This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
pod/perlre: Italicize text to indicate non-literal
[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
ed7efc79 275This modifier may be specified to be the default by C<use feature
66cbab2c
KW
276'unicode_strings>, C<use locale ':not_characters'>, or
277C<L<use 5.012|perlfunc/use VERSION>> (or higher),
808432af 278but see L</Which character set modifier is in effect?>.
b6fa137b
FC
279X</u>
280
ed7efc79
KW
281=head4 /d
282
283This modifier means to use the "Default" native rules of the platform
284except when there is cause to use Unicode rules instead, as follows:
285
286=over 4
287
288=item 1
289
290the target string is encoded in UTF-8; or
291
292=item 2
293
294the pattern is encoded in UTF-8; or
295
296=item 3
297
298the pattern explicitly mentions a code point that is above 255 (say by
299C<\x{100}>); or
300
301=item 4
b6fa137b 302
ed7efc79
KW
303the pattern uses a Unicode name (C<\N{...}>); or
304
305=item 5
306
9d1a5160
KW
307the pattern uses a Unicode property (C<\p{...}>); or
308
309=item 6
310
311the pattern uses L</C<(?[ ])>>
ed7efc79
KW
312
313=back
314
315Another mnemonic for this modifier is "Depends", as the rules actually
316used depend on various things, and as a result you can get unexpected
808432af
KW
317results. See L<perlunicode/The "Unicode Bug">. The Unicode Bug has
318become rather infamous, leading to yet another (printable) name for this
319modifier, "Dodgy".
ed7efc79 320
4b9734bf
KW
321Unless the pattern or string are encoded in UTF-8, only ASCII characters
322can match positively.
ed7efc79
KW
323
324Here are some examples of how that works on an ASCII platform:
325
326 $str = "\xDF"; # $str is not in UTF-8 format.
327 $str =~ /^\w/; # No match, as $str isn't in UTF-8 format.
328 $str .= "\x{0e0b}"; # Now $str is in UTF-8 format.
329 $str =~ /^\w/; # Match! $str is now in UTF-8 format.
330 chop $str;
331 $str =~ /^\w/; # Still a match! $str remains in UTF-8 format.
332
808432af
KW
333This modifier is automatically selected by default when none of the
334others are, so yet another name for it is "Default".
335
336Because of the unexpected behaviors associated with this modifier, you
337probably should only use it to maintain weird backward compatibilities.
338
339=head4 /a (and /aa)
340
341This modifier stands for ASCII-restrict (or ASCII-safe). This modifier,
342unlike the others, may be doubled-up to increase its effect.
343
344When it appears singly, it causes the sequences C<\d>, C<\s>, C<\w>, and
345the Posix character classes to match only in the ASCII range. They thus
346revert to their pre-5.6, pre-Unicode meanings. Under C</a>, C<\d>
347always means precisely the digits C<"0"> to C<"9">; C<\s> means the five
d28d8023
KW
348characters C<[ \f\n\r\t]>, and starting in Perl v5.18, experimentally,
349the vertical tab; C<\w> means the 63 characters
808432af
KW
350C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
351C<[[:print:]]> match only the appropriate ASCII-range characters.
352
353This modifier is useful for people who only incidentally use Unicode,
354and who do not wish to be burdened with its complexities and security
355concerns.
356
357With C</a>, one can write C<\d> with confidence that it will only match
358ASCII characters, and should the need arise to match beyond ASCII, you
359can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>). There are
360similar C<\p{...}> constructs that can match beyond ASCII both white
361space (see L<perlrecharclass/Whitespace>), and Posix classes (see
362L<perlrecharclass/POSIX Character Classes>). Thus, this modifier
363doesn't mean you can't use Unicode, it means that to get Unicode
364matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
365signals Unicode.
366
367As you would expect, this modifier causes, for example, C<\D> to mean
368the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
369C<\D>, C<\S>, and C<\W>. C<\b> still means to match at the boundary
370between C<\w> and C<\W>, using the C</a> definitions of them (similarly
371for C<\B>).
372
373Otherwise, C</a> behaves like the C</u> modifier, in that
374case-insensitive matching uses Unicode semantics; for example, "k" will
375match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
376points in the Latin1 range, above ASCII will have Unicode rules when it
377comes to case-insensitive matching.
378
379To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
380specify the "a" twice, for example C</aai> or C</aia>. (The first
381occurrence of "a" restricts the C<\d>, etc., and the second occurrence
382adds the C</i> restrictions.) But, note that code points outside the
383ASCII range will use Unicode rules for C</i> matching, so the modifier
384doesn't really restrict things to just ASCII; it just forbids the
385intermixing of ASCII and non-ASCII.
