This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: White-space only
[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
d78f32f6
FC
94In Perl 5.18 and higher this is ignored. ${^PREMATCH}, ${^MATCH}, and
95${^POSTMATCH} will be available after the match regardless of the modifier.
96
e2e6bec7
DN
97=item g and c
98X</g> X</c>
99
100Global matching, and keep the Current position after failed matching.
101Unlike i, m, s and x, these two flags affect the way the regex is used
102rather than the regex itself. See
103L<perlretut/"Using regular expressions in Perl"> for further explanation
104of the g and c modifiers.
105
b6fa137b
FC
106=item a, d, l and u
107X</a> X</d> X</l> X</u>
108
516074bb
KW
109These modifiers, all new in 5.14, affect which character-set semantics
110(Unicode, etc.) are used, as described below in
ed7efc79 111L</Character set modifiers>.
b6fa137b 112
55497cff 113=back
a0d0e21e 114
516074bb
KW
115Regular expression modifiers are usually written in documentation
116as e.g., "the C</x> modifier", even though the delimiter
b6fa137b 117in question might not really be a slash. The modifiers C</imsxadlup>
ab7bb42d 118may also be embedded within the regular expression itself using
ed7efc79 119the C<(?...)> construct, see L</Extended Patterns> below.
b6fa137b 120
ed7efc79
KW
121=head3 /x
122
b6fa137b 123C</x> tells
7b059540 124the regular expression parser to ignore most whitespace that is neither
55497cff 125backslashed nor within a character class. You can use this to break up
4633a7c4 126your regular expression into (slightly) more readable parts. The C<#>
54310121 127character is also treated as a metacharacter introducing a comment,
55497cff 128just as in ordinary Perl code. This also means that if you want real
14218588 129whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a 130class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
131escape them (using backslashes or C<\Q...\E>) or encode them using octal,
132hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
133making Perl's regular expressions more readable. Note that you have to
134be careful not to include the pattern delimiter in the comment--perl has
135no way of knowing you did not intend to close the pattern early. See
136the C-comment deletion code in L<perlop>. Also note that anything inside
7651b971 137a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
0b928c2f 138space interpretation within a single multi-character construct. For
7651b971 139example in C<\x{...}>, regardless of the C</x> modifier, there can be no
9bb1f947 140spaces. Same for a L<quantifier|/Quantifiers> such as C<{3}> or
f9e949fd
KW
141C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<?> and C<:>,
142but can between the C<(> and C<?>. Within any delimiters for such a
143construct, allowed spaces are not affected by C</x>, and depend on the
144construct. For example, C<\x{...}> can't have spaces because hexadecimal
145numbers don't have spaces in them. But, Unicode properties can have spaces, so
0b928c2f 146in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
9bb1f947 147L<perluniprops/Properties accessible through \p{} and \P{}>.
d74e8afc 148X</x>
a0d0e21e 149
ed7efc79
KW
150=head3 Character set modifiers
151
152C</d>, C</u>, C</a>, and C</l>, available starting in 5.14, are called
153the character set modifiers; they affect the character set semantics
154used for the regular expression.
155
808432af
KW
156The C</d>, C</u>, and C</l> modifiers are not likely to be of much use
157to you, and so you need not worry about them very much. They exist for
158Perl's internal use, so that complex regular expression data structures
159can be automatically serialized and later exactly reconstituted,
160including all their nuances. But, since Perl can't keep a secret, and
161there may be rare instances where they are useful, they are documented
162here.
ed7efc79 163
808432af
KW
164The C</a> modifier, on the other hand, may be useful. Its purpose is to
165allow code that is to work mostly on ASCII data to not have to concern
166itself with Unicode.
ca9560b2 167
808432af
KW
168Briefly, C</l> sets the character set to that of whatever B<L>ocale is in
169effect at the time of the execution of the pattern match.
ca9560b2 170
808432af
KW
171C</u> sets the character set to B<U>nicode.
172
173C</a> also sets the character set to Unicode, BUT adds several
174restrictions for B<A>SCII-safe matching.
175
176C</d> is the old, problematic, pre-5.14 B<D>efault character set
177behavior. Its only use is to force that old behavior.
178
179At any given time, exactly one of these modifiers is in effect. Their
180existence allows Perl to keep the originally compiled behavior of a
181regular expression, regardless of what rules are in effect when it is
182actually executed. And if it is interpolated into a larger regex, the
183original's rules continue to apply to it, and only it.
184
185The C</l> and C</u> modifiers are automatically selected for
186regular expressions compiled within the scope of various pragmas,
187and we recommend that in general, you use those pragmas instead of
188specifying these modifiers explicitly. For one thing, the modifiers
189affect only pattern matching, and do not extend to even any replacement
190done, whereas using the pragmas give consistent results for all
191appropriate operations within their scopes. For example,
192
193 s/foo/\Ubar/il
194
195will match "foo" using the locale's rules for case-insensitive matching,
196but the C</l> does not affect how the C<\U> operates. Most likely you
197want both of them to use locale rules. To do this, instead compile the
198regular expression within the scope of C<use locale>. This both
199implicitly adds the C</l> and applies locale rules to the C<\U>. The
200lesson is to C<use locale> and not C</l> explicitly.
201
202Similarly, it would be better to use C<use feature 'unicode_strings'>
203instead of,
204
205 s/foo/\Lbar/iu
206
207to get Unicode rules, as the C<\L> in the former (but not necessarily
208the latter) would also use Unicode rules.
209
210More detail on each of the modifiers follows. Most likely you don't
211need to know this detail for C</l>, C</u>, and C</d>, and can skip ahead
212to L<E<sol>a|/E<sol>a (and E<sol>aa)>.
ca9560b2 213
ed7efc79
KW
214=head4 /l
215
216means to use the current locale's rules (see L<perllocale>) when pattern
217matching. For example, C<\w> will match the "word" characters of that
218locale, and C<"/i"> case-insensitive matching will match according to
219the locale's case folding rules. The locale used will be the one in
220effect at the time of execution of the pattern match. This may not be
221the same as the compilation-time locale, and can differ from one match
222to another if there is an intervening call of the
b6fa137b 223L<setlocale() function|perllocale/The setlocale function>.
ed7efc79
KW
224
225Perl only supports single-byte locales. This means that code points
226above 255 are treated as Unicode no matter what locale is in effect.
227Under Unicode rules, there are a few case-insensitive matches that cross
228the 255/256 boundary. These are disallowed under C</l>. For example,
516074bb
KW
2290xFF (on ASCII platforms) does not caselessly match the character at
2300x178, C<LATIN CAPITAL LETTER Y WITH DIAERESIS>, because 0xFF may not be
231C<LATIN SMALL LETTER Y WITH DIAERESIS> in the current locale, and Perl
232has no way of knowing if that character even exists in the locale, much
233less what code point it is.
ed7efc79
KW
234
235This modifier may be specified to be the default by C<use locale>, but
236see L</Which character set modifier is in effect?>.
b6fa137b
FC
237X</l>
238
ed7efc79
KW
239=head4 /u
240
241means to use Unicode rules when pattern matching. On ASCII platforms,
242this means that the code points between 128 and 255 take on their
808432af
KW
243Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's).
244(Otherwise Perl considers their meanings to be undefined.) Thus,
245under this modifier, the ASCII platform effectively becomes a Unicode
246platform; and hence, for example, C<\w> will match any of the more than
247100_000 word characters in Unicode.
ed7efc79
KW
248
249Unlike most locales, which are specific to a language and country pair,
516074bb
KW
250Unicode classifies all the characters that are letters I<somewhere> in
251the world as
ed7efc79
KW
252C<\w>. For example, your locale might not think that C<LATIN SMALL
253LETTER ETH> is a letter (unless you happen to speak Icelandic), but
254Unicode does. Similarly, all the characters that are decimal digits
255somewhere in the world will match C<\d>; this is hundreds, not 10,
256possible matches. And some of those digits look like some of the 10
257ASCII digits, but mean a different number, so a human could easily think
258a number is a different quantity than it really is. For example,
259C<BENGALI DIGIT FOUR> (U+09EA) looks very much like an
260C<ASCII DIGIT EIGHT> (U+0038). And, C<\d+>, may match strings of digits
261that are a mixture from different writing systems, creating a security
67592e11 262issue. L<Unicode::UCD/num()> can be used to sort
516074bb
KW
263this out. Or the C</a> modifier can be used to force C<\d> to match
264just the ASCII 0 through 9.
ed7efc79 265
516074bb
KW
266Also, under this modifier, case-insensitive matching works on the full
267set of Unicode
ed7efc79
KW
268characters. The C<KELVIN SIGN>, for example matches the letters "k" and
269"K"; and C<LATIN SMALL LIGATURE FF> matches the sequence "ff", which,
270if you're not prepared, might make it look like a hexadecimal constant,
271presenting another potential security issue. See
272L<http://unicode.org/reports/tr36> for a detailed discussion of Unicode
273security issues.
274
6368643f
KW
275On the EBCDIC platforms that Perl handles, the native character set is
276equivalent to Latin-1. Thus this modifier changes behavior only when
ed7efc79
KW
277the C<"/i"> modifier is also specified, and it turns out it affects only
278two characters, giving them full Unicode semantics: the C<MICRO SIGN>
6368643f 279will match the Greek capital and small letters C<MU>, otherwise not; and
ed7efc79
KW
280the C<LATIN CAPITAL LETTER SHARP S> will match any of C<SS>, C<Ss>,
281C<sS>, and C<ss>, otherwise not.
282
283This modifier may be specified to be the default by C<use feature
66cbab2c
KW
284'unicode_strings>, C<use locale ':not_characters'>, or
285C<L<use 5.012|perlfunc/use VERSION>> (or higher),
808432af 286but see L</Which character set modifier is in effect?>.
b6fa137b
FC
287X</u>
288
ed7efc79
KW
289=head4 /d
290
291This modifier means to use the "Default" native rules of the platform
292except when there is cause to use Unicode rules instead, as follows:
293
294=over 4
295
296=item 1
297
298the target string is encoded in UTF-8; or
299
300=item 2
301
302the pattern is encoded in UTF-8; or
303
304=item 3
305
306the pattern explicitly mentions a code point that is above 255 (say by
307C<\x{100}>); or
308
309=item 4
b6fa137b 310
ed7efc79
KW
311the pattern uses a Unicode name (C<\N{...}>); or
312
313=item 5
314
315the pattern uses a Unicode property (C<\p{...}>)
316
317=back
318
319Another mnemonic for this modifier is "Depends", as the rules actually
320used depend on various things, and as a result you can get unexpected
808432af
KW
321results. See L<perlunicode/The "Unicode Bug">. The Unicode Bug has
322become rather infamous, leading to yet another (printable) name for this
323modifier, "Dodgy".
ed7efc79
KW
324
325On ASCII platforms, the native rules are ASCII, and on EBCDIC platforms
326(at least the ones that Perl handles), they are Latin-1.
