2 Regular Expressions
444
04838cea
RGS
445=head3 Metacharacters
446
384f06ae 447The patterns used in Perl pattern matching evolved from those supplied in
14218588 448the Version 8 regex routines. (The routines are derived
19799a22
GS
449(distantly) from Henry Spencer's freely redistributable reimplementation
450of the V8 routines.) See L<Version 8 Regular Expressions> for
451details.
a0d0e21e
LW
452
453In particular the following metacharacters have their standard I<egrep>-ish
454meanings:
d74e8afc
ITB
455X<metacharacter>
456X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
457
a0d0e21e 458
f793d64a
KW
459 \ Quote the next metacharacter
460 ^ Match the beginning of the line
461 . Match any character (except newline)
462 $ Match the end of the line (or before newline at the end)
463 | Alternation
464 () Grouping
465 [] Bracketed Character class
a0d0e21e 466
14218588
GS
467By default, the "^" character is guaranteed to match only the
468beginning of the string, the "$" character only the end (or before the
469newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
470assumption that the string contains only one line. Embedded newlines
471will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 472string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
473newline within the string (except if the newline is the last character in
474the string), and "$" will match before any newline. At the
a0d0e21e
LW
475cost of a little more overhead, you can do this by using the /m modifier
476on the pattern match operator. (Older programs did this by setting C<$*>,
db7cd43a 477but this option was removed in perl 5.10.)
d74e8afc 478X<^> X<$> X</m>
a0d0e21e 479
14218588 480To simplify multi-line substitutions, the "." character never matches a
55497cff 481newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 482the string is a single line--even if it isn't.
d74e8afc 483X<.> X</s>
a0d0e21e 484
04838cea
RGS
485=head3 Quantifiers
486
a0d0e21e 487The following standard quantifiers are recognized:
d74e8afc 488X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 489
f793d64a
KW
490 * Match 0 or more times
491 + Match 1 or more times
492 ? Match 1 or 0 times
493 {n} Match exactly n times
494 {n,} Match at least n times
495 {n,m} Match at least n but not more than m times
a0d0e21e 496
0b928c2f 497(If a curly bracket occurs in any other context and does not form part of
4d68ffa0
KW
498a backslashed sequence like C<\x{...}>, it is treated as a regular
499character. In particular, the lower quantifier bound is not optional,
500and a typo in a quantifier silently causes it to be treated as the
501literal characters. For example,
9af81bfe
KW
502
503 /o{4,3}/
504
505looks like a quantifier that matches 0 times, since 4 is greater than 3,
506but it really means to match the sequence of six characters
4d68ffa0
KW
507S<C<"o { 4 , 3 }">>. It is planned to eventually require literal uses
508of curly brackets to be escaped, say by preceding them with a backslash
509or enclosing them within square brackets, (C<"\{"> or C<"[{]">). This
510change will allow for future syntax extensions (like making the lower
511bound of a quantifier optional), and better error checking. In the
512meantime, you should get in the habit of escaping all instances where
513you mean a literal "{".)
9af81bfe
KW
514
515The "*" quantifier is equivalent to C<{0,}>, the "+"
527e91da 516quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 517to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
518This is usually 32766 on the most common platforms. The actual limit can
519be seen in the error message generated by code such as this:
520
820475bd 521 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 522
54310121
PP
523By default, a quantified subpattern is "greedy", that is, it will match as
524many times as possible (given a particular starting location) while still
525allowing the rest of the pattern to match. If you want it to match the
526minimum number of times possible, follow the quantifier with a "?". Note
527that the meanings don't change, just the "greediness":
0d017f4d 528X<metacharacter> X<greedy> X<greediness>
d74e8afc 529X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 530
f793d64a
KW
531 *? Match 0 or more times, not greedily
532 +? Match 1 or more times, not greedily
533 ?? Match 0 or 1 time, not greedily
0b928c2f 534 {n}? Match exactly n times, not greedily (redundant)
f793d64a
KW
535 {n,}? Match at least n times, not greedily
536 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 537
b9b4dddf
YO
538By default, when a quantified subpattern does not allow the rest of the
539overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 540sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
541as well.
542
f793d64a
KW
543 *+ Match 0 or more times and give nothing back
544 ++ Match 1 or more times and give nothing back
545 ?+ Match 0 or 1 time and give nothing back
546 {n}+ Match exactly n times and give nothing back (redundant)
547 {n,}+ Match at least n times and give nothing back
548 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
549
550For instance,
551
552 'aaaa' =~ /a++a/
553
554will never match, as the C<a++> will gobble up all the C<a>'s in the
555string and won't leave any for the remaining part of the pattern. This
556feature can be extremely useful to give perl hints about where it
557shouldn't backtrack. For instance, the typical "match a double-quoted
558string" problem can be most efficiently performed when written as:
559
560 /"(?:[^"\\]++|\\.)*+"/
561
0d017f4d 562as we know that if the final quote does not match, backtracking will not
0b928c2f
FC
563help. See the independent subexpression
564L</C<< (?>pattern) >>> for more details;
b9b4dddf
YO
565possessive quantifiers are just syntactic sugar for that construct. For
566instance the above example could also be written as follows:
567
568 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
569
04838cea
RGS
570=head3 Escape sequences
571
0b928c2f 572Because patterns are processed as double-quoted strings, the following
a0d0e21e
LW
573also work:
574
f793d64a
KW
575 \t tab (HT, TAB)
576 \n newline (LF, NL)
577 \r return (CR)
578 \f form feed (FF)
579 \a alarm (bell) (BEL)
580 \e escape (think troff) (ESC)
f793d64a 581 \cK control char (example: VT)
dc0d9c48 582 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 583 \N{name} named Unicode character or character sequence
f793d64a 584 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 585 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
586 \l lowercase next char (think vi)
587 \u uppercase next char (think vi)
588 \L lowercase till \E (think vi)
589 \U uppercase till \E (think vi)
590 \Q quote (disable) pattern metacharacters till \E
591 \E end either case modification or quoted section, think vi
a0d0e21e 592
9bb1f947 593Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 594
e1d1eefb 595=head3 Character Classes and other Special Escapes
04838cea 596
a0d0e21e 597In addition, Perl defines the following:
d0b16107 598X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 599
f793d64a
KW
600 Sequence Note Description
601 [...] [1] Match a character according to the rules of the
602 bracketed character class defined by the "...".
603 Example: [a-z] matches "a" or "b" or "c" ... or "z"
604 [[:...:]] [2] Match a character according to the rules of the POSIX
605 character class "..." within the outer bracketed
606 character class. Example: [[:upper:]] matches any
607 uppercase character.
572224ce 608 (?[...]) [8] Extended bracketed character class
d35dd6c6
KW
609 \w [3] Match a "word" character (alphanumeric plus "_", plus
610 other connector punctuation chars plus Unicode
0b928c2f 611 marks)
f793d64a
KW
612 \W [3] Match a non-"word" character
613 \s [3] Match a whitespace character
614 \S [3] Match a non-whitespace character
615 \d [3] Match a decimal digit character
616 \D [3] Match a non-digit character
617 \pP [3] Match P, named property. Use \p{Prop} for longer names
618 \PP [3] Match non-P
619 \X [4] Match Unicode "eXtended grapheme cluster"
620 \C Match a single C-language char (octet) even if that is
621 part of a larger UTF-8 character. Thus it breaks up
622 characters into their UTF-8 bytes, so you may end up
623 with malformed pieces of UTF-8. Unsupported in
624 lookbehind.
c27a5cfe 625 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
626 '1' may actually be any positive integer.
627 \g1 [5] Backreference to a specific or previous group,
628 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 629 previous group and may optionally be wrapped in
f793d64a
KW
630 curly brackets for safer parsing.
631 \g{name} [5] Named backreference
632 \k<name> [5] Named backreference
633 \K [6] Keep the stuff left of the \K, don't include it in $&
2171640d 634 \N [7] Any character but \n. Not affected by /s modifier
f793d64a
KW
635 \v [3] Vertical whitespace
636 \V [3] Not vertical whitespace
637 \h [3] Horizontal whitespace
638 \H [3] Not horizontal whitespace
639 \R [4] Linebreak
e1d1eefb 640
9bb1f947
KW
641=over 4
642
643=item [1]
644
645See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 646
9bb1f947 647=item [2]
b8c5462f 648
9bb1f947 649See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 650
9bb1f947 651=item [3]
5496314a 652
9bb1f947 653See L<perlrecharclass/Backslash sequences> for details.
5496314a 654
9bb1f947 655=item [4]
5496314a 656
9bb1f947 657See L<perlrebackslash/Misc> for details.
d0b16107 658
9bb1f947 659=item [5]
b8c5462f 660
c27a5cfe 661See L</Capture groups> below for details.
93733859 662
9bb1f947 663=item [6]
b8c5462f 664
9bb1f947
KW
665See L</Extended Patterns> below for details.
666
667=item [7]
668
669Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
670character or character sequence whose name is C<NAME>; and similarly
671when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
672code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947 673
572224ce
KW
674=item [8]
675
676See L<perlrecharclass/Extended Bracketed Character Classes> for details.
677
9bb1f947 678=back
d0b16107 679
04838cea
RGS
680=head3 Assertions
681
a0d0e21e 682Perl defines the following zero-width assertions:
d74e8afc
ITB
683X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
684X<regexp, zero-width assertion>
685X<regular expression, zero-width assertion>
686X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 687
9bb1f947
KW
688 \b Match a word boundary
689 \B Match except at a word boundary
690 \A Match only at beginning of string
691 \Z Match only at end of string, or before newline at the end
692 \z Match only at end of string
693 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 694 of prior m//g)
a0d0e21e 695
14218588 696A word boundary (C<\b>) is a spot between two characters
19799a22
GS
697that has a C<\w> on one side of it and a C<\W> on the other side
698of it (in either order), counting the imaginary characters off the
699beginning and end of the string as matching a C<\W>. (Within
700character classes C<\b> represents backspace rather than a word
701boundary, just as it normally does in any double-quoted string.)