386
387To summarize, this modifier provides protection for applications that
388don't wish to be exposed to all of Unicode. Specifying it twice
389gives added protection.
390
391This modifier may be specified to be the default by C<use re '/a'>
392or C<use re '/aa'>. If you do so, you may actually have occasion to use
393the C</u> modifier explictly if there are a few regular expressions
394where you do want full Unicode rules (but even here, it's best if
395everything were under feature C<"unicode_strings">, along with the
396C<use re '/aa'>). Also see L</Which character set modifier is in
397effect?>.
398X</a>
399X</aa>
400
ed7efc79
KW
401=head4 Which character set modifier is in effect?
402
403Which of these modifiers is in effect at any given point in a regular
808432af
KW
404expression depends on a fairly complex set of interactions. These have
405been designed so that in general you don't have to worry about it, but
406this section gives the gory details. As
ed7efc79
KW
407explained below in L</Extended Patterns> it is possible to explicitly
408specify modifiers that apply only to portions of a regular expression.
409The innermost always has priority over any outer ones, and one applying
6368643f
KW
410to the whole expression has priority over any of the default settings that are
411described in the remainder of this section.
ed7efc79 412
916cec3f 413The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
ed7efc79
KW
414default modifiers (including these) for regular expressions compiled
415within its scope. This pragma has precedence over the other pragmas
516074bb 416listed below that also change the defaults.
ed7efc79
KW
417
418Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
66cbab2c 419and C<L<use feature 'unicode_strings|feature>>, or
ed7efc79
KW
420C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
421C</u> when not in the same scope as either C<L<use locale|perllocale>>
66cbab2c
KW
422or C<L<use bytes|bytes>>.
423(C<L<use locale ':not_characters'|perllocale/Unicode and UTF-8>> also
424sets the default to C</u>, overriding any plain C<use locale>.)
425Unlike the mechanisms mentioned above, these
6368643f
KW
426affect operations besides regular expressions pattern matching, and so
427give more consistent results with other operators, including using
428C<\U>, C<\l>, etc. in substitution replacements.
ed7efc79
KW
429
430If none of the above apply, for backwards compatibility reasons, the
431C</d> modifier is the one in effect by default. As this can lead to
432unexpected results, it is best to specify which other rule set should be
433used.
434
435=head4 Character set modifier behavior prior to Perl 5.14
436
437Prior to 5.14, there were no explicit modifiers, but C</l> was implied
438for regexes compiled within the scope of C<use locale>, and C</d> was
439implied otherwise. However, interpolating a regex into a larger regex
440would ignore the original compilation in favor of whatever was in effect
441at the time of the second compilation. There were a number of
442inconsistencies (bugs) with the C</d> modifier, where Unicode rules
443would be used when inappropriate, and vice versa. C<\p{}> did not imply
444Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
b6fa137b 445
a0d0e21e
LW
446=head2 Regular Expressions
447
04838cea
RGS
448=head3 Metacharacters
449
384f06ae 450The patterns used in Perl pattern matching evolved from those supplied in
14218588 451the Version 8 regex routines. (The routines are derived
19799a22
GS
452(distantly) from Henry Spencer's freely redistributable reimplementation
453of the V8 routines.) See L<Version 8 Regular Expressions> for
454details.
a0d0e21e
LW
455
456In particular the following metacharacters have their standard I<egrep>-ish
457meanings:
d74e8afc
ITB
458X<metacharacter>
459X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
460
a0d0e21e 461
f793d64a
KW
462 \ Quote the next metacharacter
463 ^ Match the beginning of the line
464 . Match any character (except newline)
465 $ Match the end of the line (or before newline at the end)
466 | Alternation
467 () Grouping
468 [] Bracketed Character class
a0d0e21e 469
14218588
GS
470By default, the "^" character is guaranteed to match only the
471beginning of the string, the "$" character only the end (or before the
472newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
473assumption that the string contains only one line. Embedded newlines
474will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 475string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
476newline within the string (except if the newline is the last character in
477the string), and "$" will match before any newline. At the
a0d0e21e
LW
478cost of a little more overhead, you can do this by using the /m modifier
479on the pattern match operator. (Older programs did this by setting C<$*>,
db7cd43a 480but this option was removed in perl 5.10.)
d74e8afc 481X<^> X<$> X</m>
a0d0e21e 482
14218588 483To simplify multi-line substitutions, the "." character never matches a
55497cff 484newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 485the string is a single line--even if it isn't.