327
328Here are some examples of how that works on an ASCII platform:
329
330 $str = "\xDF"; # $str is not in UTF-8 format.
331 $str =~ /^\w/; # No match, as $str isn't in UTF-8 format.
332 $str .= "\x{0e0b}"; # Now $str is in UTF-8 format.
333 $str =~ /^\w/; # Match! $str is now in UTF-8 format.
334 chop $str;
335 $str =~ /^\w/; # Still a match! $str remains in UTF-8 format.
336
808432af
KW
337This modifier is automatically selected by default when none of the
338others are, so yet another name for it is "Default".
339
340Because of the unexpected behaviors associated with this modifier, you
341probably should only use it to maintain weird backward compatibilities.
342
343=head4 /a (and /aa)
344
345This modifier stands for ASCII-restrict (or ASCII-safe). This modifier,
346unlike the others, may be doubled-up to increase its effect.
347
348When it appears singly, it causes the sequences C<\d>, C<\s>, C<\w>, and
349the Posix character classes to match only in the ASCII range. They thus
350revert to their pre-5.6, pre-Unicode meanings. Under C</a>, C<\d>
351always means precisely the digits C<"0"> to C<"9">; C<\s> means the five
352characters C<[ \f\n\r\t]>; C<\w> means the 63 characters
353C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
354C<[[:print:]]> match only the appropriate ASCII-range characters.
355
356This modifier is useful for people who only incidentally use Unicode,
357and who do not wish to be burdened with its complexities and security
358concerns.
359
360With C</a>, one can write C<\d> with confidence that it will only match
361ASCII characters, and should the need arise to match beyond ASCII, you
362can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>). There are
363similar C<\p{...}> constructs that can match beyond ASCII both white
364space (see L<perlrecharclass/Whitespace>), and Posix classes (see
365L<perlrecharclass/POSIX Character Classes>). Thus, this modifier
366doesn't mean you can't use Unicode, it means that to get Unicode
367matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
368signals Unicode.
369
370As you would expect, this modifier causes, for example, C<\D> to mean
371the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
372C<\D>, C<\S>, and C<\W>. C<\b> still means to match at the boundary
373between C<\w> and C<\W>, using the C</a> definitions of them (similarly
374for C<\B>).
375
376Otherwise, C</a> behaves like the C</u> modifier, in that
377case-insensitive matching uses Unicode semantics; for example, "k" will
378match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
379points in the Latin1 range, above ASCII will have Unicode rules when it
380comes to case-insensitive matching.
381
382To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
383specify the "a" twice, for example C</aai> or C</aia>. (The first
384occurrence of "a" restricts the C<\d>, etc., and the second occurrence
385adds the C</i> restrictions.) But, note that code points outside the
386ASCII range will use Unicode rules for C</i> matching, so the modifier
387doesn't really restrict things to just ASCII; it just forbids the
388intermixing of ASCII and non-ASCII.
389
390To summarize, this modifier provides protection for applications that
391don't wish to be exposed to all of Unicode. Specifying it twice
392gives added protection.
393
394This modifier may be specified to be the default by C<use re '/a'>
395or C<use re '/aa'>. If you do so, you may actually have occasion to use
396the C</u> modifier explictly if there are a few regular expressions
397where you do want full Unicode rules (but even here, it's best if
398everything were under feature C<"unicode_strings">, along with the
399C<use re '/aa'>). Also see L</Which character set modifier is in
400effect?>.
401X</a>
402X</aa>
403
ed7efc79
KW
404=head4 Which character set modifier is in effect?
405
406Which of these modifiers is in effect at any given point in a regular
808432af
KW
407expression depends on a fairly complex set of interactions. These have
408been designed so that in general you don't have to worry about it, but
409this section gives the gory details. As
ed7efc79
KW
410explained below in L</Extended Patterns> it is possible to explicitly
411specify modifiers that apply only to portions of a regular expression.
412The innermost always has priority over any outer ones, and one applying
6368643f
KW
413to the whole expression has priority over any of the default settings that are
414described in the remainder of this section.
ed7efc79 415
916cec3f 416The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
ed7efc79
KW
417default modifiers (including these) for regular expressions compiled
418within its scope. This pragma has precedence over the other pragmas
516074bb 419listed below that also change the defaults.
ed7efc79
KW
420
421Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
66cbab2c 422and C<L<use feature 'unicode_strings|feature>>, or
ed7efc79
KW
423C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
424C</u> when not in the same scope as either C<L<use locale|perllocale>>
66cbab2c
KW
425or C<L<use bytes|bytes>>.
426(C<L<use locale ':not_characters'|perllocale/Unicode and UTF-8>> also
427sets the default to C</u>, overriding any plain C<use locale>.)
428Unlike the mechanisms mentioned above, these
6368643f
KW
429affect operations besides regular expressions pattern matching, and so
430give more consistent results with other operators, including using
431C<\U>, C<\l>, etc. in substitution replacements.
ed7efc79
KW
432
433If none of the above apply, for backwards compatibility reasons, the
434C</d> modifier is the one in effect by default. As this can lead to
435unexpected results, it is best to specify which other rule set should be
436used.
437
438=head4 Character set modifier behavior prior to Perl 5.14
439
440Prior to 5.14, there were no explicit modifiers, but C</l> was implied
441for regexes compiled within the scope of C<use locale>, and C</d> was
442implied otherwise. However, interpolating a regex into a larger regex
443would ignore the original compilation in favor of whatever was in effect
444at the time of the second compilation. There were a number of
445inconsistencies (bugs) with the C</d> modifier, where Unicode rules
446would be used when inappropriate, and vice versa. C<\p{}> did not imply
447Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
b6fa137b 448
a0d0e21e
LW
449=head2 Regular Expressions
450
04838cea
RGS
451=head3 Metacharacters
452
384f06ae 453The patterns used in Perl pattern matching evolved from those supplied in
14218588 454the Version 8 regex routines. (The routines are derived
19799a22
GS
455(distantly) from Henry Spencer's freely redistributable reimplementation
456of the V8 routines.) See L<Version 8 Regular Expressions> for
457details.
a0d0e21e
LW
458
459In particular the following metacharacters have their standard I<egrep>-ish
460meanings:
d74e8afc
ITB
461X<metacharacter>
462X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
463
a0d0e21e 464
f793d64a
KW
465 \ Quote the next metacharacter
466 ^ Match the beginning of the line
467 . Match any character (except newline)
468 $ Match the end of the line (or before newline at the end)
469 | Alternation
470 () Grouping
471 [] Bracketed Character class
a0d0e21e 472
14218588
GS
473By default, the "^" character is guaranteed to match only the
474beginning of the string, the "$" character only the end (or before the
475newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
476assumption that the string contains only one line. Embedded newlines
477will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 478string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
479newline within the string (except if the newline is the last character in
480the string), and "$" will match before any newline. At the
a0d0e21e
LW
481cost of a little more overhead, you can do this by using the /m modifier
482on the pattern match operator. (Older programs did this by setting C<$*>,
db7cd43a 483but this option was removed in perl 5.10.)
d74e8afc 484X<^> X<$> X</m>
a0d0e21e 485
14218588 486To simplify multi-line substitutions, the "." character never matches a
55497cff 487newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 488the string is a single line--even if it isn't.
d74e8afc 489X<.> X</s>
a0d0e21e 490
04838cea
RGS
491=head3 Quantifiers
492
a0d0e21e 493The following standard quantifiers are recognized:
d74e8afc 494X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 495
f793d64a
KW
496 * Match 0 or more times
497 + Match 1 or more times
498 ? Match 1 or 0 times
499 {n} Match exactly n times
500 {n,} Match at least n times
501 {n,m} Match at least n but not more than m times
a0d0e21e 502
0b928c2f
FC
503(If a curly bracket occurs in any other context and does not form part of
504a backslashed sequence like C<\x{...}>, it is treated
9af81bfe
KW
505as a regular character. In particular, the lower quantifier bound
506is not optional. However, in Perl v5.18, it is planned to issue a
507deprecation warning for all such occurrences, and in Perl v5.20 to
508require literal uses of a curly bracket to be escaped, say by preceding
509them with a backslash or enclosing them within square brackets, (C<"\{">
510or C<"[{]">). This change will allow for future syntax extensions (like
511making the lower bound of a quantifier optional), and better error
512checking of quantifiers. Now, a typo in a quantifier silently causes
513it to be treated as the literal characters. For example,
514
515 /o{4,3}/
516
517looks like a quantifier that matches 0 times, since 4 is greater than 3,
518but it really means to match the sequence of six characters
519S<C<"o { 4 , 3 }">>.)
520
521The "*" quantifier is equivalent to C<{0,}>, the "+"
527e91da 522quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 523to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
524This is usually 32766 on the most common platforms. The actual limit can
525be seen in the error message generated by code such as this:
526
820475bd 527 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 528
54310121
PP
529By default, a quantified subpattern is "greedy", that is, it will match as
530many times as possible (given a particular starting location) while still
531allowing the rest of the pattern to match. If you want it to match the
532minimum number of times possible, follow the quantifier with a "?". Note
533that the meanings don't change, just the "greediness":
0d017f4d 534X<metacharacter> X<greedy> X<greediness>
d74e8afc 535X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 536
f793d64a
KW
537 *? Match 0 or more times, not greedily
538 +? Match 1 or more times, not greedily
539 ?? Match 0 or 1 time, not greedily
0b928c2f 540 {n}? Match exactly n times, not greedily (redundant)
f793d64a
KW
541 {n,}? Match at least n times, not greedily
542 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 543
b9b4dddf
YO
544By default, when a quantified subpattern does not allow the rest of the
545overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 546sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
547as well.