702The C<\A> and C<\Z> are just like "^" and "$", except that they
703won't match multiple times when the C</m> modifier is used, while
704"^" and "$" will match at every internal line boundary. To match
705the actual end of the string and not ignore an optional trailing
706newline, use C<\z>.
d74e8afc 707X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
708
709The C<\G> assertion can be used to chain global matches (using
710C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
711It is also useful when writing C<lex>-like scanners, when you have
712several patterns that you want to match against consequent substrings
0b928c2f 713of your string; see the previous reference. The actual location
19799a22 714where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d 715an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
0b928c2f
FC
716matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
717is modified somewhat, in that contents to the left of C<\G> are
58e23c8d
YO
718not counted when determining the length of the match. Thus the following
719will not match forever:
d74e8afc 720X<\G>
c47ff5f1 721
e761bb84
CO
722 my $string = 'ABC';
723 pos($string) = 1;
724 while ($string =~ /(.\G)/g) {
725 print $1;
726 }
58e23c8d
YO
727
728It will print 'A' and then terminate, as it considers the match to
729be zero-width, and thus will not match at the same position twice in a
730row.
731
732It is worth noting that C<\G> improperly used can result in an infinite
733loop. Take care when using patterns that include C<\G> in an alternation.
734
c27a5cfe 735=head3 Capture groups
04838cea 736
c27a5cfe
KW
737The bracketing construct C<( ... )> creates capture groups (also referred to as
738capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
739the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
740for the second, and so on.
741This is called a I<backreference>.
d74e8afc 742X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 743X<regex, capture group> X<regexp, capture group>
d74e8afc 744X<regular expression, capture buffer> X<backreference>
c27a5cfe 745X<regular expression, capture group> X<backreference>
1f1031fe 746X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
747X<named capture buffer> X<regular expression, named capture buffer>
748X<named capture group> X<regular expression, named capture group>
749X<%+> X<$+{name}> X<< \k<name> >>
750There is no limit to the number of captured substrings that you may use.
751Groups are numbered with the leftmost open parenthesis being number 1, etc. If
752a group did not match, the associated backreference won't match either. (This
753can happen if the group is optional, or in a different branch of an
754alternation.)
755You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
756this form, described below.
757
758You can also refer to capture groups relatively, by using a negative number, so
759that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
760group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
761example:
5624f11d
YO
762
763 /
c27a5cfe
KW
764 (Y) # group 1
765 ( # group 2
766 (X) # group 3
767 \g{-1} # backref to group 3
768 \g{-3} # backref to group 1
5624f11d
YO
769 )
770 /x
771
d8b950dc
KW
772would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
773interpolate regexes into larger regexes and not have to worry about the
774capture groups being renumbered.
775
776You can dispense with numbers altogether and create named capture groups.
777The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
778reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
779also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
780I<name> must not begin with a number, nor contain hyphens.
781When different groups within the same pattern have the same name, any reference
782to that name assumes the leftmost defined group. Named groups count in
783absolute and relative numbering, and so can also be referred to by those
784numbers.
785(It's possible to do things with named capture groups that would otherwise
786require C<(??{})>.)
787
788Capture group contents are dynamically scoped and available to you outside the
789pattern until the end of the enclosing block or until the next successful
790match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
791You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
792etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
793
794Braces are required in referring to named capture groups, but are optional for
795absolute or relative numbered ones. Braces are safer when creating a regex by
796concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
797contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
798is probably not what you intended.
799
800The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
801there were no named nor relative numbered capture groups. Absolute numbered
0b928c2f
FC
802groups were referred to using C<\1>,
803C<\2>, etc., and this notation is still
d8b950dc
KW
804accepted (and likely always will be). But it leads to some ambiguities if
805there are more than 9 capture groups, as C<\10> could mean either the tenth
806capture group, or the character whose ordinal in octal is 010 (a backspace in
807ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
808only if at least 10 left parentheses have opened before it. Likewise C<\11> is
809a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
810And so on. C<\1> through C<\9> are always interpreted as backreferences.
811There are several examples below that illustrate these perils. You can avoid
812the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
813and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
814digits padded with leading zeros, since a leading zero implies an octal
815constant.
d8b950dc
KW
816
817The C<\I<digit>> notation also works in certain circumstances outside
ed7efc79 818the pattern. See L</Warning on \1 Instead of $1> below for details.
81714fb9 819
14218588 820Examples:
a0d0e21e
LW
821
822 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
823
d8b950dc 824 /(.)\g1/ # find first doubled char
81714fb9
YO
825 and print "'$1' is the first doubled character\n";
826
827 /(?<char>.)\k<char>/ # ... a different way
828 and print "'$+{char}' is the first doubled character\n";
829
d8b950dc 830 /(?'char'.)\g1/ # ... mix and match
81714fb9 831 and print "'$1' is the first doubled character\n";
c47ff5f1 832
14218588 833 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
834 $hours = $1;
835 $minutes = $2;
836 $seconds = $3;
a0d0e21e 837 }
c47ff5f1 838
9d860678
KW
839 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
840 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
841 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
842 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
843
844 $a = '(.)\1'; # Creates problems when concatenated.
845 $b = '(.)\g{1}'; # Avoids the problems.
846 "aa" =~ /${a}/; # True
847 "aa" =~ /${b}/; # True
848 "aa0" =~ /${a}0/; # False!
849 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
850 "aa\x08" =~ /${a}0/; # True!
851 "aa\x08" =~ /${b}0/; # False
9d860678 852
14218588
GS
853Several special variables also refer back to portions of the previous
854match. C<$+> returns whatever the last bracket match matched.
855C<$&> returns the entire matched string. (At one point C<$0> did
856also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
857everything before the matched string. C<$'> returns everything
858after the matched string. And C<$^N> contains whatever was matched by
859the most-recently closed group (submatch). C<$^N> can be used in
860extended patterns (see below), for example to assign a submatch to a
81714fb9 861variable.
d74e8afc 862X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 863
d8b950dc
KW
864These special variables, like the C<%+> hash and the numbered match variables
865(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
866until the end of the enclosing block or until the next successful
867match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
868X<$+> X<$^N> X<$&> X<$`> X<$'>
869X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
870
0d017f4d 871B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 872which makes it easier to write code that tests for a series of more
665e98b9
JH
873specific cases and remembers the best match.
874
e1fd4132 875B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
14218588 876C<$'> anywhere in the program, it has to provide them for every
e1fd4132 877pattern match. This may substantially slow your program. Perl
d8b950dc 878uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
879price for each pattern that contains capturing parentheses. (To
880avoid this cost while retaining the grouping behaviour, use the
881extended regular expression C<(?: ... )> instead.) But if you never
882use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
883parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
884if you can, but if you can't (and some algorithms really appreciate
885them), once you've used them once, use them at will, because you've
e1fd4132
DM
886already paid the price. As of 5.17.4, the presence of each of the three
887variables in a program is recorded separately, and depending on
888circumstances, perl may be able be more efficient knowing that only C<$&>
889rather than all three have been seen, for example.
d74e8afc 890X<$&> X<$`> X<$'>
68dc0745 891
e1fd4132 892As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
893C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
894and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 895successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
896The use of these variables incurs no global performance penalty, unlike
897their punctuation char equivalents, however at the trade-off that you
e1fd4132 898have to tell perl when you want to use them.
87e95b7f 899X</p> X<p modifier>
cde0cee5 900
9d727203
KW
901=head2 Quoting metacharacters
902
19799a22
GS
903Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
904C<\w>, C<\n>. Unlike some other regular expression languages, there
905are no backslashed symbols that aren't alphanumeric. So anything
0f264506 906that looks like \\, \(, \), \[, \], \{, or \} is always
19799a22
GS
907interpreted as a literal character, not a metacharacter. This was
908once used in a common idiom to disable or quote the special meanings
909of regular expression metacharacters in a string that you want to
36bbe248 910use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
911
912 $pattern =~ s/(\W)/\\$1/g;
913
f1cbbd6e 914(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
915Today it is more common to use the quotemeta() function or the C<\Q>
916metaquoting escape sequence to disable all metacharacters' special
917meanings like this:
a0d0e21e
LW
918
919 /$unquoted\Q$quoted\E$unquoted/
920
9da458fc
IZ
921Beware that if you put literal backslashes (those not inside
922interpolated variables) between C<\Q> and C<\E>, double-quotish
923backslash interpolation may lead to confusing results. If you
924I<need> to use literal backslashes within C<\Q...\E>,
925consult L<perlop/"Gory details of parsing quoted constructs">.
926
736fe711
KW
927C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
928
19799a22
GS
929=head2 Extended Patterns
930
14218588 931Perl also defines a consistent extension syntax for features not
0b928c2f
FC
932found in standard tools like B<awk> and
933B<lex>. The syntax for most of these is a
14218588
GS
934pair of parentheses with a question mark as the first thing within
935the parentheses. The character after the question mark indicates
936the extension.
19799a22 937
14218588
GS
938The stability of these extensions varies widely. Some have been
939part of the core language for many years. Others are experimental
940and may change without warning or be completely removed. Check
941the documentation on an individual feature to verify its current
942status.