d74e8afc 486X<.> X</s>
a0d0e21e 487
04838cea
RGS
488=head3 Quantifiers
489
a0d0e21e 490The following standard quantifiers are recognized:
d74e8afc 491X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 492
f793d64a
KW
493 * Match 0 or more times
494 + Match 1 or more times
495 ? Match 1 or 0 times
496 {n} Match exactly n times
497 {n,} Match at least n times
498 {n,m} Match at least n but not more than m times
a0d0e21e 499
0b928c2f 500(If a curly bracket occurs in any other context and does not form part of
4d68ffa0
KW
501a backslashed sequence like C<\x{...}>, it is treated as a regular
502character. In particular, the lower quantifier bound is not optional,
503and a typo in a quantifier silently causes it to be treated as the
504literal characters. For example,
9af81bfe
KW
505
506 /o{4,3}/
507
508looks like a quantifier that matches 0 times, since 4 is greater than 3,
509but it really means to match the sequence of six characters
4d68ffa0
KW
510S<C<"o { 4 , 3 }">>. It is planned to eventually require literal uses
511of curly brackets to be escaped, say by preceding them with a backslash
512or enclosing them within square brackets, (C<"\{"> or C<"[{]">). This
513change will allow for future syntax extensions (like making the lower
514bound of a quantifier optional), and better error checking. In the
515meantime, you should get in the habit of escaping all instances where
516you mean a literal "{".)
9af81bfe
KW
517
518The "*" quantifier is equivalent to C<{0,}>, the "+"
527e91da 519quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 520to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
521This is usually 32766 on the most common platforms. The actual limit can
522be seen in the error message generated by code such as this:
523
820475bd 524 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 525
54310121
PP
526By default, a quantified subpattern is "greedy", that is, it will match as
527many times as possible (given a particular starting location) while still
528allowing the rest of the pattern to match. If you want it to match the
529minimum number of times possible, follow the quantifier with a "?". Note
530that the meanings don't change, just the "greediness":
0d017f4d 531X<metacharacter> X<greedy> X<greediness>
d74e8afc 532X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 533
f793d64a
KW
534 *? Match 0 or more times, not greedily
535 +? Match 1 or more times, not greedily
536 ?? Match 0 or 1 time, not greedily
0b928c2f 537 {n}? Match exactly n times, not greedily (redundant)
f793d64a
KW
538 {n,}? Match at least n times, not greedily
539 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 540
b9b4dddf
YO
541By default, when a quantified subpattern does not allow the rest of the
542overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 543sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
544as well.
545
f793d64a
KW
546 *+ Match 0 or more times and give nothing back
547 ++ Match 1 or more times and give nothing back
548 ?+ Match 0 or 1 time and give nothing back
549 {n}+ Match exactly n times and give nothing back (redundant)
550 {n,}+ Match at least n times and give nothing back
551 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
552
553For instance,
554
555 'aaaa' =~ /a++a/
556
557will never match, as the C<a++> will gobble up all the C<a>'s in the
558string and won't leave any for the remaining part of the pattern. This
559feature can be extremely useful to give perl hints about where it
560shouldn't backtrack. For instance, the typical "match a double-quoted
561string" problem can be most efficiently performed when written as:
562
563 /"(?:[^"\\]++|\\.)*+"/
564
0d017f4d 565as we know that if the final quote does not match, backtracking will not
0b928c2f
FC
566help. See the independent subexpression
567L</C<< (?>pattern) >>> for more details;
b9b4dddf
YO
568possessive quantifiers are just syntactic sugar for that construct. For
569instance the above example could also be written as follows:
570
571 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
572
04838cea
RGS
573=head3 Escape sequences
574
0b928c2f 575Because patterns are processed as double-quoted strings, the following
a0d0e21e
LW
576also work:
577
f793d64a
KW
578 \t tab (HT, TAB)
579 \n newline (LF, NL)
580 \r return (CR)
581 \f form feed (FF)
582 \a alarm (bell) (BEL)
583 \e escape (think troff) (ESC)
f793d64a 584 \cK control char (example: VT)
dc0d9c48 585 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 586 \N{name} named Unicode character or character sequence
f793d64a 587 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 588 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
589 \l lowercase next char (think vi)
590 \u uppercase next char (think vi)
591 \L lowercase till \E (think vi)
592 \U uppercase till \E (think vi)
593 \Q quote (disable) pattern metacharacters till \E
594 \E end either case modification or quoted section, think vi
a0d0e21e 595
9bb1f947 596Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 597
e1d1eefb 598=head3 Character Classes and other Special Escapes
04838cea 599
a0d0e21e 600In addition, Perl defines the following:
d0b16107 601X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 602
f793d64a
KW
603 Sequence Note Description
604 [...] [1] Match a character according to the rules of the
605 bracketed character class defined by the "...".
606 Example: [a-z] matches "a" or "b" or "c" ... or "z"
607 [[:...:]] [2] Match a character according to the rules of the POSIX
608 character class "..." within the outer bracketed
609 character class. Example: [[:upper:]] matches any
610 uppercase character.