548
f793d64a
KW
549 *+ Match 0 or more times and give nothing back
550 ++ Match 1 or more times and give nothing back
551 ?+ Match 0 or 1 time and give nothing back
552 {n}+ Match exactly n times and give nothing back (redundant)
553 {n,}+ Match at least n times and give nothing back
554 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
555
556For instance,
557
558 'aaaa' =~ /a++a/
559
560will never match, as the C<a++> will gobble up all the C<a>'s in the
561string and won't leave any for the remaining part of the pattern. This
562feature can be extremely useful to give perl hints about where it
563shouldn't backtrack. For instance, the typical "match a double-quoted
564string" problem can be most efficiently performed when written as:
565
566 /"(?:[^"\\]++|\\.)*+"/
567
0d017f4d 568as we know that if the final quote does not match, backtracking will not
0b928c2f
FC
569help. See the independent subexpression
570L</C<< (?>pattern) >>> for more details;
b9b4dddf
YO
571possessive quantifiers are just syntactic sugar for that construct. For
572instance the above example could also be written as follows:
573
574 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
575
04838cea
RGS
576=head3 Escape sequences
577
0b928c2f 578Because patterns are processed as double-quoted strings, the following
a0d0e21e
LW
579also work:
580
f793d64a
KW
581 \t tab (HT, TAB)
582 \n newline (LF, NL)
583 \r return (CR)
584 \f form feed (FF)
585 \a alarm (bell) (BEL)
586 \e escape (think troff) (ESC)
f793d64a 587 \cK control char (example: VT)
dc0d9c48 588 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 589 \N{name} named Unicode character or character sequence
f793d64a 590 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 591 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
592 \l lowercase next char (think vi)
593 \u uppercase next char (think vi)
594 \L lowercase till \E (think vi)
595 \U uppercase till \E (think vi)
596 \Q quote (disable) pattern metacharacters till \E
597 \E end either case modification or quoted section, think vi
a0d0e21e 598
9bb1f947 599Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 600
e1d1eefb 601=head3 Character Classes and other Special Escapes
04838cea 602
a0d0e21e 603In addition, Perl defines the following:
d0b16107 604X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 605
f793d64a
KW
606 Sequence Note Description
607 [...] [1] Match a character according to the rules of the
608 bracketed character class defined by the "...".
609 Example: [a-z] matches "a" or "b" or "c" ... or "z"
610 [[:...:]] [2] Match a character according to the rules of the POSIX
611 character class "..." within the outer bracketed
612 character class. Example: [[:upper:]] matches any
613 uppercase character.
d35dd6c6
KW
614 \w [3] Match a "word" character (alphanumeric plus "_", plus
615 other connector punctuation chars plus Unicode
0b928c2f 616 marks)
f793d64a
KW
617 \W [3] Match a non-"word" character
618 \s [3] Match a whitespace character
619 \S [3] Match a non-whitespace character
620 \d [3] Match a decimal digit character
621 \D [3] Match a non-digit character
622 \pP [3] Match P, named property. Use \p{Prop} for longer names
623 \PP [3] Match non-P
624 \X [4] Match Unicode "eXtended grapheme cluster"
625 \C Match a single C-language char (octet) even if that is
626 part of a larger UTF-8 character. Thus it breaks up
627 characters into their UTF-8 bytes, so you may end up
628 with malformed pieces of UTF-8. Unsupported in
629 lookbehind.
c27a5cfe 630 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
631 '1' may actually be any positive integer.
632 \g1 [5] Backreference to a specific or previous group,
633 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 634 previous group and may optionally be wrapped in
f793d64a
KW
635 curly brackets for safer parsing.
636 \g{name} [5] Named backreference
637 \k<name> [5] Named backreference
638 \K [6] Keep the stuff left of the \K, don't include it in $&
639 \N [7] Any character but \n (experimental). Not affected by
640 /s modifier
641 \v [3] Vertical whitespace
642 \V [3] Not vertical whitespace
643 \h [3] Horizontal whitespace
644 \H [3] Not horizontal whitespace
645 \R [4] Linebreak
e1d1eefb 646
9bb1f947
KW
647=over 4
648
649=item [1]
650
651See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 652
9bb1f947 653=item [2]
b8c5462f 654
9bb1f947 655See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 656
9bb1f947 657=item [3]
5496314a 658
9bb1f947 659See L<perlrecharclass/Backslash sequences> for details.
5496314a 660
9bb1f947 661=item [4]
5496314a 662
9bb1f947 663See L<perlrebackslash/Misc> for details.
d0b16107 664
9bb1f947 665=item [5]
b8c5462f 666
c27a5cfe 667See L</Capture groups> below for details.
93733859 668
9bb1f947 669=item [6]
b8c5462f 670
9bb1f947
KW
671See L</Extended Patterns> below for details.
672
673=item [7]
674
675Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
676character or character sequence whose name is C<NAME>; and similarly
677when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
678code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947
KW
679
680=back
d0b16107 681
04838cea
RGS
682=head3 Assertions
683
a0d0e21e 684Perl defines the following zero-width assertions:
d74e8afc
ITB
685X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
686X<regexp, zero-width assertion>
687X<regular expression, zero-width assertion>
688X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 689
9bb1f947
KW
690 \b Match a word boundary
691 \B Match except at a word boundary
692 \A Match only at beginning of string
693 \Z Match only at end of string, or before newline at the end
694 \z Match only at end of string
695 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 696 of prior m//g)
a0d0e21e 697
14218588 698A word boundary (C<\b>) is a spot between two characters
19799a22
GS
699that has a C<\w> on one side of it and a C<\W> on the other side
700of it (in either order), counting the imaginary characters off the
701beginning and end of the string as matching a C<\W>. (Within
702character classes C<\b> represents backspace rather than a word
703boundary, just as it normally does in any double-quoted string.)
704The C<\A> and C<\Z> are just like "^" and "$", except that they
705won't match multiple times when the C</m> modifier is used, while
706"^" and "$" will match at every internal line boundary. To match
707the actual end of the string and not ignore an optional trailing
708newline, use C<\z>.
d74e8afc 709X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
710
711The C<\G> assertion can be used to chain global matches (using
712C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
713It is also useful when writing C<lex>-like scanners, when you have
714several patterns that you want to match against consequent substrings
0b928c2f 715of your string; see the previous reference. The actual location
19799a22 716where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d 717an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
0b928c2f
FC
718matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
719is modified somewhat, in that contents to the left of C<\G> are
58e23c8d
YO
720not counted when determining the length of the match. Thus the following
721will not match forever:
d74e8afc 722X<\G>
c47ff5f1 723
e761bb84
CO
724 my $string = 'ABC';
725 pos($string) = 1;
726 while ($string =~ /(.\G)/g) {
727 print $1;
728 }
58e23c8d
YO
729
730It will print 'A' and then terminate, as it considers the match to
731be zero-width, and thus will not match at the same position twice in a
732row.
733
734It is worth noting that C<\G> improperly used can result in an infinite
735loop. Take care when using patterns that include C<\G> in an alternation.
736
c27a5cfe 737=head3 Capture groups
04838cea 738
c27a5cfe
KW
739The bracketing construct C<( ... )> creates capture groups (also referred to as
740capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
741the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
742for the second, and so on.
743This is called a I<backreference>.
d74e8afc 744X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 745X<regex, capture group> X<regexp, capture group>
d74e8afc 746X<regular expression, capture buffer> X<backreference>
c27a5cfe 747X<regular expression, capture group> X<backreference>
1f1031fe 748X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
749X<named capture buffer> X<regular expression, named capture buffer>
750X<named capture group> X<regular expression, named capture group>
751X<%+> X<$+{name}> X<< \k<name> >>
752There is no limit to the number of captured substrings that you may use.
753Groups are numbered with the leftmost open parenthesis being number 1, etc. If
754a group did not match, the associated backreference won't match either. (This
755can happen if the group is optional, or in a different branch of an
756alternation.)
757You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
758this form, described below.
759
760You can also refer to capture groups relatively, by using a negative number, so
761that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
762group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
763example:
5624f11d
YO
764
765 /
c27a5cfe
KW
766 (Y) # group 1
767 ( # group 2
768 (X) # group 3
769 \g{-1} # backref to group 3
770 \g{-3} # backref to group 1
5624f11d
YO
771 )
772 /x
773
d8b950dc
KW
774would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
775interpolate regexes into larger regexes and not have to worry about the
776capture groups being renumbered.
777
778You can dispense with numbers altogether and create named capture groups.
779The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
780reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
781also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
782I<name> must not begin with a number, nor contain hyphens.
783When different groups within the same pattern have the same name, any reference
784to that name assumes the leftmost defined group. Named groups count in
785absolute and relative numbering, and so can also be referred to by those
786numbers.
787(It's possible to do things with named capture groups that would otherwise
788require C<(??{})>.)
789
790Capture group contents are dynamically scoped and available to you outside the
791pattern until the end of the enclosing block or until the next successful
792match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
793You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
794etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
795
796Braces are required in referring to named capture groups, but are optional for
797absolute or relative numbered ones. Braces are safer when creating a regex by
798concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
799contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
800is probably not what you intended.
801
802The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
803there were no named nor relative numbered capture groups. Absolute numbered
0b928c2f
FC
804groups were referred to using C<\1>,
805C<\2>, etc., and this notation is still
d8b950dc
KW
806accepted (and likely always will be). But it leads to some ambiguities if
807there are more than 9 capture groups, as C<\10> could mean either the tenth
808capture group, or the character whose ordinal in octal is 010 (a backspace in
809ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
810only if at least 10 left parentheses have opened before it. Likewise C<\11> is
811a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
812And so on. C<\1> through C<\9> are always interpreted as backreferences.
813There are several examples below that illustrate these perils. You can avoid
814the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
815and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
816digits padded with leading zeros, since a leading zero implies an octal
817constant.
d8b950dc
KW
818
819The C<\I<digit>> notation also works in certain circumstances outside
ed7efc79 820the pattern. See L</Warning on \1 Instead of $1> below for details.