19799a22 943
14218588
GS
944A question mark was chosen for this and for the minimal-matching
945construct because 1) question marks are rare in older regular
946expressions, and 2) whenever you see one, you should stop and
0b928c2f 947"question" exactly what is going on. That's psychology....
a0d0e21e 948
70ca8714 949=over 4
a0d0e21e 950
cc6b7395 951=item C<(?#text)>
d74e8afc 952X<(?#)>
a0d0e21e 953
14218588 954A comment. The text is ignored. If the C</x> modifier enables
19799a22 955whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
956the comment as soon as it sees a C<)>, so there is no way to put a literal
957C<)> in the comment.
a0d0e21e 958
cfaf538b 959=item C<(?adlupimsx-imsx)>
fb85c044 960
cfaf538b 961=item C<(?^alupimsx)>
fb85c044 962X<(?)> X<(?^)>
19799a22 963
0b6d1084
JH
964One or more embedded pattern-match modifiers, to be turned on (or
965turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
966the remainder of the enclosing pattern group (if any).
967
fb85c044 968This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 969configuration file, taken from an argument, or specified in a table
0b928c2f
FC
970somewhere. Consider the case where some patterns want to be
971case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 972include C<(?i)> at the front of the pattern. For example:
19799a22
GS
973
974 $pattern = "foobar";
5d458dd8 975 if ( /$pattern/i ) { }
19799a22
GS
976
977 # more flexible:
978
979 $pattern = "(?i)foobar";
5d458dd8 980 if ( /$pattern/ ) { }
19799a22 981
0b6d1084 982These modifiers are restored at the end of the enclosing group. For example,
19799a22 983
d8b950dc 984 ( (?i) blah ) \s+ \g1
19799a22 985
0d017f4d
WL
986will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
987repetition of the previous word, assuming the C</x> modifier, and no C</i>
988modifier outside this group.
19799a22 989
8eb5594e 990These modifiers do not carry over into named subpatterns called in the
dd72e27b 991enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
8eb5594e
DR
992change the case-sensitivity of the "NAME" pattern.
993
dc925305
KW
994Any of these modifiers can be set to apply globally to all regular
995expressions compiled within the scope of a C<use re>. See
a0bbd6ff 996L<re/"'/flags' mode">.
dc925305 997
9de15fec
KW
998Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
999after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
1000C<"d">) may follow the caret to override it.
1001But a minus sign is not legal with it.
1002
dc925305 1003Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 1004that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 1005C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
1006others, and a maximum of one (or two C<a>'s) may appear in the
1007construct. Thus, for
0b928c2f 1008example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 1009C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1010
1011Note also that the C<p> modifier is special in that its presence
1012anywhere in a pattern has a global effect.
cde0cee5 1013
5a964f20 1014=item C<(?:pattern)>
d74e8afc 1015X<(?:)>
a0d0e21e 1016
cfaf538b 1017=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1018
cfaf538b 1019=item C<(?^aluimsx:pattern)>
fb85c044
KW
1020X<(?^:)>
1021
5a964f20
TC
1022This is for clustering, not capturing; it groups subexpressions like
1023"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1024
5a964f20 1025 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1026
1027is like
1028
5a964f20 1029 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1030
19799a22
GS
1031but doesn't spit out extra fields. It's also cheaper not to capture
1032characters if you don't need to.
a0d0e21e 1033
19799a22 1034Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1035C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1036
1037 /(?s-i:more.*than).*million/i
1038
14218588 1039is equivalent to the more verbose
ca9dfc88
IZ
1040
1041 /(?:(?s-i)more.*than).*million/i
1042
fb85c044 1043Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1044after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1045flags (except C<"d">) may follow the caret, so
fb85c044
KW
1046
1047 (?^x:foo)
1048
1049is equivalent to
1050
1051 (?x-ims:foo)
1052
1053The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1054surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1055modified by any flags specified.
1056
1057The caret allows for simpler stringification of compiled regular
1058expressions. These look like
1059
1060 (?^:pattern)
1061
1062with any non-default flags appearing between the caret and the colon.
1063A test that looks at such stringification thus doesn't need to have the
1064system default flags hard-coded in it, just the caret. If new flags are
1065added to Perl, the meaning of the caret's expansion will change to include
1066the default for those flags, so the test will still work, unchanged.
1067
1068Specifying a negative flag after the caret is an error, as the flag is
1069redundant.
1070
1071Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1072to match at the beginning.
1073
594d7033
YO
1074=item C<(?|pattern)>
1075X<(?|)> X<Branch reset>
1076
1077This is the "branch reset" pattern, which has the special property
c27a5cfe 1078that the capture groups are numbered from the same starting point
99d59c4d 1079in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1080
c27a5cfe 1081Capture groups are numbered from left to right, but inside this
693596a8 1082construct the numbering is restarted for each branch.
4deaaa80 1083
c27a5cfe 1084The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1085following this construct will be numbered as though the construct
1086contained only one branch, that being the one with the most capture
c27a5cfe 1087groups in it.
4deaaa80 1088
0b928c2f 1089This construct is useful when you want to capture one of a
4deaaa80
PJ
1090number of alternative matches.
1091
1092Consider the following pattern. The numbers underneath show in
c27a5cfe 1093which group the captured content will be stored.
594d7033
YO
1094
1095
1096 # before ---------------branch-reset----------- after
1097 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1098 # 1 2 2 3 2 3 4
1099
ab106183
A
1100Be careful when using the branch reset pattern in combination with
1101named captures. Named captures are implemented as being aliases to
c27a5cfe 1102numbered groups holding the captures, and that interferes with the
ab106183
A
1103implementation of the branch reset pattern. If you are using named
1104captures in a branch reset pattern, it's best to use the same names,
1105in the same order, in each of the alternations:
1106
1107 /(?| (?<a> x ) (?<b> y )
1108 | (?<a> z ) (?<b> w )) /x
1109
1110Not doing so may lead to surprises:
1111
1112 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1113 say $+ {a}; # Prints '12'
1114 say $+ {b}; # *Also* prints '12'.
1115
c27a5cfe
KW
1116The problem here is that both the group named C<< a >> and the group
1117named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1118
ee9b8eae
YO
1119=item Look-Around Assertions
1120X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1121
0b928c2f 1122Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1123pattern without including it in C<$&>. Positive assertions match when
1124their subpattern matches, negative assertions match when their subpattern
1125fails. Look-behind matches text up to the current match position,
1126look-ahead matches text following the current match position.
1127
1128=over 4
1129
5a964f20 1130=item C<(?=pattern)>
d74e8afc 1131X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1132
19799a22 1133A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1134matches a word followed by a tab, without including the tab in C<$&>.
1135
5a964f20 1136=item C<(?!pattern)>
d74e8afc 1137X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1138
19799a22 1139A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1140matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1141however that look-ahead and look-behind are NOT the same thing. You cannot
1142use this for look-behind.
7b8d334a 1143
5a964f20 1144If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1145will not do what you want. That's because the C<(?!foo)> is just saying that
1146the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1147match. Use look-behind instead (see below).
c277df42 1148
ee9b8eae
YO
1149=item C<(?<=pattern)> C<\K>
1150X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1151
c47ff5f1 1152A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1153matches a word that follows a tab, without including the tab in C<$&>.
1154Works only for fixed-width look-behind.
c277df42 1155
ee9b8eae
YO
1156There is a special form of this construct, called C<\K>, which causes the
1157regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1158not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1159look-behind. The use of C<\K> inside of another look-around assertion
1160is allowed, but the behaviour is currently not well defined.
1161
c62285ac 1162For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1163equivalent C<< (?<=...) >> construct, and it is especially useful in
1164situations where you want to efficiently remove something following
1165something else in a string. For instance
1166
1167 s/(foo)bar/$1/g;
1168
1169can be rewritten as the much more efficient
1170
1171 s/foo\Kbar//g;
1172
5a964f20 1173=item C<(?<!pattern)>
d74e8afc 1174X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1175
19799a22
GS
1176A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1177matches any occurrence of "foo" that does not follow "bar". Works
1178only for fixed-width look-behind.
c277df42 1179
ee9b8eae
YO
1180=back
1181
81714fb9
YO
1182=item C<(?'NAME'pattern)>
1183
1184=item C<< (?<NAME>pattern) >>
1185X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1186
c27a5cfe 1187A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1188parentheses C<()> but for the additional fact that the group
1189can be referred to by name in various regular expression
1190constructs (like C<\g{NAME}>) and can be accessed by name
1191after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1192for more details on the C<%+> and C<%-> hashes.
81714fb9 1193
c27a5cfe
KW
1194If multiple distinct capture groups have the same name then the
1195$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1196
0d017f4d 1197The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1198
1199B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1200function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1201numbered sequentially regardless of being named or not. Thus in the
1202pattern
1203
1204 /(x)(?<foo>y)(z)/
1205
1206$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1207the opposite which is what a .NET regex hacker might expect.
1208
1f1031fe
YO
1209Currently NAME is restricted to simple identifiers only.
1210In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1211its Unicode extension (see L<utf8>),
1212though it isn't extended by the locale (see L<perllocale>).
81714fb9 1213
1f1031fe 1214B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1215with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1216may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1217support the use of single quotes as a delimiter for the name.
81714fb9 1218
1f1031fe
YO
1219=item C<< \k<NAME> >>
1220
1221=item C<< \k'NAME' >>
81714fb9
YO
1222
1223Named backreference. Similar to numeric backreferences, except that
1224the group is designated by name and not number. If multiple groups
1225have the same name then it refers to the leftmost defined group in
1226the current match.
1227
0d017f4d 1228It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1229earlier in the pattern.
1230
1231Both forms are equivalent.
1232
1f1031fe 1233B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1234with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1235may be used instead of C<< \k<NAME> >>.