572224ce 611 (?[...]) [8] Extended bracketed character class
d35dd6c6
KW
612 \w [3] Match a "word" character (alphanumeric plus "_", plus
613 other connector punctuation chars plus Unicode
0b928c2f 614 marks)
f793d64a
KW
615 \W [3] Match a non-"word" character
616 \s [3] Match a whitespace character
617 \S [3] Match a non-whitespace character
618 \d [3] Match a decimal digit character
619 \D [3] Match a non-digit character
620 \pP [3] Match P, named property. Use \p{Prop} for longer names
621 \PP [3] Match non-P
622 \X [4] Match Unicode "eXtended grapheme cluster"
623 \C Match a single C-language char (octet) even if that is
624 part of a larger UTF-8 character. Thus it breaks up
625 characters into their UTF-8 bytes, so you may end up
626 with malformed pieces of UTF-8. Unsupported in
627 lookbehind.
c27a5cfe 628 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
629 '1' may actually be any positive integer.
630 \g1 [5] Backreference to a specific or previous group,
631 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 632 previous group and may optionally be wrapped in
f793d64a
KW
633 curly brackets for safer parsing.
634 \g{name} [5] Named backreference
635 \k<name> [5] Named backreference
636 \K [6] Keep the stuff left of the \K, don't include it in $&
2171640d 637 \N [7] Any character but \n. Not affected by /s modifier
f793d64a
KW
638 \v [3] Vertical whitespace
639 \V [3] Not vertical whitespace
640 \h [3] Horizontal whitespace
641 \H [3] Not horizontal whitespace
642 \R [4] Linebreak
e1d1eefb 643
9bb1f947
KW
644=over 4
645
646=item [1]
647
648See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 649
9bb1f947 650=item [2]
b8c5462f 651
9bb1f947 652See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 653
9bb1f947 654=item [3]
5496314a 655
9bb1f947 656See L<perlrecharclass/Backslash sequences> for details.
5496314a 657
9bb1f947 658=item [4]
5496314a 659
9bb1f947 660See L<perlrebackslash/Misc> for details.
d0b16107 661
9bb1f947 662=item [5]
b8c5462f 663
c27a5cfe 664See L</Capture groups> below for details.
93733859 665
9bb1f947 666=item [6]
b8c5462f 667
9bb1f947
KW
668See L</Extended Patterns> below for details.
669
670=item [7]
671
672Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
673character or character sequence whose name is C<NAME>; and similarly
674when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
675code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947 676
572224ce
KW
677=item [8]
678
679See L<perlrecharclass/Extended Bracketed Character Classes> for details.
680
9bb1f947 681=back
d0b16107 682
04838cea
RGS
683=head3 Assertions
684
a0d0e21e 685Perl defines the following zero-width assertions:
d74e8afc
ITB
686X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
687X<regexp, zero-width assertion>
688X<regular expression, zero-width assertion>
689X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 690
9bb1f947
KW
691 \b Match a word boundary
692 \B Match except at a word boundary
693 \A Match only at beginning of string
694 \Z Match only at end of string, or before newline at the end
695 \z Match only at end of string
696 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 697 of prior m//g)
a0d0e21e 698
14218588 699A word boundary (C<\b>) is a spot between two characters
19799a22
GS
700that has a C<\w> on one side of it and a C<\W> on the other side
701of it (in either order), counting the imaginary characters off the
702beginning and end of the string as matching a C<\W>. (Within
703character classes C<\b> represents backspace rather than a word
704boundary, just as it normally does in any double-quoted string.)
705The C<\A> and C<\Z> are just like "^" and "$", except that they
706won't match multiple times when the C</m> modifier is used, while
707"^" and "$" will match at every internal line boundary. To match
708the actual end of the string and not ignore an optional trailing
709newline, use C<\z>.
d74e8afc 710X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
711
712The C<\G> assertion can be used to chain global matches (using
713C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
714It is also useful when writing C<lex>-like scanners, when you have
715several patterns that you want to match against consequent substrings
0b928c2f 716of your string; see the previous reference. The actual location
19799a22 717where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d 718an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
0b928c2f
FC
719matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
720is modified somewhat, in that contents to the left of C<\G> are
58e23c8d
YO
721not counted when determining the length of the match. Thus the following
722will not match forever:
d74e8afc 723X<\G>
c47ff5f1 724
e761bb84
CO
725 my $string = 'ABC';
726 pos($string) = 1;
727 while ($string =~ /(.\G)/g) {
728 print $1;
729 }
58e23c8d
YO
730
731It will print 'A' and then terminate, as it considers the match to
732be zero-width, and thus will not match at the same position twice in a
733row.