81714fb9 821
14218588 822Examples:
a0d0e21e
LW
823
824 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
825
d8b950dc 826 /(.)\g1/ # find first doubled char
81714fb9
YO
827 and print "'$1' is the first doubled character\n";
828
829 /(?<char>.)\k<char>/ # ... a different way
830 and print "'$+{char}' is the first doubled character\n";
831
d8b950dc 832 /(?'char'.)\g1/ # ... mix and match
81714fb9 833 and print "'$1' is the first doubled character\n";
c47ff5f1 834
14218588 835 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
836 $hours = $1;
837 $minutes = $2;
838 $seconds = $3;
a0d0e21e 839 }
c47ff5f1 840
9d860678
KW
841 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
842 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
843 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
844 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
845
846 $a = '(.)\1'; # Creates problems when concatenated.
847 $b = '(.)\g{1}'; # Avoids the problems.
848 "aa" =~ /${a}/; # True
849 "aa" =~ /${b}/; # True
850 "aa0" =~ /${a}0/; # False!
851 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
852 "aa\x08" =~ /${a}0/; # True!
853 "aa\x08" =~ /${b}0/; # False
9d860678 854
14218588
GS
855Several special variables also refer back to portions of the previous
856match. C<$+> returns whatever the last bracket match matched.
857C<$&> returns the entire matched string. (At one point C<$0> did
858also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
859everything before the matched string. C<$'> returns everything
860after the matched string. And C<$^N> contains whatever was matched by
861the most-recently closed group (submatch). C<$^N> can be used in
862extended patterns (see below), for example to assign a submatch to a
81714fb9 863variable.
d74e8afc 864X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 865
d8b950dc
KW
866These special variables, like the C<%+> hash and the numbered match variables
867(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
868until the end of the enclosing block or until the next successful
869match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
870X<$+> X<$^N> X<$&> X<$`> X<$'>
871X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
872
0d017f4d 873B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 874which makes it easier to write code that tests for a series of more
665e98b9
JH
875specific cases and remembers the best match.
876
d78f32f6
FC
877B<WARNING>: If your code is to run on Perl 5.16 or earlier,
878beware that once Perl sees that you need one of C<$&>, C<$`>, or
14218588 879C<$'> anywhere in the program, it has to provide them for every
d78f32f6
FC
880pattern match. This may substantially slow your program. (In Perl 5.18 a
881more efficient mechanism is used, eliminating any slowdown.) Perl
d8b950dc 882uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
883price for each pattern that contains capturing parentheses. (To
884avoid this cost while retaining the grouping behaviour, use the
885extended regular expression C<(?: ... )> instead.) But if you never
886use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
887parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
888if you can, but if you can't (and some algorithms really appreciate
889them), once you've used them once, use them at will, because you've
d78f32f6 890already paid the price.
d74e8afc 891X<$&> X<$`> X<$'>
68dc0745 892
d78f32f6 893As a workaround for this problem, Perl 5.10.0 introduced C<${^PREMATCH}>,
cde0cee5
YO
894C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
895and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 896successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
897The use of these variables incurs no global performance penalty, unlike
898their punctuation char equivalents, however at the trade-off that you
d78f32f6
FC
899have to tell perl when you want to use them. As of Perl 5.18, these three
900variables are equivalent to C<$`>, C<$&> and C<$'>, and C</p> is ignored.
87e95b7f 901X</p> X<p modifier>
cde0cee5 902
9d727203
KW
903=head2 Quoting metacharacters
904
19799a22
GS
905Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
906C<\w>, C<\n>. Unlike some other regular expression languages, there
907are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 908that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
909interpreted as a literal character, not a metacharacter. This was
910once used in a common idiom to disable or quote the special meanings
911of regular expression metacharacters in a string that you want to
36bbe248 912use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
913
914 $pattern =~ s/(\W)/\\$1/g;
915
f1cbbd6e 916(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
917Today it is more common to use the quotemeta() function or the C<\Q>
918metaquoting escape sequence to disable all metacharacters' special
919meanings like this:
a0d0e21e
LW
920
921 /$unquoted\Q$quoted\E$unquoted/
922
9da458fc
IZ
923Beware that if you put literal backslashes (those not inside
924interpolated variables) between C<\Q> and C<\E>, double-quotish
925backslash interpolation may lead to confusing results. If you
926I<need> to use literal backslashes within C<\Q...\E>,
927consult L<perlop/"Gory details of parsing quoted constructs">.
928
736fe711
KW
929C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
930
19799a22
GS
931=head2 Extended Patterns
932
14218588 933Perl also defines a consistent extension syntax for features not
0b928c2f
FC
934found in standard tools like B<awk> and
935B<lex>. The syntax for most of these is a
14218588
GS
936pair of parentheses with a question mark as the first thing within
937the parentheses. The character after the question mark indicates
938the extension.
19799a22 939
14218588
GS
940The stability of these extensions varies widely. Some have been
941part of the core language for many years. Others are experimental
942and may change without warning or be completely removed. Check
943the documentation on an individual feature to verify its current
944status.
19799a22 945
14218588
GS
946A question mark was chosen for this and for the minimal-matching
947construct because 1) question marks are rare in older regular
948expressions, and 2) whenever you see one, you should stop and
0b928c2f 949"question" exactly what is going on. That's psychology....
a0d0e21e 950
70ca8714 951=over 4
a0d0e21e 952
cc6b7395 953=item C<(?#text)>
d74e8afc 954X<(?#)>
a0d0e21e 955
14218588 956A comment. The text is ignored. If the C</x> modifier enables
19799a22 957whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
958the comment as soon as it sees a C<)>, so there is no way to put a literal
959C<)> in the comment.
a0d0e21e 960
cfaf538b 961=item C<(?adlupimsx-imsx)>
fb85c044 962
cfaf538b 963=item C<(?^alupimsx)>
fb85c044 964X<(?)> X<(?^)>
19799a22 965
0b6d1084
JH
966One or more embedded pattern-match modifiers, to be turned on (or
967turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
968the remainder of the enclosing pattern group (if any).
969
fb85c044 970This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 971configuration file, taken from an argument, or specified in a table
0b928c2f
FC
972somewhere. Consider the case where some patterns want to be
973case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 974include C<(?i)> at the front of the pattern. For example:
19799a22
GS
975
976 $pattern = "foobar";
5d458dd8 977 if ( /$pattern/i ) { }
19799a22
GS
978
979 # more flexible:
980
981 $pattern = "(?i)foobar";
5d458dd8 982 if ( /$pattern/ ) { }
19799a22 983
0b6d1084 984These modifiers are restored at the end of the enclosing group. For example,
19799a22 985
d8b950dc 986 ( (?i) blah ) \s+ \g1
19799a22 987
0d017f4d
WL
988will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
989repetition of the previous word, assuming the C</x> modifier, and no C</i>
990modifier outside this group.
19799a22 991
8eb5594e 992These modifiers do not carry over into named subpatterns called in the
dd72e27b 993enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
8eb5594e
DR
994change the case-sensitivity of the "NAME" pattern.
995
dc925305
KW
996Any of these modifiers can be set to apply globally to all regular
997expressions compiled within the scope of a C<use re>. See
a0bbd6ff 998L<re/"'/flags' mode">.
dc925305 999
9de15fec
KW
1000Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1001after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
1002C<"d">) may follow the caret to override it.
1003But a minus sign is not legal with it.
1004
dc925305 1005Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 1006that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 1007C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
1008others, and a maximum of one (or two C<a>'s) may appear in the
1009construct. Thus, for
0b928c2f 1010example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 1011C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1012
1013Note also that the C<p> modifier is special in that its presence
1014anywhere in a pattern has a global effect.
cde0cee5 1015
5a964f20 1016=item C<(?:pattern)>
d74e8afc 1017X<(?:)>
a0d0e21e 1018
cfaf538b 1019=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1020
cfaf538b 1021=item C<(?^aluimsx:pattern)>
fb85c044
KW
1022X<(?^:)>
1023
5a964f20
TC
1024This is for clustering, not capturing; it groups subexpressions like
1025"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1026
5a964f20 1027 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1028
1029is like
1030
5a964f20 1031 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1032
19799a22
GS
1033but doesn't spit out extra fields. It's also cheaper not to capture
1034characters if you don't need to.
a0d0e21e 1035
19799a22 1036Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1037C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1038
1039 /(?s-i:more.*than).*million/i
1040
14218588 1041is equivalent to the more verbose
ca9dfc88
IZ
1042
1043 /(?:(?s-i)more.*than).*million/i
1044
fb85c044 1045Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1046after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1047flags (except C<"d">) may follow the caret, so
fb85c044
KW
1048
1049 (?^x:foo)
1050
1051is equivalent to
1052
1053 (?x-ims:foo)
1054
1055The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1056surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1057modified by any flags specified.
1058
1059The caret allows for simpler stringification of compiled regular
1060expressions. These look like
1061
1062 (?^:pattern)
1063
1064with any non-default flags appearing between the caret and the colon.
1065A test that looks at such stringification thus doesn't need to have the
1066system default flags hard-coded in it, just the caret. If new flags are
1067added to Perl, the meaning of the caret's expansion will change to include
1068the default for those flags, so the test will still work, unchanged.
1069
1070Specifying a negative flag after the caret is an error, as the flag is
1071redundant.
1072
1073Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1074to match at the beginning.
1075
594d7033
YO
1076=item C<(?|pattern)>
1077X<(?|)> X<Branch reset>
1078
1079This is the "branch reset" pattern, which has the special property
c27a5cfe 1080that the capture groups are numbered from the same starting point
99d59c4d 1081in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1082
c27a5cfe 1083Capture groups are numbered from left to right, but inside this
693596a8 1084construct the numbering is restarted for each branch.
4deaaa80 1085
c27a5cfe 1086The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1087following this construct will be numbered as though the construct
1088contained only one branch, that being the one with the most capture
c27a5cfe 1089groups in it.
4deaaa80 1090
0b928c2f 1091This construct is useful when you want to capture one of a
4deaaa80
PJ
1092number of alternative matches.
1093
1094Consider the following pattern. The numbers underneath show in
c27a5cfe 1095which group the captured content will be stored.