1f1031fe 1236
cc6b7395 1237=item C<(?{ code })>
d74e8afc 1238X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1239
19799a22 1240B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1241experimental, and may be changed without notice. Code executed that
1242has side effects may not perform identically from version to version
e128ab2c
DM
1243due to the effect of future optimisations in the regex engine. The
1244implementation of this feature was radically overhauled for the 5.18.0
1245release, and its behaviour in earlier versions of perl was much buggier,
1246especially in relation to parsing, lexical vars, scoping, recursion and
1247reentrancy.
c277df42 1248
e128ab2c
DM
1249This zero-width assertion executes any embedded Perl code. It always
1250succeeds, and its return value is set as C<$^R>.
19799a22 1251
e128ab2c
DM
1252In literal patterns, the code is parsed at the same time as the
1253surrounding code. While within the pattern, control is passed temporarily
1254back to the perl parser, until the logically-balancing closing brace is
1255encountered. This is similar to the way that an array index expression in
1256a literal string is handled, for example
77ea4f6d 1257
e128ab2c
DM
1258 "abc$array[ 1 + f('[') + g()]def"
1259
1260In particular, braces do not need to be balanced:
1261
576fa024 1262 s/abc(?{ f('{'); })/def/
e128ab2c
DM
1263
1264Even in a pattern that is interpolated and compiled at run-time, literal
1265code blocks will be compiled once, at perl compile time; the following
1266prints "ABCD":
1267
1268 print "D";
1269 my $qr = qr/(?{ BEGIN { print "A" } })/;
1270 my $foo = "foo";
1271 /$foo$qr(?{ BEGIN { print "B" } })/;
1272 BEGIN { print "C" }
1273
1274In patterns where the text of the code is derived from run-time
1275information rather than appearing literally in a source code /pattern/,
1276the code is compiled at the same time that the pattern is compiled, and
5771dda0 1277for reasons of security, C<use re 'eval'> must be in scope. This is to
e128ab2c
DM
1278stop user-supplied patterns containing code snippets from being
1279executable.
1280
5771dda0 1281In situations where you need to enable this with C<use re 'eval'>, you should
e128ab2c
DM
1282also have taint checking enabled. Better yet, use the carefully
1283constrained evaluation within a Safe compartment. See L<perlsec> for
1284details about both these mechanisms.
1285
1286From the viewpoint of parsing, lexical variable scope and closures,
1287
1288 /AAA(?{ BBB })CCC/
1289
1290behaves approximately like
1291
1292 /AAA/ && do { BBB } && /CCC/
1293
1294Similarly,
1295
1296 qr/AAA(?{ BBB })CCC/
1297
1298behaves approximately like
77ea4f6d 1299
e128ab2c
DM
1300 sub { /AAA/ && do { BBB } && /CCC/ }
1301
1302In particular:
1303
1304 { my $i = 1; $r = qr/(?{ print $i })/ }
1305 my $i = 2;
1306 /$r/; # prints "1"
1307
1308Inside a C<(?{...})> block, C<$_> refers to the string the regular
754091cb 1309expression is matching against. You can also use C<pos()> to know what is
fa11829f 1310the current position of matching within this string.
754091cb 1311
e128ab2c
DM
1312The code block introduces a new scope from the perspective of lexical
1313variable declarations, but B<not> from the perspective of C<local> and
1314similar localizing behaviours. So later code blocks within the same
1315pattern will still see the values which were localized in earlier blocks.
1316These accumulated localizations are undone either at the end of a
1317successful match, or if the assertion is backtracked (compare
1318L<"Backtracking">). For example,
b9ac3b5b
GS
1319
1320 $_ = 'a' x 8;
5d458dd8 1321 m<
d1fbf752 1322 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1323 (
5d458dd8 1324 a
b9ac3b5b 1325 (?{
d1fbf752
KW
1326 local $cnt = $cnt + 1; # Update $cnt,
1327 # backtracking-safe.
b9ac3b5b 1328 })
5d458dd8 1329 )*
b9ac3b5b 1330 aaaa
d1fbf752
KW
1331 (?{ $res = $cnt }) # On success copy to
1332 # non-localized location.
b9ac3b5b
GS
1333 >x;
1334
e128ab2c
DM
1335will initially increment C<$cnt> up to 8; then during backtracking, its
1336value will be unwound back to 4, which is the value assigned to C<$res>.
1337At the end of the regex execution, $cnt will be wound back to its initial
1338value of 0.
1339
1340This assertion may be used as the condition in a
b9ac3b5b 1341
e128ab2c
DM
1342 (?(condition)yes-pattern|no-pattern)
1343
1344switch. If I<not> used in this way, the result of evaluation of C<code>
1345is put into the special variable C<$^R>. This happens immediately, so
1346C<$^R> can be used from other C<(?{ code })> assertions inside the same
1347regular expression.
b9ac3b5b 1348
19799a22
GS
1349The assignment to C<$^R> above is properly localized, so the old
1350value of C<$^R> is restored if the assertion is backtracked; compare
1351L<"Backtracking">.
b9ac3b5b 1352
e128ab2c
DM
1353Note that the special variable C<$^N> is particularly useful with code
1354blocks to capture the results of submatches in variables without having to
1355keep track of the number of nested parentheses. For example:
1356
1357 $_ = "The brown fox jumps over the lazy dog";
1358 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1359 print "color = $color, animal = $animal\n";
1360
8988a1bb 1361
14455d6c 1362=item C<(??{ code })>
d74e8afc
ITB
1363X<(??{})>
1364X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1365
19799a22 1366B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1367experimental, and may be changed without notice. Code executed that
1368has side effects may not perform identically from version to version
1369due to the effect of future optimisations in the regex engine.
0f5d15d6 1370
e128ab2c
DM
1371This is a "postponed" regular subexpression. It behaves in I<exactly> the
1372same way as a C<(?{ code })> code block as described above, except that
1373its return value, rather than being assigned to C<$^R>, is treated as a
1374pattern, compiled if it's a string (or used as-is if its a qr// object),
1375then matched as if it were inserted instead of this construct.
6bda09f9 1376
e128ab2c
DM
1377During the matching of this sub-pattern, it has its own set of
1378captures which are valid during the sub-match, but are discarded once
1379control returns to the main pattern. For example, the following matches,
1380with the inner pattern capturing "B" and matching "BB", while the outer
1381pattern captures "A";
1382
1383 my $inner = '(.)\1';
1384 "ABBA" =~ /^(.)(??{ $inner })\1/;
1385 print $1; # prints "A";
6bda09f9 1386
e128ab2c
DM
1387Note that this means that there is no way for the inner pattern to refer
1388to a capture group defined outside. (The code block itself can use C<$1>,
1389etc., to refer to the enclosing pattern's capture groups.) Thus, although
0f5d15d6 1390
e128ab2c
DM
1391 ('a' x 100)=~/(??{'(.)' x 100})/
1392
1393I<will> match, it will I<not> set $1 on exit.
19799a22
GS
1394
1395The following pattern matches a parenthesized group:
0f5d15d6 1396
d1fbf752
KW
1397 $re = qr{
1398 \(
1399 (?:
1400 (?> [^()]+ ) # Non-parens without backtracking
1401 |
1402 (??{ $re }) # Group with matching parens
1403 )*
1404 \)
1405 }x;
0f5d15d6 1406
93f313ef
KW
1407See also
1408L<C<(?I<PARNO>)>|/(?PARNO) (?-PARNO) (?+PARNO) (?R) (?0)>
1409for a different, more efficient way to accomplish
6bda09f9
YO
1410the same task.
1411
e128ab2c
DM
1412Executing a postponed regular expression 50 times without consuming any
1413input string will result in a fatal error. The maximum depth is compiled
1414into perl, so changing it requires a custom build.
6bda09f9 1415
93f313ef 1416=item C<(?I<PARNO>)> C<(?-I<PARNO>)> C<(?+I<PARNO>)> C<(?R)> C<(?0)>
542fa716 1417X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1418X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1419X<regex, relative recursion>
6bda09f9 1420
e128ab2c
DM
1421Similar to C<(??{ code })> except that it does not involve executing any
1422code or potentially compiling a returned pattern string; instead it treats
1423the part of the current pattern contained within a specified capture group
1424as an independent pattern that must match at the current position.
1425Capture groups contained by the pattern will have the value as determined
1426by the outermost recursion.
6bda09f9 1427
93f313ef 1428I<PARNO> is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1429the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1430the beginning of the whole pattern. C<(?0)> is an alternate syntax for
93f313ef 1431C<(?R)>. If I<PARNO> is preceded by a plus or minus sign then it is assumed
c27a5cfe 1432to be relative, with negative numbers indicating preceding capture groups
542fa716 1433and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1434declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1435Note that the counting for relative recursion differs from that of
c27a5cfe 1436relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1437included.
6bda09f9 1438
81714fb9 1439The following pattern matches a function foo() which may contain
f145b7e9 1440balanced parentheses as the argument.
6bda09f9 1441
d1fbf752 1442 $re = qr{ ( # paren group 1 (full function)
81714fb9 1443 foo
d1fbf752 1444 ( # paren group 2 (parens)
6bda09f9 1445 \(
d1fbf752 1446 ( # paren group 3 (contents of parens)
6bda09f9 1447 (?:
d1fbf752 1448 (?> [^()]+ ) # Non-parens without backtracking
6bda09f9 1449 |
d1fbf752 1450 (?2) # Recurse to start of paren group 2
6bda09f9
YO
1451 )*
1452 )
1453 \)
1454 )
1455 )
1456 }x;
1457
1458If the pattern was used as follows
1459
1460 'foo(bar(baz)+baz(bop))'=~/$re/
1461 and print "\$1 = $1\n",
1462 "\$2 = $2\n",
1463 "\$3 = $3\n";
1464
1465the output produced should be the following:
1466
1467 $1 = foo(bar(baz)+baz(bop))
1468 $2 = (bar(baz)+baz(bop))
81714fb9 1469 $3 = bar(baz)+baz(bop)
6bda09f9 1470
c27a5cfe 1471If there is no corresponding capture group defined, then it is a
61528107 1472fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1473string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1474into perl, so changing it requires a custom build.