734
735It is worth noting that C<\G> improperly used can result in an infinite
736loop. Take care when using patterns that include C<\G> in an alternation.
737
c27a5cfe 738=head3 Capture groups
04838cea 739
c27a5cfe
KW
740The bracketing construct C<( ... )> creates capture groups (also referred to as
741capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
742the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
743for the second, and so on.
744This is called a I<backreference>.
d74e8afc 745X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 746X<regex, capture group> X<regexp, capture group>
d74e8afc 747X<regular expression, capture buffer> X<backreference>
c27a5cfe 748X<regular expression, capture group> X<backreference>
1f1031fe 749X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
750X<named capture buffer> X<regular expression, named capture buffer>
751X<named capture group> X<regular expression, named capture group>
752X<%+> X<$+{name}> X<< \k<name> >>
753There is no limit to the number of captured substrings that you may use.
754Groups are numbered with the leftmost open parenthesis being number 1, etc. If
755a group did not match, the associated backreference won't match either. (This
756can happen if the group is optional, or in a different branch of an
757alternation.)
758You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
759this form, described below.
760
761You can also refer to capture groups relatively, by using a negative number, so
762that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
763group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
764example:
5624f11d
YO
765
766 /
c27a5cfe
KW
767 (Y) # group 1
768 ( # group 2
769 (X) # group 3
770 \g{-1} # backref to group 3
771 \g{-3} # backref to group 1
5624f11d
YO
772 )
773 /x
774
d8b950dc
KW
775would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
776interpolate regexes into larger regexes and not have to worry about the
777capture groups being renumbered.
778
779You can dispense with numbers altogether and create named capture groups.
780The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
781reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
782also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
783I<name> must not begin with a number, nor contain hyphens.
784When different groups within the same pattern have the same name, any reference
785to that name assumes the leftmost defined group. Named groups count in
786absolute and relative numbering, and so can also be referred to by those
787numbers.
788(It's possible to do things with named capture groups that would otherwise
789require C<(??{})>.)
790
791Capture group contents are dynamically scoped and available to you outside the
792pattern until the end of the enclosing block or until the next successful
793match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
794You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
795etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
796
797Braces are required in referring to named capture groups, but are optional for
798absolute or relative numbered ones. Braces are safer when creating a regex by
799concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
800contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
801is probably not what you intended.
802
803The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
804there were no named nor relative numbered capture groups. Absolute numbered
0b928c2f
FC
805groups were referred to using C<\1>,
806C<\2>, etc., and this notation is still
d8b950dc
KW
807accepted (and likely always will be). But it leads to some ambiguities if
808there are more than 9 capture groups, as C<\10> could mean either the tenth
809capture group, or the character whose ordinal in octal is 010 (a backspace in
810ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
811only if at least 10 left parentheses have opened before it. Likewise C<\11> is
812a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
813And so on. C<\1> through C<\9> are always interpreted as backreferences.
814There are several examples below that illustrate these perils. You can avoid
815the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
816and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
817digits padded with leading zeros, since a leading zero implies an octal
818constant.
d8b950dc
KW
819
820The C<\I<digit>> notation also works in certain circumstances outside
ed7efc79 821the pattern. See L</Warning on \1 Instead of $1> below for details.
81714fb9 822
14218588 823Examples:
a0d0e21e
LW
824
825 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
826
d8b950dc 827 /(.)\g1/ # find first doubled char
81714fb9
YO
828 and print "'$1' is the first doubled character\n";
829
830 /(?<char>.)\k<char>/ # ... a different way
831 and print "'$+{char}' is the first doubled character\n";
832
d8b950dc 833 /(?'char'.)\g1/ # ... mix and match
81714fb9 834 and print "'$1' is the first doubled character\n";
c47ff5f1 835
14218588 836 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
837 $hours = $1;
838 $minutes = $2;
839 $seconds = $3;
a0d0e21e 840 }
c47ff5f1 841
9d860678
KW
842 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
843 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
844 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
845 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
846
847 $a = '(.)\1'; # Creates problems when concatenated.
848 $b = '(.)\g{1}'; # Avoids the problems.
849 "aa" =~ /${a}/; # True
850 "aa" =~ /${b}/; # True
851 "aa0" =~ /${a}0/; # False!
852 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
853 "aa\x08" =~ /${a}0/; # True!
854 "aa\x08" =~ /${b}0/; # False
9d860678 855
14218588
GS
856Several special variables also refer back to portions of the previous
857match. C<$+> returns whatever the last bracket match matched.