594d7033
YO
1096
1097
1098 # before ---------------branch-reset----------- after
1099 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1100 # 1 2 2 3 2 3 4
1101
ab106183
A
1102Be careful when using the branch reset pattern in combination with
1103named captures. Named captures are implemented as being aliases to
c27a5cfe 1104numbered groups holding the captures, and that interferes with the
ab106183
A
1105implementation of the branch reset pattern. If you are using named
1106captures in a branch reset pattern, it's best to use the same names,
1107in the same order, in each of the alternations:
1108
1109 /(?| (?<a> x ) (?<b> y )
1110 | (?<a> z ) (?<b> w )) /x
1111
1112Not doing so may lead to surprises:
1113
1114 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1115 say $+ {a}; # Prints '12'
1116 say $+ {b}; # *Also* prints '12'.
1117
c27a5cfe
KW
1118The problem here is that both the group named C<< a >> and the group
1119named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1120
ee9b8eae
YO
1121=item Look-Around Assertions
1122X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1123
0b928c2f 1124Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1125pattern without including it in C<$&>. Positive assertions match when
1126their subpattern matches, negative assertions match when their subpattern
1127fails. Look-behind matches text up to the current match position,
1128look-ahead matches text following the current match position.
1129
1130=over 4
1131
5a964f20 1132=item C<(?=pattern)>
d74e8afc 1133X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1134
19799a22 1135A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1136matches a word followed by a tab, without including the tab in C<$&>.
1137
5a964f20 1138=item C<(?!pattern)>
d74e8afc 1139X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1140
19799a22 1141A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1142matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1143however that look-ahead and look-behind are NOT the same thing. You cannot
1144use this for look-behind.
7b8d334a 1145
5a964f20 1146If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1147will not do what you want. That's because the C<(?!foo)> is just saying that
1148the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1149match. Use look-behind instead (see below).
c277df42 1150
ee9b8eae
YO
1151=item C<(?<=pattern)> C<\K>
1152X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1153
c47ff5f1 1154A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1155matches a word that follows a tab, without including the tab in C<$&>.
1156Works only for fixed-width look-behind.
c277df42 1157
ee9b8eae
YO
1158There is a special form of this construct, called C<\K>, which causes the
1159regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1160not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1161look-behind. The use of C<\K> inside of another look-around assertion
1162is allowed, but the behaviour is currently not well defined.
1163
c62285ac 1164For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1165equivalent C<< (?<=...) >> construct, and it is especially useful in
1166situations where you want to efficiently remove something following
1167something else in a string. For instance
1168
1169 s/(foo)bar/$1/g;
1170
1171can be rewritten as the much more efficient
1172
1173 s/foo\Kbar//g;
1174
5a964f20 1175=item C<(?<!pattern)>
d74e8afc 1176X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1177
19799a22
GS
1178A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1179matches any occurrence of "foo" that does not follow "bar". Works
1180only for fixed-width look-behind.
c277df42 1181
ee9b8eae
YO
1182=back
1183
81714fb9
YO
1184=item C<(?'NAME'pattern)>
1185
1186=item C<< (?<NAME>pattern) >>
1187X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1188
c27a5cfe 1189A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1190parentheses C<()> but for the additional fact that the group
1191can be referred to by name in various regular expression
1192constructs (like C<\g{NAME}>) and can be accessed by name
1193after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1194for more details on the C<%+> and C<%-> hashes.
81714fb9 1195
c27a5cfe
KW
1196If multiple distinct capture groups have the same name then the
1197$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1198
0d017f4d 1199The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1200
1201B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1202function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1203numbered sequentially regardless of being named or not. Thus in the
1204pattern
1205
1206 /(x)(?<foo>y)(z)/
1207
1208$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1209the opposite which is what a .NET regex hacker might expect.
1210
1f1031fe
YO
1211Currently NAME is restricted to simple identifiers only.
1212In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1213its Unicode extension (see L<utf8>),
1214though it isn't extended by the locale (see L<perllocale>).
81714fb9 1215
1f1031fe 1216B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1217with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1218may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1219support the use of single quotes as a delimiter for the name.
81714fb9 1220
1f1031fe
YO
1221=item C<< \k<NAME> >>
1222
1223=item C<< \k'NAME' >>
81714fb9
YO
1224
1225Named backreference. Similar to numeric backreferences, except that
1226the group is designated by name and not number. If multiple groups
1227have the same name then it refers to the leftmost defined group in
1228the current match.
1229
0d017f4d 1230It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1231earlier in the pattern.
1232
1233Both forms are equivalent.
1234
1f1031fe 1235B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1236with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1237may be used instead of C<< \k<NAME> >>.
1f1031fe 1238
cc6b7395 1239=item C<(?{ code })>
d74e8afc 1240X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1241
19799a22 1242B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1243experimental, and may be changed without notice. Code executed that
1244has side effects may not perform identically from version to version
e128ab2c
DM
1245due to the effect of future optimisations in the regex engine. The
1246implementation of this feature was radically overhauled for the 5.18.0
1247release, and its behaviour in earlier versions of perl was much buggier,
1248especially in relation to parsing, lexical vars, scoping, recursion and
1249reentrancy.
c277df42 1250
e128ab2c
DM
1251This zero-width assertion executes any embedded Perl code. It always
1252succeeds, and its return value is set as C<$^R>.
19799a22 1253
e128ab2c
DM
1254In literal patterns, the code is parsed at the same time as the
1255surrounding code. While within the pattern, control is passed temporarily
1256back to the perl parser, until the logically-balancing closing brace is
1257encountered. This is similar to the way that an array index expression in
1258a literal string is handled, for example
77ea4f6d 1259
e128ab2c
DM
1260 "abc$array[ 1 + f('[') + g()]def"
1261
1262In particular, braces do not need to be balanced:
1263
576fa024 1264 s/abc(?{ f('{'); })/def/
e128ab2c
DM
1265
1266Even in a pattern that is interpolated and compiled at run-time, literal
1267code blocks will be compiled once, at perl compile time; the following
1268prints "ABCD":
1269
1270 print "D";
1271 my $qr = qr/(?{ BEGIN { print "A" } })/;
1272 my $foo = "foo";
1273 /$foo$qr(?{ BEGIN { print "B" } })/;
1274 BEGIN { print "C" }
1275
1276In patterns where the text of the code is derived from run-time
1277information rather than appearing literally in a source code /pattern/,
1278the code is compiled at the same time that the pattern is compiled, and
5771dda0 1279for reasons of security, C<use re 'eval'> must be in scope. This is to
e128ab2c
DM
1280stop user-supplied patterns containing code snippets from being
1281executable.
1282
5771dda0 1283In situations where you need to enable this with C<use re 'eval'>, you should
e128ab2c
DM
1284also have taint checking enabled. Better yet, use the carefully
1285constrained evaluation within a Safe compartment. See L<perlsec> for
1286details about both these mechanisms.
1287
1288From the viewpoint of parsing, lexical variable scope and closures,
1289
1290 /AAA(?{ BBB })CCC/
1291
1292behaves approximately like
1293
1294 /AAA/ && do { BBB } && /CCC/
1295
1296Similarly,
1297
1298 qr/AAA(?{ BBB })CCC/
1299
1300behaves approximately like
77ea4f6d 1301
e128ab2c
DM
1302 sub { /AAA/ && do { BBB } && /CCC/ }
1303
1304In particular:
1305
1306 { my $i = 1; $r = qr/(?{ print $i })/ }
1307 my $i = 2;
1308 /$r/; # prints "1"
1309
1310Inside a C<(?{...})> block, C<$_> refers to the string the regular
754091cb 1311expression is matching against. You can also use C<pos()> to know what is
fa11829f 1312the current position of matching within this string.
754091cb 1313
e128ab2c
DM
1314The code block introduces a new scope from the perspective of lexical
1315variable declarations, but B<not> from the perspective of C<local> and
1316similar localizing behaviours. So later code blocks within the same
1317pattern will still see the values which were localized in earlier blocks.
1318These accumulated localizations are undone either at the end of a
1319successful match, or if the assertion is backtracked (compare
1320L<"Backtracking">). For example,
b9ac3b5b
GS
1321
1322 $_ = 'a' x 8;
5d458dd8 1323 m<
d1fbf752 1324 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1325 (
5d458dd8 1326 a
b9ac3b5b 1327 (?{
d1fbf752
KW
1328 local $cnt = $cnt + 1; # Update $cnt,
1329 # backtracking-safe.
b9ac3b5b 1330 })
5d458dd8 1331 )*
b9ac3b5b 1332 aaaa
d1fbf752
KW
1333 (?{ $res = $cnt }) # On success copy to
1334 # non-localized location.
b9ac3b5b
GS
1335 >x;
1336
e128ab2c
DM
1337will initially increment C<$cnt> up to 8; then during backtracking, its
1338value will be unwound back to 4, which is the value assigned to C<$res>.
1339At the end of the regex execution, $cnt will be wound back to its initial
1340value of 0.
1341
1342This assertion may be used as the condition in a
b9ac3b5b 1343
e128ab2c
DM
1344 (?(condition)yes-pattern|no-pattern)
1345
1346switch. If I<not> used in this way, the result of evaluation of C<code>
1347is put into the special variable C<$^R>. This happens immediately, so
1348C<$^R> can be used from other C<(?{ code })> assertions inside the same
1349regular expression.
b9ac3b5b 1350
19799a22
GS
1351The assignment to C<$^R> above is properly localized, so the old
1352value of C<$^R> is restored if the assertion is backtracked; compare
1353L<"Backtracking">.
b9ac3b5b 1354
e128ab2c
DM
1355Note that the special variable C<$^N> is particularly useful with code
1356blocks to capture the results of submatches in variables without having to
1357keep track of the number of nested parentheses. For example:
1358
1359 $_ = "The brown fox jumps over the lazy dog";
1360 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1361 print "color = $color, animal = $animal\n";
1362
8988a1bb 1363
14455d6c 1364=item C<(??{ code })>
d74e8afc
ITB
1365X<(??{})>
1366X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1367
19799a22 1368B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1369experimental, and may be changed without notice. Code executed that
1370has side effects may not perform identically from version to version
1371due to the effect of future optimisations in the regex engine.