1475
542fa716
YO
1476The following shows how using negative indexing can make it
1477easier to embed recursive patterns inside of a C<qr//> construct
1478for later use:
1479
1480 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
c77257ed 1481 if (/foo $parens \s+ \+ \s+ bar $parens/x) {
542fa716
YO
1482 # do something here...
1483 }
1484
81714fb9 1485B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1486PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1487a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1488as atomic. Also, modifiers are resolved at compile time, so constructs
1489like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1490be processed.
6bda09f9 1491
894be9b7
YO
1492=item C<(?&NAME)>
1493X<(?&NAME)>
1494
93f313ef 1495Recurse to a named subpattern. Identical to C<(?I<PARNO>)> except that the
0d017f4d 1496parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1497the same name, then it recurses to the leftmost.
1498
1499It is an error to refer to a name that is not declared somewhere in the
1500pattern.
1501
1f1031fe
YO
1502B<NOTE:> In order to make things easier for programmers with experience
1503with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1504may be used instead of C<< (?&NAME) >>.
1f1031fe 1505
e2e6a0f1
YO
1506=item C<(?(condition)yes-pattern|no-pattern)>
1507X<(?()>
286f584a 1508
e2e6a0f1 1509=item C<(?(condition)yes-pattern)>
286f584a 1510
41ef34de
ML
1511Conditional expression. Matches C<yes-pattern> if C<condition> yields
1512a true value, matches C<no-pattern> otherwise. A missing pattern always
1513matches.
1514
25e26d77 1515C<(condition)> should be one of: 1) an integer in
e2e6a0f1 1516parentheses (which is valid if the corresponding pair of parentheses
25e26d77 1517matched); 2) a look-ahead/look-behind/evaluate zero-width assertion; 3) a
c27a5cfe 1518name in angle brackets or single quotes (which is valid if a group
25e26d77 1519with the given name matched); or 4) the special symbol (R) (true when
e2e6a0f1
YO
1520evaluated inside of recursion or eval). Additionally the R may be
1521followed by a number, (which will be true when evaluated when recursing
1522inside of the appropriate group), or by C<&NAME>, in which case it will
1523be true only when evaluated during recursion in the named group.
1524
1525Here's a summary of the possible predicates:
1526
1527=over 4
1528
1529=item (1) (2) ...
1530
c27a5cfe 1531Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1532
1533=item (<NAME>) ('NAME')
1534
c27a5cfe 1535Checks if a group with the given name has matched something.
e2e6a0f1 1536
f01cd190
FC
1537=item (?=...) (?!...) (?<=...) (?<!...)
1538
1539Checks whether the pattern matches (or does not match, for the '!'
1540variants).
1541
e2e6a0f1
YO
1542=item (?{ CODE })
1543
f01cd190 1544Treats the return value of the code block as the condition.
e2e6a0f1
YO
1545
1546=item (R)
1547
1548Checks if the expression has been evaluated inside of recursion.
1549
1550=item (R1) (R2) ...
1551
1552Checks if the expression has been evaluated while executing directly
1553inside of the n-th capture group. This check is the regex equivalent of
1554
1555 if ((caller(0))[3] eq 'subname') { ... }
1556
1557In other words, it does not check the full recursion stack.
1558
1559=item (R&NAME)
1560
1561Similar to C<(R1)>, this predicate checks to see if we're executing
1562directly inside of the leftmost group with a given name (this is the same
1563logic used by C<(?&NAME)> to disambiguate). It does not check the full
1564stack, but only the name of the innermost active recursion.
1565
1566=item (DEFINE)
1567
1568In this case, the yes-pattern is never directly executed, and no
1569no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1570See below for details.
1571
1572=back
1573
1574For example:
1575
1576 m{ ( \( )?
1577 [^()]+
1578 (?(1) \) )
1579 }x
1580
1581matches a chunk of non-parentheses, possibly included in parentheses
1582themselves.
1583
0b928c2f
FC
1584A special form is the C<(DEFINE)> predicate, which never executes its
1585yes-pattern directly, and does not allow a no-pattern. This allows one to
1586define subpatterns which will be executed only by the recursion mechanism.
e2e6a0f1
YO
1587This way, you can define a set of regular expression rules that can be
1588bundled into any pattern you choose.
1589
1590It is recommended that for this usage you put the DEFINE block at the
1591end of the pattern, and that you name any subpatterns defined within it.
1592
1593Also, it's worth noting that patterns defined this way probably will
31dc26d6 1594not be as efficient, as the optimizer is not very clever about
e2e6a0f1
YO
1595handling them.
1596
1597An example of how this might be used is as follows:
1598
2bf803e2 1599 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1600 (?(DEFINE)
2bf803e2
YO
1601 (?<NAME_PAT>....)
1602 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1603 )/x
1604
c27a5cfe
KW
1605Note that capture groups matched inside of recursion are not accessible
1606after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1607necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1608C<$+{NAME}> would be.
286f584a 1609
51a1303c
BF
1610Finally, keep in mind that subpatterns created inside a DEFINE block
1611count towards the absolute and relative number of captures, so this:
1612
1613 my @captures = "a" =~ /(.) # First capture
1614 (?(DEFINE)
1615 (?<EXAMPLE> 1 ) # Second capture
1616 )/x;
1617 say scalar @captures;
1618
1619Will output 2, not 1. This is particularly important if you intend to
1620compile the definitions with the C<qr//> operator, and later
1621interpolate them in another pattern.
1622
c47ff5f1 1623=item C<< (?>pattern) >>
6bda09f9 1624X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1625
19799a22
GS
1626An "independent" subexpression, one which matches the substring
1627that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1628position, and it matches I<nothing other than this substring>. This
19799a22
GS
1629construct is useful for optimizations of what would otherwise be
1630"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1631It may also be useful in places where the "grab all you can, and do not
1632give anything back" semantic is desirable.
19799a22 1633
c47ff5f1 1634For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1635(anchored at the beginning of string, as above) will match I<all>
1636characters C<a> at the beginning of string, leaving no C<a> for
1637C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1638since the match of the subgroup C<a*> is influenced by the following
1639group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1640C<a*ab> will match fewer characters than a standalone C<a*>, since
1641this makes the tail match.
1642
0b928c2f
FC
1643C<< (?>pattern) >> does not disable backtracking altogether once it has
1644matched. It is still possible to backtrack past the construct, but not
1645into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1646
c47ff5f1 1647An effect similar to C<< (?>pattern) >> may be achieved by writing
0b928c2f
FC
1648C<(?=(pattern))\g{-1}>. This matches the same substring as a standalone
1649C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
c47ff5f1 1650makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1651(The difference between these two constructs is that the second one
1652uses a capturing group, thus shifting ordinals of backreferences
1653in the rest of a regular expression.)
1654
1655Consider this pattern:
c277df42 1656
871b0233 1657 m{ \(
e2e6a0f1 1658 (
f793d64a 1659 [^()]+ # x+
e2e6a0f1 1660 |
871b0233
IZ
1661 \( [^()]* \)
1662 )+
e2e6a0f1 1663 \)
871b0233 1664 }x
5a964f20 1665
19799a22
GS
1666That will efficiently match a nonempty group with matching parentheses
1667two levels deep or less. However, if there is no such group, it
1668will take virtually forever on a long string. That's because there
1669are so many different ways to split a long string into several
1670substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1671to a subpattern of the above pattern. Consider how the pattern
1672above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1673seconds, but that each extra letter doubles this time. This
1674exponential performance will make it appear that your program has
14218588 1675hung. However, a tiny change to this pattern
5a964f20 1676
e2e6a0f1
YO
1677 m{ \(
1678 (
f793d64a 1679 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1680 |
871b0233
IZ
1681 \( [^()]* \)
1682 )+
e2e6a0f1 1683 \)
871b0233 1684 }x
c277df42 1685
c47ff5f1 1686which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1687this yourself would be a productive exercise), but finishes in a fourth
1688the time when used on a similar string with 1000000 C<a>s. Be aware,
0b928c2f
FC
1689however, that, when this construct is followed by a
1690quantifier, it currently triggers a warning message under
9f1b1f2d 1691the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1692C<"matches null string many times in regex">.
c277df42 1693
c47ff5f1 1694On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1695effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1696This was only 4 times slower on a string with 1000000 C<a>s.
1697
9da458fc
IZ
1698The "grab all you can, and do not give anything back" semantic is desirable
1699in many situations where on the first sight a simple C<()*> looks like
1700the correct solution. Suppose we parse text with comments being delimited
1701by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1702its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1703the comment delimiter, because it may "give up" some whitespace if
1704the remainder of the pattern can be made to match that way. The correct
1705answer is either one of these:
1706
1707 (?>#[ \t]*)
1708 #[ \t]*(?![ \t])
1709
1710For example, to grab non-empty comments into $1, one should use either
1711one of these:
1712
1713 / (?> \# [ \t]* ) ( .+ ) /x;
1714 / \# [ \t]* ( [^ \t] .* ) /x;
1715
1716Which one you pick depends on which of these expressions better reflects
1717the above specification of comments.
1718
6bda09f9
YO
1719In some literature this construct is called "atomic matching" or
1720"possessive matching".