858C<$&> returns the entire matched string. (At one point C<$0> did
859also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
860everything before the matched string. C<$'> returns everything
861after the matched string. And C<$^N> contains whatever was matched by
862the most-recently closed group (submatch). C<$^N> can be used in
863extended patterns (see below), for example to assign a submatch to a
81714fb9 864variable.
d74e8afc 865X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 866
d8b950dc
KW
867These special variables, like the C<%+> hash and the numbered match variables
868(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
869until the end of the enclosing block or until the next successful
870match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
871X<$+> X<$^N> X<$&> X<$`> X<$'>
872X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
873
0d017f4d 874B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 875which makes it easier to write code that tests for a series of more
665e98b9
JH
876specific cases and remembers the best match.
877
d78f32f6
FC
878B<WARNING>: If your code is to run on Perl 5.16 or earlier,
879beware that once Perl sees that you need one of C<$&>, C<$`>, or
14218588 880C<$'> anywhere in the program, it has to provide them for every
d78f32f6
FC
881pattern match. This may substantially slow your program. (In Perl 5.18 a
882more efficient mechanism is used, eliminating any slowdown.) Perl
d8b950dc 883uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
884price for each pattern that contains capturing parentheses. (To
885avoid this cost while retaining the grouping behaviour, use the
886extended regular expression C<(?: ... )> instead.) But if you never
887use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
888parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
889if you can, but if you can't (and some algorithms really appreciate
890them), once you've used them once, use them at will, because you've
d78f32f6 891already paid the price.
d74e8afc 892X<$&> X<$`> X<$'>
68dc0745 893
d78f32f6 894As a workaround for this problem, Perl 5.10.0 introduced C<${^PREMATCH}>,
cde0cee5
YO
895C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
896and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 897successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
898The use of these variables incurs no global performance penalty, unlike
899their punctuation char equivalents, however at the trade-off that you
d78f32f6
FC
900have to tell perl when you want to use them. As of Perl 5.18, these three
901variables are equivalent to C<$`>, C<$&> and C<$'>, and C</p> is ignored.
87e95b7f 902X</p> X<p modifier>
cde0cee5 903
9d727203
KW
904=head2 Quoting metacharacters
905
19799a22
GS
906Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
907C<\w>, C<\n>. Unlike some other regular expression languages, there
908are no backslashed symbols that aren't alphanumeric. So anything
0f264506 909that looks like \\, \(, \), \[, \], \{, or \} is always
19799a22
GS
910interpreted as a literal character, not a metacharacter. This was
911once used in a common idiom to disable or quote the special meanings
912of regular expression metacharacters in a string that you want to
36bbe248 913use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
914
915 $pattern =~ s/(\W)/\\$1/g;
916
f1cbbd6e 917(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
918Today it is more common to use the quotemeta() function or the C<\Q>
919metaquoting escape sequence to disable all metacharacters' special
920meanings like this:
a0d0e21e
LW
921
922 /$unquoted\Q$quoted\E$unquoted/
923
9da458fc
IZ
924Beware that if you put literal backslashes (those not inside
925interpolated variables) between C<\Q> and C<\E>, double-quotish
926backslash interpolation may lead to confusing results. If you
927I<need> to use literal backslashes within C<\Q...\E>,
928consult L<perlop/"Gory details of parsing quoted constructs">.
929
736fe711
KW
930C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
931
19799a22
GS
932=head2 Extended Patterns
933
14218588 934Perl also defines a consistent extension syntax for features not
0b928c2f
FC
935found in standard tools like B<awk> and
936B<lex>. The syntax for most of these is a
14218588
GS
937pair of parentheses with a question mark as the first thing within
938the parentheses. The character after the question mark indicates
939the extension.
19799a22 940
14218588
GS
941The stability of these extensions varies widely. Some have been
942part of the core language for many years. Others are experimental
943and may change without warning or be completely removed. Check
944the documentation on an individual feature to verify its current
945status.
19799a22 946
14218588
GS
947A question mark was chosen for this and for the minimal-matching
948construct because 1) question marks are rare in older regular
949expressions, and 2) whenever you see one, you should stop and
0b928c2f 950"question" exactly what is going on. That's psychology....
a0d0e21e 951
70ca8714 952=over 4
a0d0e21e 953
cc6b7395 954=item C<(?#text)>
d74e8afc 955X<(?#)>
a0d0e21e 956
14218588 957A comment. The text is ignored. If the C</x> modifier enables
19799a22 958whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
959the comment as soon as it sees a C<)>, so there is no way to put a literal
960C<)> in the comment.
a0d0e21e 961
cfaf538b 962=item C<(?adlupimsx-imsx)>
fb85c044 963
cfaf538b 964=item C<(?^alupimsx)>
fb85c044 965X<(?)> X<(?^)>
19799a22 966
0b6d1084
JH
967One or more embedded pattern-match modifiers, to be turned on (or
968turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
969the remainder of the enclosing pattern group (if any).