0f5d15d6 1372
e128ab2c
DM
1373This is a "postponed" regular subexpression. It behaves in I<exactly> the
1374same way as a C<(?{ code })> code block as described above, except that
1375its return value, rather than being assigned to C<$^R>, is treated as a
1376pattern, compiled if it's a string (or used as-is if its a qr// object),
1377then matched as if it were inserted instead of this construct.
6bda09f9 1378
e128ab2c
DM
1379During the matching of this sub-pattern, it has its own set of
1380captures which are valid during the sub-match, but are discarded once
1381control returns to the main pattern. For example, the following matches,
1382with the inner pattern capturing "B" and matching "BB", while the outer
1383pattern captures "A";
1384
1385 my $inner = '(.)\1';
1386 "ABBA" =~ /^(.)(??{ $inner })\1/;
1387 print $1; # prints "A";
6bda09f9 1388
e128ab2c
DM
1389Note that this means that there is no way for the inner pattern to refer
1390to a capture group defined outside. (The code block itself can use C<$1>,
1391etc., to refer to the enclosing pattern's capture groups.) Thus, although
0f5d15d6 1392
e128ab2c
DM
1393 ('a' x 100)=~/(??{'(.)' x 100})/
1394
1395I<will> match, it will I<not> set $1 on exit.
19799a22
GS
1396
1397The following pattern matches a parenthesized group:
0f5d15d6 1398
d1fbf752
KW
1399 $re = qr{
1400 \(
1401 (?:
1402 (?> [^()]+ ) # Non-parens without backtracking
1403 |
1404 (??{ $re }) # Group with matching parens
1405 )*
1406 \)
1407 }x;
0f5d15d6 1408
6bda09f9
YO
1409See also C<(?PARNO)> for a different, more efficient way to accomplish
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
542fa716
YO
1416=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
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
894be9b7 1428PARNO 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
542fa716 1431C<(?R)>. If 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
0d017f4d
WL
1495Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
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
1594not be as efficient, as the optimiser is not very clever about
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 $1, one should use either
1711one of these:
1712
1713 / (?> \# [ \t]* ) ( .+ ) /x;
1714 / \# [ \t]* ( [^ \t] .* ) /x;
1715
1716Which one you pick depends on which of these expressions better reflects
1717the above specification of comments.
1718
6bda09f9
YO
1719In some literature this construct is called "atomic matching" or
1720"possessive matching".
1721
b9b4dddf
YO
1722Possessive quantifiers are equivalent to putting the item they are applied
1723to inside of one of these constructs. The following equivalences apply:
1724
1725 Quantifier Form Bracketing Form
1726 --------------- ---------------
1727 PAT*+ (?>PAT*)
1728 PAT++ (?>PAT+)
1729 PAT?+ (?>PAT?)
1730 PAT{min,max}+ (?>PAT{min,max})
1731
e2e6a0f1
YO
1732=back
1733
1734=head2 Special Backtracking Control Verbs
1735
1736B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1737removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1738be noted to avoid problems during upgrades.
1739
1740These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1741otherwise stated the ARG argument is optional; in some cases, it is
1742forbidden.
1743
1744Any pattern containing a special backtracking verb that allows an argument
e1020413 1745has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1746C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1747rules apply:
e2e6a0f1 1748
5d458dd8
YO
1749On failure, the C<$REGERROR> variable will be set to the ARG value of the
1750verb pattern, if the verb was involved in the failure of the match. If the
1751ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1752name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1753none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1754
5d458dd8
YO
1755On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1756the C<$REGMARK> variable will be set to the name of the last
1757C<(*MARK:NAME)> pattern executed. See the explanation for the
1758C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1759
5d458dd8 1760B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1761and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1762readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1763Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1764
1765If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1766argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1767
70ca8714 1768=over 3
e2e6a0f1
YO
1769
1770=item Verbs that take an argument
1771
1772=over 4
1773
5d458dd8 1774=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1775X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1776
5d458dd8
YO
1777This zero-width pattern prunes the backtracking tree at the current point
1778when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1779where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1780A may backtrack as necessary to match. Once it is reached, matching
1781continues in B, which may also backtrack as necessary; however, should B
1782not match, then no further backtracking will take place, and the pattern
1783will fail outright at the current starting position.
54612592
YO
1784
1785The following example counts all the possible matching strings in a
1786pattern (without actually matching any of them).
1787
e2e6a0f1 1788 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1789 print "Count=$count\n";
1790
1791which produces:
1792
1793 aaab
1794 aaa
1795 aa
1796 a
1797 aab
1798 aa
1799 a
1800 ab
1801 a
1802 Count=9
1803
5d458dd8 1804If we add a C<(*PRUNE)> before the count like the following
54612592 1805
5d458dd8 1806 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1807 print "Count=$count\n";
1808
0b928c2f 1809we prevent backtracking and find the count of the longest matching string
353c6505 1810at each matching starting point like so:
54612592
YO
1811
1812 aaab
1813 aab
1814 ab
1815 Count=3
1816
5d458dd8 1817Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1818
5d458dd8
YO
1819See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1820control backtracking. In some cases, the use of C<(*PRUNE)> can be
1821replaced with a C<< (?>pattern) >> with no functional difference; however,
1822C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1823C<< (?>pattern) >> alone.
54612592 1824
5d458dd8
YO
1825=item C<(*SKIP)> C<(*SKIP:NAME)>
1826X<(*SKIP)>
e2e6a0f1 1827
5d458dd8 1828This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1829failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1830to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1831of this pattern. This effectively means that the regex engine "skips" forward
1832to this position on failure and tries to match again, (assuming that
1833there is sufficient room to match).
1834
1835The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1836C<(*MARK:NAME)> was encountered while matching, then it is that position
1837which is used as the "skip point". If no C<(*MARK)> of that name was
1838encountered, then the C<(*SKIP)> operator has no effect. When used
1839without a name the "skip point" is where the match point was when
1840executing the (*SKIP) pattern.
1841
0b928c2f 1842Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1843is twice as long:
1844
d1fbf752
KW
1845 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1846 print "Count=$count\n";
24b23f37
YO
1847
1848outputs
1849
1850 aaab
1851 aaab
1852 Count=2
1853
5d458dd8 1854Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1855executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1856C<(*SKIP)> was executed.
1857
5d458dd8 1858=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1859X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1860
1861This zero-width pattern can be used to mark the point reached in a string
1862when a certain part of the pattern has been successfully matched. This
1863mark may be given a name. A later C<(*SKIP)> pattern will then skip
1864forward to that point if backtracked into on failure. Any number of
b4222fa9 1865C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1866
1867In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1868can be used to "label" a pattern branch, so that after matching, the
1869program can determine which branches of the pattern were involved in the
1870match.
1871
1872When a match is successful, the C<$REGMARK> variable will be set to the
1873name of the most recently executed C<(*MARK:NAME)> that was involved
1874in the match.
1875
1876This can be used to determine which branch of a pattern was matched
c27a5cfe 1877without using a separate capture group for each branch, which in turn
5d458dd8
YO
1878can result in a performance improvement, as perl cannot optimize
1879C</(?:(x)|(y)|(z))/> as efficiently as something like
1880C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1881
1882When a match has failed, and unless another verb has been involved in
1883failing the match and has provided its own name to use, the C<$REGERROR>
1884variable will be set to the name of the most recently executed
1885C<(*MARK:NAME)>.
1886
42ac7c82 1887See L</(*SKIP)> for more details.
5d458dd8 1888
b62d2d15
YO
1889As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1890
5d458dd8
YO
1891=item C<(*THEN)> C<(*THEN:NAME)>
1892
ac9d8485 1893This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1894C<(*PRUNE)>, this verb always matches, and when backtracked into on
1895failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
1896innermost enclosing group (capturing or otherwise) that has alternations.
1897The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1898count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
1899
1900Its name comes from the observation that this operation combined with the
1901alternation operator (C<|>) can be used to create what is essentially a
1902pattern-based if/then/else block:
1903
1904 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1905
1906Note that if this operator is used and NOT inside of an alternation then
1907it acts exactly like the C<(*PRUNE)> operator.
1908
1909 / A (*PRUNE) B /
1910
1911is the same as
1912
1913 / A (*THEN) B /
1914
1915but
1916
25e26d77 1917 / ( A (*THEN) B | C ) /
5d458dd8
YO
1918
1919is not the same as
1920
25e26d77 1921 / ( A (*PRUNE) B | C ) /
5d458dd8
YO
1922
1923as after matching the A but failing on the B the C<(*THEN)> verb will
1924backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1925
cbeadc21
JV
1926=back
1927
1928=item Verbs without an argument
1929
1930=over 4
1931
e2e6a0f1
YO
1932=item C<(*COMMIT)>
1933X<(*COMMIT)>
24b23f37 1934
241e7389 1935This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1936zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1937into on failure it causes the match to fail outright. No further attempts
1938to find a valid match by advancing the start pointer will occur again.
1939For example,
24b23f37 1940
d1fbf752
KW
1941 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
1942 print "Count=$count\n";
24b23f37
YO
1943
1944outputs
1945
1946 aaab
1947 Count=1
1948
e2e6a0f1
YO
1949In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1950does not match, the regex engine will not try any further matching on the
1951rest of the string.
c277df42 1952
e2e6a0f1
YO
1953=item C<(*FAIL)> C<(*F)>
1954X<(*FAIL)> X<(*F)>
9af228c6 1955
e2e6a0f1
YO
1956This pattern matches nothing and always fails. It can be used to force the
1957engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1958fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1959
e2e6a0f1 1960It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1961
e2e6a0f1
YO
1962=item C<(*ACCEPT)>
1963X<(*ACCEPT)>
9af228c6 1964
e2e6a0f1
YO
1965B<WARNING:> This feature is highly experimental. It is not recommended
1966for production code.
9af228c6 1967
e2e6a0f1
YO
1968This pattern matches nothing and causes the end of successful matching at
1969the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1970whether there is actually more to match in the string. When inside of a
0d017f4d 1971nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1972via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1973
c27a5cfe 1974If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1975marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1976For instance:
9af228c6 1977
e2e6a0f1 1978 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1979
e2e6a0f1 1980will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 1981be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 1982string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1983
1984=back
c277df42 1985
a0d0e21e
LW
1986=back
1987
c07a80fd 1988=head2 Backtracking
d74e8afc 1989X<backtrack> X<backtracking>
c07a80fd 1990
35a734be
IZ
1991NOTE: This section presents an abstract approximation of regular
1992expression behavior. For a more rigorous (and complicated) view of
1993the rules involved in selecting a match among possible alternatives,
0d017f4d 1994see L<Combining RE Pieces>.