1721
b9b4dddf
YO
1722Possessive quantifiers are equivalent to putting the item they are applied
1723to inside of one of these constructs. The following equivalences apply:
1724
1725 Quantifier Form Bracketing Form
1726 --------------- ---------------
1727 PAT*+ (?>PAT*)
1728 PAT++ (?>PAT+)
1729 PAT?+ (?>PAT?)
1730 PAT{min,max}+ (?>PAT{min,max})
1731
9d1a5160 1732=item C<(?[ ])>
f4f5fe57 1733
572224ce 1734See L<perlrecharclass/Extended Bracketed Character Classes>.
9d1a5160 1735
e2e6a0f1
YO
1736=back
1737
1738=head2 Special Backtracking Control Verbs
1739
1740B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1741removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1742be noted to avoid problems during upgrades.
1743
1744These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1745otherwise stated the ARG argument is optional; in some cases, it is
1746forbidden.
1747
1748Any pattern containing a special backtracking verb that allows an argument
e1020413 1749has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1750C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1751rules apply:
e2e6a0f1 1752
5d458dd8
YO
1753On failure, the C<$REGERROR> variable will be set to the ARG value of the
1754verb pattern, if the verb was involved in the failure of the match. If the
1755ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1756name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1757none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1758
5d458dd8
YO
1759On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1760the C<$REGMARK> variable will be set to the name of the last
1761C<(*MARK:NAME)> pattern executed. See the explanation for the
1762C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1763
5d458dd8 1764B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1765and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1766readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1767Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1768
1769If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1770argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1771
70ca8714 1772=over 3
e2e6a0f1
YO
1773
1774=item Verbs that take an argument
1775
1776=over 4
1777
5d458dd8 1778=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1779X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1780
5d458dd8
YO
1781This zero-width pattern prunes the backtracking tree at the current point
1782when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1783where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1784A may backtrack as necessary to match. Once it is reached, matching
1785continues in B, which may also backtrack as necessary; however, should B
1786not match, then no further backtracking will take place, and the pattern
1787will fail outright at the current starting position.
54612592
YO
1788
1789The following example counts all the possible matching strings in a
1790pattern (without actually matching any of them).
1791
e2e6a0f1 1792 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1793 print "Count=$count\n";
1794
1795which produces:
1796
1797 aaab
1798 aaa
1799 aa
1800 a
1801 aab
1802 aa
1803 a
1804 ab
1805 a
1806 Count=9
1807
5d458dd8 1808If we add a C<(*PRUNE)> before the count like the following
54612592 1809
5d458dd8 1810 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1811 print "Count=$count\n";
1812
0b928c2f 1813we prevent backtracking and find the count of the longest matching string
353c6505 1814at each matching starting point like so:
54612592
YO
1815
1816 aaab
1817 aab
1818 ab
1819 Count=3
1820
5d458dd8 1821Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1822
5d458dd8
YO
1823See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1824control backtracking. In some cases, the use of C<(*PRUNE)> can be
1825replaced with a C<< (?>pattern) >> with no functional difference; however,
1826C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1827C<< (?>pattern) >> alone.
54612592 1828
5d458dd8
YO
1829=item C<(*SKIP)> C<(*SKIP:NAME)>
1830X<(*SKIP)>
e2e6a0f1 1831
5d458dd8 1832This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1833failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1834to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1835of this pattern. This effectively means that the regex engine "skips" forward
1836to this position on failure and tries to match again, (assuming that
1837there is sufficient room to match).
1838
1839The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1840C<(*MARK:NAME)> was encountered while matching, then it is that position
1841which is used as the "skip point". If no C<(*MARK)> of that name was
1842encountered, then the C<(*SKIP)> operator has no effect. When used
1843without a name the "skip point" is where the match point was when
1844executing the (*SKIP) pattern.
1845
0b928c2f 1846Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1847is twice as long:
1848
d1fbf752
KW
1849 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1850 print "Count=$count\n";
24b23f37
YO
1851
1852outputs
1853
1854 aaab
1855 aaab
1856 Count=2
1857
5d458dd8 1858Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1859executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1860C<(*SKIP)> was executed.
1861
5d458dd8 1862=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1863X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1864
1865This zero-width pattern can be used to mark the point reached in a string
1866when a certain part of the pattern has been successfully matched. This
1867mark may be given a name. A later C<(*SKIP)> pattern will then skip
1868forward to that point if backtracked into on failure. Any number of
b4222fa9 1869C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1870
1871In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1872can be used to "label" a pattern branch, so that after matching, the
1873program can determine which branches of the pattern were involved in the
1874match.
1875
1876When a match is successful, the C<$REGMARK> variable will be set to the
1877name of the most recently executed C<(*MARK:NAME)> that was involved
1878in the match.
1879
1880This can be used to determine which branch of a pattern was matched
c27a5cfe 1881without using a separate capture group for each branch, which in turn
5d458dd8
YO
1882can result in a performance improvement, as perl cannot optimize
1883C</(?:(x)|(y)|(z))/> as efficiently as something like
1884C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1885
1886When a match has failed, and unless another verb has been involved in
1887failing the match and has provided its own name to use, the C<$REGERROR>
1888variable will be set to the name of the most recently executed
1889C<(*MARK:NAME)>.
1890
42ac7c82 1891See L</(*SKIP)> for more details.
5d458dd8 1892
b62d2d15
YO
1893As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1894
5d458dd8
YO
1895=item C<(*THEN)> C<(*THEN:NAME)>
1896
ac9d8485 1897This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1898C<(*PRUNE)>, this verb always matches, and when backtracked into on
1899failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
1900innermost enclosing group (capturing or otherwise) that has alternations.
1901The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1902count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
1903
1904Its name comes from the observation that this operation combined with the
1905alternation operator (C<|>) can be used to create what is essentially a
1906pattern-based if/then/else block:
1907
1908 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1909
1910Note that if this operator is used and NOT inside of an alternation then
1911it acts exactly like the C<(*PRUNE)> operator.
1912
1913 / A (*PRUNE) B /
1914
1915is the same as
1916
1917 / A (*THEN) B /
1918
1919but
1920
25e26d77 1921 / ( A (*THEN) B | C ) /
5d458dd8
YO
1922
1923is not the same as
1924
25e26d77 1925 / ( A (*PRUNE) B | C ) /
5d458dd8
YO
1926
1927as after matching the A but failing on the B the C<(*THEN)> verb will
1928backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1929
cbeadc21
JV
1930=back
1931
1932=item Verbs without an argument
1933
1934=over 4
1935
e2e6a0f1
YO
1936=item C<(*COMMIT)>
1937X<(*COMMIT)>
24b23f37 1938
241e7389 1939This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1940zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1941into on failure it causes the match to fail outright. No further attempts
1942to find a valid match by advancing the start pointer will occur again.
1943For example,
24b23f37 1944
d1fbf752
KW
1945 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
1946 print "Count=$count\n";
24b23f37
YO
1947
1948outputs
1949
1950 aaab
1951 Count=1
1952
e2e6a0f1
YO
1953In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1954does not match, the regex engine will not try any further matching on the
1955rest of the string.
c277df42 1956
e2e6a0f1
YO
1957=item C<(*FAIL)> C<(*F)>
1958X<(*FAIL)> X<(*F)>
9af228c6 1959
e2e6a0f1
YO
1960This pattern matches nothing and always fails. It can be used to force the
1961engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1962fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1963
e2e6a0f1 1964It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1965
e2e6a0f1
YO
1966=item C<(*ACCEPT)>
1967X<(*ACCEPT)>
9af228c6 1968
e2e6a0f1
YO
1969B<WARNING:> This feature is highly experimental. It is not recommended
1970for production code.
9af228c6 1971
e2e6a0f1
YO
1972This pattern matches nothing and causes the end of successful matching at
1973the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1974whether there is actually more to match in the string. When inside of a
0d017f4d 1975nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1976via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1977
c27a5cfe 1978If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1979marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1980For instance:
9af228c6 1981
e2e6a0f1 1982 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1983
e2e6a0f1 1984will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 1985be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 1986string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1987
1988=back
c277df42 1989
a0d0e21e
LW
1990=back
1991
c07a80fd 1992=head2 Backtracking
d74e8afc 1993X<backtrack> X<backtracking>
c07a80fd 1994
35a734be
IZ
1995NOTE: This section presents an abstract approximation of regular
1996expression behavior. For a more rigorous (and complicated) view of
1997the rules involved in selecting a match among possible alternatives,
0d017f4d 1998see L<Combining RE Pieces>.
35a734be 1999
c277df42 2000A fundamental feature of regular expression matching involves the
5a964f20 2001notion called I<backtracking>, which is currently used (when needed)
0d017f4d 2002by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
2003C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
2004internally, but the general principle outlined here is valid.
c07a80fd
PP
2005
2006For a regular expression to match, the I<entire> regular expression must
2007match, not just part of it. So if the beginning of a pattern containing a
2008quantifier succeeds in a way that causes later parts in the pattern to
2009fail, the matching engine backs up and recalculates the beginning
2010part--that's why it's called backtracking.
2011
2012Here is an example of backtracking: Let's say you want to find the
2013word following "foo" in the string "Food is on the foo table.":
2014
2015 $_ = "Food is on the foo table.";
2016 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 2017 print "$2 follows $1.\n";
c07a80fd
PP
2018 }
2019
2020When the match runs, the first part of the regular expression (C<\b(foo)>)
2021finds a possible match right at the beginning of the string, and loads up
2022$1 with "Foo". However, as soon as the matching engine sees that there's
2023no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 2024mistake and starts over again one character after where it had the
c07a80fd
PP
2025tentative match. This time it goes all the way until the next occurrence
2026of "foo". The complete regular expression matches this time, and you get
2027the expected output of "table follows foo."