970
fb85c044 971This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 972configuration file, taken from an argument, or specified in a table
0b928c2f
FC
973somewhere. Consider the case where some patterns want to be
974case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 975include C<(?i)> at the front of the pattern. For example:
19799a22
GS
976
977 $pattern = "foobar";
5d458dd8 978 if ( /$pattern/i ) { }
19799a22
GS
979
980 # more flexible:
981
982 $pattern = "(?i)foobar";
5d458dd8 983 if ( /$pattern/ ) { }
19799a22 984
0b6d1084 985These modifiers are restored at the end of the enclosing group. For example,
19799a22 986
d8b950dc 987 ( (?i) blah ) \s+ \g1
19799a22 988
0d017f4d
WL
989will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
990repetition of the previous word, assuming the C</x> modifier, and no C</i>
991modifier outside this group.
19799a22 992
8eb5594e 993These modifiers do not carry over into named subpatterns called in the
dd72e27b 994enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
8eb5594e
DR
995change the case-sensitivity of the "NAME" pattern.
996
dc925305
KW
997Any of these modifiers can be set to apply globally to all regular
998expressions compiled within the scope of a C<use re>. See
a0bbd6ff 999L<re/"'/flags' mode">.
dc925305 1000
9de15fec
KW
1001Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1002after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
1003C<"d">) may follow the caret to override it.
1004But a minus sign is not legal with it.
1005
dc925305 1006Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 1007that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 1008C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
1009others, and a maximum of one (or two C<a>'s) may appear in the
1010construct. Thus, for
0b928c2f 1011example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 1012C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1013
1014Note also that the C<p> modifier is special in that its presence
1015anywhere in a pattern has a global effect.
cde0cee5 1016
5a964f20 1017=item C<(?:pattern)>
d74e8afc 1018X<(?:)>
a0d0e21e 1019
cfaf538b 1020=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1021
cfaf538b 1022=item C<(?^aluimsx:pattern)>
fb85c044
KW
1023X<(?^:)>
1024
5a964f20
TC
1025This is for clustering, not capturing; it groups subexpressions like
1026"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1027
5a964f20 1028 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1029
1030is like
1031
5a964f20 1032 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1033
19799a22
GS
1034but doesn't spit out extra fields. It's also cheaper not to capture
1035characters if you don't need to.
a0d0e21e 1036
19799a22 1037Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1038C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1039
1040 /(?s-i:more.*than).*million/i
1041
14218588 1042is equivalent to the more verbose
ca9dfc88
IZ
1043
1044 /(?:(?s-i)more.*than).*million/i
1045
fb85c044 1046Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1047after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1048flags (except C<"d">) may follow the caret, so
fb85c044
KW
1049
1050 (?^x:foo)
1051
1052is equivalent to
1053
1054 (?x-ims:foo)
1055
1056The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1057surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1058modified by any flags specified.
1059
1060The caret allows for simpler stringification of compiled regular
1061expressions. These look like
1062
1063 (?^:pattern)
1064
1065with any non-default flags appearing between the caret and the colon.
1066A test that looks at such stringification thus doesn't need to have the
1067system default flags hard-coded in it, just the caret. If new flags are
1068added to Perl, the meaning of the caret's expansion will change to include
1069the default for those flags, so the test will still work, unchanged.
1070
1071Specifying a negative flag after the caret is an error, as the flag is
1072redundant.
1073
1074Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1075to match at the beginning.
1076
594d7033
YO
1077=item C<(?|pattern)>
1078X<(?|)> X<Branch reset>
1079
1080This is the "branch reset" pattern, which has the special property
c27a5cfe 1081that the capture groups are numbered from the same starting point
99d59c4d 1082in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1083
c27a5cfe 1084Capture groups are numbered from left to right, but inside this
693596a8 1085construct the numbering is restarted for each branch.
4deaaa80 1086
c27a5cfe 1087The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1088following this construct will be numbered as though the construct
1089contained only one branch, that being the one with the most capture
c27a5cfe 1090groups in it.
4deaaa80 1091
0b928c2f 1092This construct is useful when you want to capture one of a
4deaaa80
PJ
1093number of alternative matches.
1094
1095Consider the following pattern. The numbers underneath show in
c27a5cfe 1096which group the captured content will be stored.
594d7033
YO
1097
1098
1099 # before ---------------branch-reset----------- after
1100 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1101 # 1 2 2 3 2 3 4
1102
ab106183
A
1103Be careful when using the branch reset pattern in combination with
1104named captures. Named captures are implemented as being aliases to
c27a5cfe 1105numbered groups holding the captures, and that interferes with the
ab106183
A
1106implementation of the branch reset pattern. If you are using named
1107captures in a branch reset pattern, it's best to use the same names,
1108in the same order, in each of the alternations:
1109
1110 /(?| (?<a> x ) (?<b> y )
1111 | (?<a> z ) (?<b> w )) /x
1112
1113Not doing so may lead to surprises:
1114
1115 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1116 say $+ {a}; # Prints '12'
1117 say $+ {b}; # *Also* prints '12'.