35a734be 1995
c277df42 1996A fundamental feature of regular expression matching involves the
5a964f20 1997notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1998by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1999C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
2000internally, but the general principle outlined here is valid.
c07a80fd
PP
2001
2002For a regular expression to match, the I<entire> regular expression must
2003match, not just part of it. So if the beginning of a pattern containing a
2004quantifier succeeds in a way that causes later parts in the pattern to
2005fail, the matching engine backs up and recalculates the beginning
2006part--that's why it's called backtracking.
2007
2008Here is an example of backtracking: Let's say you want to find the
2009word following "foo" in the string "Food is on the foo table.":
2010
2011 $_ = "Food is on the foo table.";
2012 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 2013 print "$2 follows $1.\n";
c07a80fd
PP
2014 }
2015
2016When the match runs, the first part of the regular expression (C<\b(foo)>)
2017finds a possible match right at the beginning of the string, and loads up
2018$1 with "Foo". However, as soon as the matching engine sees that there's
2019no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 2020mistake and starts over again one character after where it had the
c07a80fd
PP
2021tentative match. This time it goes all the way until the next occurrence
2022of "foo". The complete regular expression matches this time, and you get
2023the expected output of "table follows foo."
2024
2025Sometimes minimal matching can help a lot. Imagine you'd like to match
2026everything between "foo" and "bar". Initially, you write something
2027like this:
2028
2029 $_ = "The food is under the bar in the barn.";
2030 if ( /foo(.*)bar/ ) {
f793d64a 2031 print "got <$1>\n";
c07a80fd
PP
2032 }
2033
2034Which perhaps unexpectedly yields:
2035
2036 got <d is under the bar in the >
2037
2038That's because C<.*> was greedy, so you get everything between the
14218588 2039I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
2040to use minimal matching to make sure you get the text between a "foo"
2041and the first "bar" thereafter.
2042
2043 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2044 got <d is under the >
2045
0d017f4d 2046Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2047of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2048So you write this:
2049
2050 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2051 if ( /(.*)(\d*)/ ) { # Wrong!
2052 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2053 }
2054
2055That won't work at all, because C<.*> was greedy and gobbled up the
2056whole string. As C<\d*> can match on an empty string the complete
2057regular expression matched successfully.
2058
8e1088bc 2059 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2060
2061Here are some variants, most of which don't work:
2062
2063 $_ = "I have 2 numbers: 53147";
2064 @pats = qw{
f793d64a
KW
2065 (.*)(\d*)
2066 (.*)(\d+)
2067 (.*?)(\d*)
2068 (.*?)(\d+)
2069 (.*)(\d+)$
2070 (.*?)(\d+)$
2071 (.*)\b(\d+)$
2072 (.*\D)(\d+)$
c07a80fd
PP
2073 };
2074
2075 for $pat (@pats) {
f793d64a
KW
2076 printf "%-12s ", $pat;
2077 if ( /$pat/ ) {
2078 print "<$1> <$2>\n";
2079 } else {
2080 print "FAIL\n";
2081 }
c07a80fd
PP
2082 }
2083
2084That will print out:
2085
2086 (.*)(\d*) <I have 2 numbers: 53147> <>
2087 (.*)(\d+) <I have 2 numbers: 5314> <7>
2088 (.*?)(\d*) <> <>
2089 (.*?)(\d+) <I have > <2>
2090 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2091 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2092 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2093 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2094
2095As you see, this can be a bit tricky. It's important to realize that a
2096regular expression is merely a set of assertions that gives a definition
2097of success. There may be 0, 1, or several different ways that the
2098definition might succeed against a particular string. And if there are
5a964f20
TC
2099multiple ways it might succeed, you need to understand backtracking to
2100know which variety of success you will achieve.
c07a80fd 2101
19799a22 2102When using look-ahead assertions and negations, this can all get even
8b19b778 2103trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2104followed by "123". You might try to write that as
2105
871b0233 2106 $_ = "ABC123";
f793d64a
KW
2107 if ( /^\D*(?!123)/ ) { # Wrong!
2108 print "Yup, no 123 in $_\n";
871b0233 2109 }
c07a80fd
PP
2110
2111But that isn't going to match; at least, not the way you're hoping. It
2112claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2113why that pattern matches, contrary to popular expectations:
c07a80fd 2114
4358a253
SS
2115 $x = 'ABC123';
2116 $y = 'ABC445';
c07a80fd 2117
4358a253
SS
2118 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2119 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2120
4358a253
SS
2121 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2122 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2123
2124This prints
2125
2126 2: got ABC
2127 3: got AB
2128 4: got ABC
2129
5f05dabc 2130You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2131general purpose version of test 1. The important difference between
2132them is that test 3 contains a quantifier (C<\D*>) and so can use
2133backtracking, whereas test 1 will not. What's happening is
2134that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2135non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2136let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2137fail.
14218588 2138
c07a80fd 2139The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2140try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2141a quantifier (C<\D*>) has been used in the regular expression, the
2142search engine can backtrack and retry the match differently
54310121 2143in the hope of matching the complete regular expression.
c07a80fd 2144
5a964f20
TC
2145The pattern really, I<really> wants to succeed, so it uses the
2146standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2147time. Now there's indeed something following "AB" that is not
14218588 2148"123". It's "C123", which suffices.
c07a80fd 2149
14218588
GS
2150We can deal with this by using both an assertion and a negation.
2151We'll say that the first part in $1 must be followed both by a digit
2152and by something that's not "123". Remember that the look-aheads
2153are zero-width expressions--they only look, but don't consume any
2154of the string in their match. So rewriting this way produces what
c07a80fd
PP
2155you'd expect; that is, case 5 will fail, but case 6 succeeds:
2156
4358a253
SS
2157 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2158 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2159
2160 6: got ABC
2161
5a964f20 2162In other words, the two zero-width assertions next to each other work as though
19799a22 2163they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2164matches only if you're at the beginning of the line AND the end of the
2165line simultaneously. The deeper underlying truth is that juxtaposition in
2166regular expressions always means AND, except when you write an explicit OR
2167using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2168although the attempted matches are made at different positions because "a"
2169is not a zero-width assertion, but a one-width assertion.
2170
0d017f4d 2171B<WARNING>: Particularly complicated regular expressions can take
14218588 2172exponential time to solve because of the immense number of possible
0d017f4d 2173ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2174internal optimizations done by the regular expression engine, this will
2175take a painfully long time to run:
c07a80fd 2176
e1901655
IZ
2177 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2178
2179And if you used C<*>'s in the internal groups instead of limiting them
2180to 0 through 5 matches, then it would take forever--or until you ran
2181out of stack space. Moreover, these internal optimizations are not
2182always applicable. For example, if you put C<{0,5}> instead of C<*>
2183on the external group, no current optimization is applicable, and the
2184match takes a long time to finish.
c07a80fd 2185
9da458fc
IZ
2186A powerful tool for optimizing such beasts is what is known as an
2187"independent group",
96090e4f 2188which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2189zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2190the tail match, since they are in "logical" context: only
14218588 2191whether they match is considered relevant. For an example
9da458fc 2192where side-effects of look-ahead I<might> have influenced the
96090e4f 2193following match, see L</C<< (?>pattern) >>>.
c277df42 2194
a0d0e21e 2195=head2 Version 8 Regular Expressions
d74e8afc 2196X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2197
5a964f20 2198In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2199routines, here are the pattern-matching rules not described above.
2200
54310121 2201Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2202with a special meaning described here or above. You can cause
5a964f20 2203characters that normally function as metacharacters to be interpreted
5f05dabc 2204literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2205character; "\\" matches a "\"). This escape mechanism is also required
2206for the character used as the pattern delimiter.
2207
2208A series of characters matches that series of characters in the target
0b928c2f 2209string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2210string.
a0d0e21e
LW
2211
2212You can specify a character class, by enclosing a list of characters
5d458dd8 2213in C<[]>, which will match any character from the list. If the
a0d0e21e 2214first character after the "[" is "^", the class matches any character not
14218588 2215in the list. Within a list, the "-" character specifies a
5a964f20 2216range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2217inclusive. If you want either "-" or "]" itself to be a member of a
2218class, put it at the start of the list (possibly after a "^"), or
2219escape it with a backslash. "-" is also taken literally when it is
2220at the end of the list, just before the closing "]". (The
84850974
DD
2221following all specify the same class of three characters: C<[-az]>,
2222C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2223specifies a class containing twenty-six characters, even on EBCDIC-based
2224character sets.) Also, if you try to use the character
2225classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2226a range, the "-" is understood literally.
a0d0e21e 2227
8ada0baa
JH
2228Note also that the whole range idea is rather unportable between
2229character sets--and even within character sets they may cause results
2230you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2231that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2232[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2233spell out the character sets in full.
2234
54310121 2235Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2236used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2237"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2238of three octal digits, matches the character whose coded character set value
5d458dd8 2239is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2240matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2241matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2242matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2243
2244You can specify a series of alternatives for a pattern using "|" to
2245separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2246or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2247first alternative includes everything from the last pattern delimiter
0b928c2f 2248("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2249the last alternative contains everything from the last "|" to the next
0b928c2f 2250closing pattern delimiter. That's why it's common practice to include
14218588 2251alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2252start and end.
2253
5a964f20 2254Alternatives are tried from left to right, so the first
a3cb178b
GS
2255alternative found for which the entire expression matches, is the one that
2256is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2257example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2258part will match, as that is the first alternative tried, and it successfully
2259matches the target string. (This might not seem important, but it is
2260important when you are capturing matched text using parentheses.)