2028
2029Sometimes minimal matching can help a lot. Imagine you'd like to match
2030everything between "foo" and "bar". Initially, you write something
2031like this:
2032
2033 $_ = "The food is under the bar in the barn.";
2034 if ( /foo(.*)bar/ ) {
f793d64a 2035 print "got <$1>\n";
c07a80fd
PP
2036 }
2037
2038Which perhaps unexpectedly yields:
2039
2040 got <d is under the bar in the >
2041
2042That's because C<.*> was greedy, so you get everything between the
14218588 2043I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
2044to use minimal matching to make sure you get the text between a "foo"
2045and the first "bar" thereafter.
2046
2047 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2048 got <d is under the >
2049
0d017f4d 2050Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2051of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2052So you write this:
2053
2054 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2055 if ( /(.*)(\d*)/ ) { # Wrong!
2056 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2057 }
2058
2059That won't work at all, because C<.*> was greedy and gobbled up the
2060whole string. As C<\d*> can match on an empty string the complete
2061regular expression matched successfully.
2062
8e1088bc 2063 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2064
2065Here are some variants, most of which don't work:
2066
2067 $_ = "I have 2 numbers: 53147";
2068 @pats = qw{
f793d64a
KW
2069 (.*)(\d*)
2070 (.*)(\d+)
2071 (.*?)(\d*)
2072 (.*?)(\d+)
2073 (.*)(\d+)$
2074 (.*?)(\d+)$
2075 (.*)\b(\d+)$
2076 (.*\D)(\d+)$
c07a80fd
PP
2077 };
2078
2079 for $pat (@pats) {
f793d64a
KW
2080 printf "%-12s ", $pat;
2081 if ( /$pat/ ) {
2082 print "<$1> <$2>\n";
2083 } else {
2084 print "FAIL\n";
2085 }
c07a80fd
PP
2086 }
2087
2088That will print out:
2089
2090 (.*)(\d*) <I have 2 numbers: 53147> <>
2091 (.*)(\d+) <I have 2 numbers: 5314> <7>
2092 (.*?)(\d*) <> <>
2093 (.*?)(\d+) <I have > <2>
2094 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2095 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2096 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2097 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2098
2099As you see, this can be a bit tricky. It's important to realize that a
2100regular expression is merely a set of assertions that gives a definition
2101of success. There may be 0, 1, or several different ways that the
2102definition might succeed against a particular string. And if there are
5a964f20
TC
2103multiple ways it might succeed, you need to understand backtracking to
2104know which variety of success you will achieve.
c07a80fd 2105
19799a22 2106When using look-ahead assertions and negations, this can all get even
8b19b778 2107trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2108followed by "123". You might try to write that as
2109
871b0233 2110 $_ = "ABC123";
f793d64a
KW
2111 if ( /^\D*(?!123)/ ) { # Wrong!
2112 print "Yup, no 123 in $_\n";
871b0233 2113 }
c07a80fd
PP
2114
2115But that isn't going to match; at least, not the way you're hoping. It
2116claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2117why that pattern matches, contrary to popular expectations:
c07a80fd 2118
4358a253
SS
2119 $x = 'ABC123';
2120 $y = 'ABC445';
c07a80fd 2121
4358a253
SS
2122 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2123 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2124
4358a253
SS
2125 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2126 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2127
2128This prints
2129
2130 2: got ABC
2131 3: got AB
2132 4: got ABC
2133
5f05dabc 2134You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2135general purpose version of test 1. The important difference between
2136them is that test 3 contains a quantifier (C<\D*>) and so can use
2137backtracking, whereas test 1 will not. What's happening is
2138that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2139non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2140let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2141fail.
14218588 2142
c07a80fd 2143The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2144try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2145a quantifier (C<\D*>) has been used in the regular expression, the
2146search engine can backtrack and retry the match differently
54310121 2147in the hope of matching the complete regular expression.
c07a80fd 2148
5a964f20
TC
2149The pattern really, I<really> wants to succeed, so it uses the
2150standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2151time. Now there's indeed something following "AB" that is not
14218588 2152"123". It's "C123", which suffices.
c07a80fd 2153
14218588
GS
2154We can deal with this by using both an assertion and a negation.
2155We'll say that the first part in $1 must be followed both by a digit
2156and by something that's not "123". Remember that the look-aheads
2157are zero-width expressions--they only look, but don't consume any
2158of the string in their match. So rewriting this way produces what
c07a80fd
PP
2159you'd expect; that is, case 5 will fail, but case 6 succeeds:
2160
4358a253
SS
2161 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2162 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2163
2164 6: got ABC
2165
5a964f20 2166In other words, the two zero-width assertions next to each other work as though
19799a22 2167they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2168matches only if you're at the beginning of the line AND the end of the
2169line simultaneously. The deeper underlying truth is that juxtaposition in
2170regular expressions always means AND, except when you write an explicit OR
2171using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2172although the attempted matches are made at different positions because "a"
2173is not a zero-width assertion, but a one-width assertion.
2174
0d017f4d 2175B<WARNING>: Particularly complicated regular expressions can take
14218588 2176exponential time to solve because of the immense number of possible
0d017f4d 2177ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2178internal optimizations done by the regular expression engine, this will
2179take a painfully long time to run:
c07a80fd 2180
e1901655
IZ
2181 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2182
2183And if you used C<*>'s in the internal groups instead of limiting them
2184to 0 through 5 matches, then it would take forever--or until you ran
2185out of stack space. Moreover, these internal optimizations are not
2186always applicable. For example, if you put C<{0,5}> instead of C<*>
2187on the external group, no current optimization is applicable, and the
2188match takes a long time to finish.
c07a80fd 2189
9da458fc
IZ
2190A powerful tool for optimizing such beasts is what is known as an
2191"independent group",
96090e4f 2192which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2193zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2194the tail match, since they are in "logical" context: only
14218588 2195whether they match is considered relevant. For an example
9da458fc 2196where side-effects of look-ahead I<might> have influenced the
96090e4f 2197following match, see L</C<< (?>pattern) >>>.
c277df42 2198
a0d0e21e 2199=head2 Version 8 Regular Expressions
d74e8afc 2200X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2201
5a964f20 2202In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2203routines, here are the pattern-matching rules not described above.
2204
54310121 2205Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2206with a special meaning described here or above. You can cause
5a964f20 2207characters that normally function as metacharacters to be interpreted
5f05dabc 2208literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2209character; "\\" matches a "\"). This escape mechanism is also required
2210for the character used as the pattern delimiter.
2211
2212A series of characters matches that series of characters in the target
0b928c2f 2213string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2214string.
a0d0e21e
LW
2215
2216You can specify a character class, by enclosing a list of characters
5d458dd8 2217in C<[]>, which will match any character from the list. If the
a0d0e21e 2218first character after the "[" is "^", the class matches any character not
14218588 2219in the list. Within a list, the "-" character specifies a
5a964f20 2220range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2221inclusive. If you want either "-" or "]" itself to be a member of a
2222class, put it at the start of the list (possibly after a "^"), or
2223escape it with a backslash. "-" is also taken literally when it is
2224at the end of the list, just before the closing "]". (The
84850974
DD
2225following all specify the same class of three characters: C<[-az]>,
2226C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2227specifies a class containing twenty-six characters, even on EBCDIC-based
2228character sets.) Also, if you try to use the character
2229classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2230a range, the "-" is understood literally.
a0d0e21e 2231
8ada0baa
JH
2232Note also that the whole range idea is rather unportable between
2233character sets--and even within character sets they may cause results
2234you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2235that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2236[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2237spell out the character sets in full.
2238
54310121 2239Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2240used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2241"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2242of three octal digits, matches the character whose coded character set value
5d458dd8 2243is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2244matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2245matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2246matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2247
2248You can specify a series of alternatives for a pattern using "|" to
2249separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2250or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2251first alternative includes everything from the last pattern delimiter
0b928c2f 2252("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2253the last alternative contains everything from the last "|" to the next
0b928c2f 2254closing pattern delimiter. That's why it's common practice to include
14218588 2255alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2256start and end.
2257
5a964f20 2258Alternatives are tried from left to right, so the first
a3cb178b
GS
2259alternative found for which the entire expression matches, is the one that
2260is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2261example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2262part will match, as that is the first alternative tried, and it successfully
2263matches the target string. (This might not seem important, but it is
2264important when you are capturing matched text using parentheses.)
2265
5a964f20 2266Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2267so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2268
14218588
GS
2269Within a pattern, you may designate subpatterns for later reference
2270by enclosing them in parentheses, and you may refer back to the
2271I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2272\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2273of their opening parenthesis. A backreference matches whatever
2274actually matched the subpattern in the string being examined, not
d8b950dc 2275the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2276match "0x1234 0x4321", but not "0x1234 01234", because subpattern
22771 matched "0x", even though the rule C<0|0x> could potentially match
2278the leading 0 in the second number.
cb1a09d0 2279
0d017f4d 2280=head2 Warning on \1 Instead of $1
cb1a09d0 2281
5a964f20 2282Some people get too used to writing things like:
cb1a09d0
AD
2283
2284 $pattern =~ s/(\W)/\\\1/g;
2285
3ff1c45a
KW
2286This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2287shocking the
cb1a09d0 2288B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2289PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2290the usual double-quoted string means a control-A. The customary Unix
2291meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2292of doing that, you get yourself into trouble if you then add an C</e>
2293modifier.
2294
f793d64a 2295 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2296
2297Or if you try to do
2298
2299 s/(\d+)/\1000/;
2300
2301You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2302C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2303with the operation of matching a backreference. Certainly they mean two
2304different things on the I<left> side of the C<s///>.
9fa51da4 2305
0d017f4d 2306=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2307
19799a22 2308B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2309
2310Regular expressions provide a terse and powerful programming language. As
2311with most other power tools, power comes together with the ability
2312to wreak havoc.
2313
2314A common abuse of this power stems from the ability to make infinite
628afcb5 2315loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2316
2317 'foo' =~ m{ ( o? )* }x;
2318
0d017f4d 2319The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2320in the string is not moved by the match, C<o?> would match again and again
527e91da 2321because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2322is with the looping modifier C<//g>:
2323
2324 @matches = ( 'foo' =~ m{ o? }xg );
2325
2326or
2327
2328 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2329
2330or the loop implied by split().
2331
2332However, long experience has shown that many programming tasks may
14218588
GS
2333be significantly simplified by using repeated subexpressions that
2334may match zero-length substrings. Here's a simple example being:
c84d73f1 2335
d1fbf752 2336 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2337 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2338
9da458fc 2339Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2340the infinite loop>. The rules for this are different for lower-level
527e91da 2341loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2342ones like the C</g> modifier or split() operator.
2343
19799a22
GS
2344The lower-level loops are I<interrupted> (that is, the loop is
2345broken) when Perl detects that a repeated expression matched a
2346zero-length substring. Thus
c84d73f1
IZ
2347
2348 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2349
5d458dd8 2350is made equivalent to
c84d73f1 2351
0b928c2f
FC
2352 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2353
2354For example, this program
2355
2356 #!perl -l
2357 "aaaaab" =~ /
2358 (?:
2359 a # non-zero
2360 | # or
2361 (?{print "hello"}) # print hello whenever this
2362 # branch is tried
2363 (?=(b)) # zero-width assertion
2364 )* # any number of times
2365 /x;
2366 print $&;
2367 print $1;
c84d73f1 2368
0b928c2f
FC
2369prints
2370
2371 hello
2372 aaaaa
2373 b
2374
2375Notice that "hello" is only printed once, as when Perl sees that the sixth
2376iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2377the C<*>.
2378
2379The higher-level loops preserve an additional state between iterations:
5d458dd8 2380whether the last match was zero-length. To break the loop, the following
c84d73f1 2381match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2382This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2383and so the I<second best> match is chosen if the I<best> match is of
2384zero length.
2385
19799a22 2386For example:
c84d73f1
IZ
2387
2388 $_ = 'bar';
2389 s/\w??/<$&>/g;
2390
20fb949f 2391results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2392match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2393best> match is what is matched by C<\w>. Thus zero-length matches
2394alternate with one-character-long matches.
2395
5d458dd8 2396Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2397position one notch further in the string.
2398
19799a22 2399The additional state of being I<matched with zero-length> is associated with
c84d73f1 2400the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2401Zero-length matches at the end of the previous match are ignored
2402during C<split>.
c84d73f1 2403
0d017f4d 2404=head2 Combining RE Pieces
35a734be
IZ
2405
2406Each of the elementary pieces of regular expressions which were described
2407before (such as C<ab> or C<\Z>) could match at most one substring
2408at the given position of the input string. However, in a typical regular
2409expression these elementary pieces are combined into more complicated
0b928c2f 2410patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2411(in these examples C<S> and C<T> are regular subexpressions).
2412
2413Such combinations can include alternatives, leading to a problem of choice:
2414if we match a regular expression C<a|ab> against C<"abc">, will it match
2415substring C<"a"> or C<"ab">? One way to describe which substring is
2416actually matched is the concept of backtracking (see L<"Backtracking">).
2417However, this description is too low-level and makes you think
2418in terms of a particular implementation.
2419
2420Another description starts with notions of "better"/"worse". All the
2421substrings which may be matched by the given regular expression can be
2422sorted from the "best" match to the "worst" match, and it is the "best"
2423match which is chosen. This substitutes the question of "what is chosen?"
2424by the question of "which matches are better, and which are worse?".
2425
2426Again, for elementary pieces there is no such question, since at most
2427one match at a given position is possible. This section describes the
2428notion of better/worse for combining operators. In the description
2429below C<S> and C<T> are regular subexpressions.
2430
13a2d996 2431=over 4
35a734be
IZ
2432
2433=item C<ST>
2434
2435Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2436substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2437which can be matched by C<T>.
35a734be 2438
0b928c2f 2439If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2440match than C<A'B'>.
2441
2442If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2443C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2444
2445=item C<S|T>
2446
2447When C<S> can match, it is a better match than when only C<T> can match.
2448
2449Ordering of two matches for C<S> is the same as for C<S>. Similar for
2450two matches for C<T>.
2451
2452=item C<S{REPEAT_COUNT}>
2453
2454Matches as C<SSS...S> (repeated as many times as necessary).
2455
2456=item C<S{min,max}>
2457
2458Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2459
2460=item C<S{min,max}?>
2461
2462Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2463
2464=item C<S?>, C<S*>, C<S+>
2465
2466Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2467
2468=item C<S??>, C<S*?>, C<S+?>
2469
2470Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2471
c47ff5f1 2472=item C<< (?>S) >>
35a734be
IZ
2473
2474Matches the best match for C<S> and only that.
2475
2476=item C<(?=S)>, C<(?<=S)>
2477
2478Only the best match for C<S> is considered. (This is important only if
2479C<S> has capturing parentheses, and backreferences are used somewhere
2480else in the whole regular expression.)
2481
2482=item C<(?!S)>, C<(?<!S)>
2483
2484For this grouping operator there is no need to describe the ordering, since
2485only whether or not C<S> can match is important.
2486
93f313ef 2487=item C<(??{ EXPR })>, C<(?I<PARNO>)>
35a734be
IZ
2488
2489The ordering is the same as for the regular expression which is
93f313ef 2490the result of EXPR, or the pattern contained by capture group I<PARNO>.
35a734be
IZ
2491
2492=item C<(?(condition)yes-pattern|no-pattern)>
2493
2494Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2495already determined. The ordering of the matches is the same as for the
2496chosen subexpression.
2497
2498=back
2499
2500The above recipes describe the ordering of matches I<at a given position>.
2501One more rule is needed to understand how a match is determined for the
2502whole regular expression: a match at an earlier position is always better
2503than a match at a later position.
2504
0d017f4d 2505=head2 Creating Custom RE Engines
c84d73f1 2506
0b928c2f
FC
2507As of Perl 5.10.0, one can create custom regular expression engines. This
2508is not for the faint of heart, as they have to plug in at the C level. See
2509L<perlreapi> for more details.
2510
2511As an alternative, overloaded constants (see L<overload>) provide a simple
2512way to extend the functionality of the RE engine, by substituting one
2513pattern for another.
c84d73f1
IZ
2514
2515Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2516matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2517characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2518at these positions, so we want to have each C<\Y|> in the place of the
2519more complicated version. We can create a module C<customre> to do
2520this:
2521
2522 package customre;
2523 use overload;
2524
2525 sub import {
2526 shift;
2527 die "No argument to customre::import allowed" if @_;
2528 overload::constant 'qr' => \&convert;
2529 }
2530
2531 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2532
580a9fe1
RGS
2533 # We must also take care of not escaping the legitimate \\Y|
2534 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2535 my %rules = ( '\\' => '\\\\',
f793d64a 2536 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2537 sub convert {
2538 my $re = shift;
5d458dd8 2539 $re =~ s{
c84d73f1
IZ
2540 \\ ( \\ | Y . )
2541 }
5d458dd8 2542 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2543 return $re;
2544 }
2545
2546Now C<use customre> enables the new escape in constant regular
2547expressions, i.e., those without any runtime variable interpolations.
2548As documented in L<overload>, this conversion will work only over
2549literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2550part of this regular expression needs to be converted explicitly
2551(but only if the special meaning of C<\Y|> should be enabled inside $re):
2552
2553 use customre;
2554 $re = <>;
2555 chomp $re;
2556 $re = customre::convert $re;
2557 /\Y|$re\Y|/;
2558
0b928c2f 2559=head2 PCRE/Python Support
1f1031fe 2560
0b928c2f 2561As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2562to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2563Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2564
2565=over 4
2566
ae5648b3 2567=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2568
c27a5cfe 2569Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2570
2571=item C<< (?P=NAME) >>
2572
c27a5cfe 2573Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2574
2575=item C<< (?P>NAME) >>
2576
c27a5cfe 2577Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2578
ee9b8eae 2579=back
1f1031fe 2580
19799a22
GS
2581=head1 BUGS
2582
88c9975e
KW
2583Many regular expression constructs don't work on EBCDIC platforms.
2584
ed7efc79
KW
2585There are a number of issues with regard to case-insensitive matching
2586in Unicode rules. See C<i> under L</Modifiers> above.
2587
9da458fc
IZ
2588This document varies from difficult to understand to completely
2589and utterly opaque. The wandering prose riddled with jargon is
2590hard to fathom in several places.
2591
2592This document needs a rewrite that separates the tutorial content
2593from the reference content.
19799a22
GS
2594
2595=head1 SEE ALSO
9fa51da4 2596
91e0c79e
MJD
2597L<perlrequick>.
2598
2599L<perlretut>.
2600
9b599b2a
GS
2601L<perlop/"Regexp Quote-Like Operators">.
2602
1e66bd83
PP
2603L<perlop/"Gory details of parsing quoted constructs">.
2604
14218588
GS
2605L<perlfaq6>.
2606
9b599b2a
GS
2607L<perlfunc/pos>.
2608
2609L<perllocale>.
2610
fb55449c
JH
2611L<perlebcdic>.
2612
14218588
GS
2613I<Mastering Regular Expressions> by Jeffrey Friedl, published
2614by O'Reilly and Associates.