2261
5a964f20 2262Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2263so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2264
14218588
GS
2265Within a pattern, you may designate subpatterns for later reference
2266by enclosing them in parentheses, and you may refer back to the
2267I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2268\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2269of their opening parenthesis. A backreference matches whatever
2270actually matched the subpattern in the string being examined, not
d8b950dc 2271the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2272match "0x1234 0x4321", but not "0x1234 01234", because subpattern
22731 matched "0x", even though the rule C<0|0x> could potentially match
2274the leading 0 in the second number.
cb1a09d0 2275
0d017f4d 2276=head2 Warning on \1 Instead of $1
cb1a09d0 2277
5a964f20 2278Some people get too used to writing things like:
cb1a09d0
AD
2279
2280 $pattern =~ s/(\W)/\\\1/g;
2281
3ff1c45a
KW
2282This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2283shocking the
cb1a09d0 2284B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2285PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2286the usual double-quoted string means a control-A. The customary Unix
2287meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2288of doing that, you get yourself into trouble if you then add an C</e>
2289modifier.
2290
f793d64a 2291 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2292
2293Or if you try to do
2294
2295 s/(\d+)/\1000/;
2296
2297You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2298C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2299with the operation of matching a backreference. Certainly they mean two
2300different things on the I<left> side of the C<s///>.
9fa51da4 2301
0d017f4d 2302=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2303
19799a22 2304B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2305
2306Regular expressions provide a terse and powerful programming language. As
2307with most other power tools, power comes together with the ability
2308to wreak havoc.
2309
2310A common abuse of this power stems from the ability to make infinite
628afcb5 2311loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2312
2313 'foo' =~ m{ ( o? )* }x;
2314
0d017f4d 2315The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2316in the string is not moved by the match, C<o?> would match again and again
527e91da 2317because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2318is with the looping modifier C<//g>:
2319
2320 @matches = ( 'foo' =~ m{ o? }xg );
2321
2322or
2323
2324 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2325
2326or the loop implied by split().
2327
2328However, long experience has shown that many programming tasks may
14218588
GS
2329be significantly simplified by using repeated subexpressions that
2330may match zero-length substrings. Here's a simple example being:
c84d73f1 2331
d1fbf752 2332 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2333 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2334
9da458fc 2335Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2336the infinite loop>. The rules for this are different for lower-level
527e91da 2337loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2338ones like the C</g> modifier or split() operator.
2339
19799a22
GS
2340The lower-level loops are I<interrupted> (that is, the loop is
2341broken) when Perl detects that a repeated expression matched a
2342zero-length substring. Thus
c84d73f1
IZ
2343
2344 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2345
5d458dd8 2346is made equivalent to
c84d73f1 2347
0b928c2f
FC
2348 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2349
2350For example, this program
2351
2352 #!perl -l
2353 "aaaaab" =~ /
2354 (?:
2355 a # non-zero
2356 | # or
2357 (?{print "hello"}) # print hello whenever this
2358 # branch is tried
2359 (?=(b)) # zero-width assertion
2360 )* # any number of times
2361 /x;
2362 print $&;
2363 print $1;
c84d73f1 2364
0b928c2f
FC
2365prints
2366
2367 hello
2368 aaaaa
2369 b
2370
2371Notice that "hello" is only printed once, as when Perl sees that the sixth
2372iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2373the C<*>.
2374
2375The higher-level loops preserve an additional state between iterations:
5d458dd8 2376whether the last match was zero-length. To break the loop, the following
c84d73f1 2377match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2378This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2379and so the I<second best> match is chosen if the I<best> match is of
2380zero length.
2381
19799a22 2382For example:
c84d73f1
IZ
2383
2384 $_ = 'bar';
2385 s/\w??/<$&>/g;
2386
20fb949f 2387results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2388match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2389best> match is what is matched by C<\w>. Thus zero-length matches
2390alternate with one-character-long matches.
2391
5d458dd8 2392Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2393position one notch further in the string.
2394
19799a22 2395The additional state of being I<matched with zero-length> is associated with
c84d73f1 2396the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2397Zero-length matches at the end of the previous match are ignored
2398during C<split>.
c84d73f1 2399
0d017f4d 2400=head2 Combining RE Pieces
35a734be
IZ
2401
2402Each of the elementary pieces of regular expressions which were described
2403before (such as C<ab> or C<\Z>) could match at most one substring
2404at the given position of the input string. However, in a typical regular
2405expression these elementary pieces are combined into more complicated
0b928c2f 2406patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2407(in these examples C<S> and C<T> are regular subexpressions).
2408
2409Such combinations can include alternatives, leading to a problem of choice:
2410if we match a regular expression C<a|ab> against C<"abc">, will it match
2411substring C<"a"> or C<"ab">? One way to describe which substring is
2412actually matched is the concept of backtracking (see L<"Backtracking">).
2413However, this description is too low-level and makes you think
2414in terms of a particular implementation.
2415
2416Another description starts with notions of "better"/"worse". All the
2417substrings which may be matched by the given regular expression can be
2418sorted from the "best" match to the "worst" match, and it is the "best"
2419match which is chosen. This substitutes the question of "what is chosen?"
2420by the question of "which matches are better, and which are worse?".
2421
2422Again, for elementary pieces there is no such question, since at most
2423one match at a given position is possible. This section describes the
2424notion of better/worse for combining operators. In the description
2425below C<S> and C<T> are regular subexpressions.
2426
13a2d996 2427=over 4
35a734be
IZ
2428
2429=item C<ST>
2430
2431Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2432substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2433which can be matched by C<T>.
35a734be 2434
0b928c2f 2435If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2436match than C<A'B'>.
2437
2438If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2439C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2440
2441=item C<S|T>
2442
2443When C<S> can match, it is a better match than when only C<T> can match.
2444
2445Ordering of two matches for C<S> is the same as for C<S>. Similar for
2446two matches for C<T>.
2447
2448=item C<S{REPEAT_COUNT}>
2449
2450Matches as C<SSS...S> (repeated as many times as necessary).
2451
2452=item C<S{min,max}>
2453
2454Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2455
2456=item C<S{min,max}?>
2457
2458Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2459
2460=item C<S?>, C<S*>, C<S+>
2461
2462Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
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
c47ff5f1 2468=item C<< (?>S) >>
35a734be
IZ
2469
2470Matches the best match for C<S> and only that.
2471
2472=item C<(?=S)>, C<(?<=S)>
2473
2474Only the best match for C<S> is considered. (This is important only if
2475C<S> has capturing parentheses, and backreferences are used somewhere
2476else in the whole regular expression.)
2477
2478=item C<(?!S)>, C<(?<!S)>
2479
2480For this grouping operator there is no need to describe the ordering, since
2481only whether or not C<S> can match is important.
2482
6bda09f9 2483=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2484
2485The ordering is the same as for the regular expression which is
c27a5cfe 2486the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
2487
2488=item C<(?(condition)yes-pattern|no-pattern)>
2489
2490Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2491already determined. The ordering of the matches is the same as for the
2492chosen subexpression.
2493
2494=back
2495
2496The above recipes describe the ordering of matches I<at a given position>.
2497One more rule is needed to understand how a match is determined for the
2498whole regular expression: a match at an earlier position is always better
2499than a match at a later position.
2500
0d017f4d 2501=head2 Creating Custom RE Engines
c84d73f1 2502
0b928c2f
FC
2503As of Perl 5.10.0, one can create custom regular expression engines. This
2504is not for the faint of heart, as they have to plug in at the C level. See
2505L<perlreapi> for more details.
2506
2507As an alternative, overloaded constants (see L<overload>) provide a simple
2508way to extend the functionality of the RE engine, by substituting one
2509pattern for another.
c84d73f1
IZ
2510
2511Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2512matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2513characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2514at these positions, so we want to have each C<\Y|> in the place of the
2515more complicated version. We can create a module C<customre> to do
2516this:
2517
2518 package customre;
2519 use overload;
2520
2521 sub import {
2522 shift;
2523 die "No argument to customre::import allowed" if @_;
2524 overload::constant 'qr' => \&convert;
2525 }
2526
2527 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2528
580a9fe1
RGS
2529 # We must also take care of not escaping the legitimate \\Y|
2530 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2531 my %rules = ( '\\' => '\\\\',
f793d64a 2532 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2533 sub convert {
2534 my $re = shift;
5d458dd8 2535 $re =~ s{
c84d73f1
IZ
2536 \\ ( \\ | Y . )
2537 }
5d458dd8 2538 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2539 return $re;
2540 }
2541
2542Now C<use customre> enables the new escape in constant regular
2543expressions, i.e., those without any runtime variable interpolations.
2544As documented in L<overload>, this conversion will work only over
2545literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2546part of this regular expression needs to be converted explicitly
2547(but only if the special meaning of C<\Y|> should be enabled inside $re):
2548
2549 use customre;
2550 $re = <>;
2551 chomp $re;
2552 $re = customre::convert $re;
2553 /\Y|$re\Y|/;
2554
0b928c2f 2555=head2 PCRE/Python Support
1f1031fe 2556
0b928c2f 2557As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2558to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2559Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2560
2561=over 4
2562
ae5648b3 2563=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2564
c27a5cfe 2565Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2566
2567=item C<< (?P=NAME) >>
2568
c27a5cfe 2569Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2570
2571=item C<< (?P>NAME) >>
2572
c27a5cfe 2573Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2574
ee9b8eae 2575=back
1f1031fe 2576
19799a22
GS
2577=head1 BUGS
2578
88c9975e
KW
2579Many regular expression constructs don't work on EBCDIC platforms.
2580
ed7efc79
KW
2581There are a number of issues with regard to case-insensitive matching
2582in Unicode rules. See C<i> under L</Modifiers> above.
2583
9da458fc
IZ
2584This document varies from difficult to understand to completely
2585and utterly opaque. The wandering prose riddled with jargon is
2586hard to fathom in several places.
2587
2588This document needs a rewrite that separates the tutorial content
2589from the reference content.
19799a22
GS
2590
2591=head1 SEE ALSO
9fa51da4 2592
91e0c79e
MJD
2593L<perlrequick>.
2594
2595L<perlretut>.
2596
9b599b2a
GS
2597L<perlop/"Regexp Quote-Like Operators">.
2598
1e66bd83
PP
2599L<perlop/"Gory details of parsing quoted constructs">.
2600
14218588
GS
2601L<perlfaq6>.
2602
9b599b2a
GS
2603L<perlfunc/pos>.
2604
2605L<perllocale>.
2606
fb55449c
JH
2607L<perlebcdic>.
2608
14218588
GS
2609I<Mastering Regular Expressions> by Jeffrey Friedl, published
2610by O'Reilly and Associates.