1118
c27a5cfe
KW
1119The problem here is that both the group named C<< a >> and the group
1120named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1121
ee9b8eae
YO
1122=item Look-Around Assertions
1123X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1124
0b928c2f 1125Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1126pattern without including it in C<$&>. Positive assertions match when
1127their subpattern matches, negative assertions match when their subpattern
1128fails. Look-behind matches text up to the current match position,
1129look-ahead matches text following the current match position.
1130
1131=over 4
1132
5a964f20 1133=item C<(?=pattern)>
d74e8afc 1134X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1135
19799a22 1136A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1137matches a word followed by a tab, without including the tab in C<$&>.
1138
5a964f20 1139=item C<(?!pattern)>
d74e8afc 1140X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1141
19799a22 1142A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1143matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1144however that look-ahead and look-behind are NOT the same thing. You cannot
1145use this for look-behind.
7b8d334a 1146
5a964f20 1147If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1148will not do what you want. That's because the C<(?!foo)> is just saying that
1149the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1150match. Use look-behind instead (see below).
c277df42 1151
ee9b8eae
YO
1152=item C<(?<=pattern)> C<\K>
1153X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1154
c47ff5f1 1155A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1156matches a word that follows a tab, without including the tab in C<$&>.
1157Works only for fixed-width look-behind.
c277df42 1158
ee9b8eae
YO
1159There is a special form of this construct, called C<\K>, which causes the
1160regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1161not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1162look-behind. The use of C<\K> inside of another look-around assertion
1163is allowed, but the behaviour is currently not well defined.
1164
c62285ac 1165For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1166equivalent C<< (?<=...) >> construct, and it is especially useful in
1167situations where you want to efficiently remove something following
1168something else in a string. For instance
1169
1170 s/(foo)bar/$1/g;
1171
1172can be rewritten as the much more efficient
1173
1174 s/foo\Kbar//g;
1175
5a964f20 1176=item C<(?<!pattern)>
d74e8afc 1177X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1178
19799a22
GS
1179A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1180matches any occurrence of "foo" that does not follow "bar". Works
1181only for fixed-width look-behind.
c277df42 1182
ee9b8eae
YO
1183=back
1184
81714fb9
YO
1185=item C<(?'NAME'pattern)>
1186
1187=item C<< (?<NAME>pattern) >>
1188X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1189
c27a5cfe 1190A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1191parentheses C<()> but for the additional fact that the group
1192can be referred to by name in various regular expression
1193constructs (like C<\g{NAME}>) and can be accessed by name
1194after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1195for more details on the C<%+> and C<%-> hashes.
81714fb9 1196
c27a5cfe
KW
1197If multiple distinct capture groups have the same name then the
1198$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1199
0d017f4d 1200The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1201
1202B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1203function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1204numbered sequentially regardless of being named or not. Thus in the
1205pattern
1206
1207 /(x)(?<foo>y)(z)/
1208
1209$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1210the opposite which is what a .NET regex hacker might expect.
1211
1f1031fe
YO
1212Currently NAME is restricted to simple identifiers only.
1213In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1214its Unicode extension (see L<utf8>),
1215though it isn't extended by the locale (see L<perllocale>).
81714fb9 1216
1f1031fe 1217B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1218with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1219may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1220support the use of single quotes as a delimiter for the name.
81714fb9 1221
1f1031fe
YO
1222=item C<< \k<NAME> >>
1223
1224=item C<< \k'NAME' >>
81714fb9
YO
1225
1226Named backreference. Similar to numeric backreferences, except that
1227the group is designated by name and not number. If multiple groups
1228have the same name then it refers to the leftmost defined group in
1229the current match.
1230
0d017f4d 1231It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1232earlier in the pattern.
1233
1234Both forms are equivalent.
1235
1f1031fe 1236B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1237with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1238may be used instead of C<< \k<NAME> >>.
1f1031fe 1239
cc6b7395 1240=item C<(?{ code })>
d74e8afc 1241X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1242
19799a22 1243B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1244experimental, and may be changed without notice. Code executed that
1245has side effects may not perform identically from version to version
e128ab2c
DM
1246due to the effect of future optimisations in the regex engine. The
1247implementation of this feature was radically overhauled for the 5.18.0
1248release, and its behaviour in earlier versions of perl was much buggier,
1249especially in relation to parsing, lexical vars, scoping, recursion and
1250reentrancy.
c277df42 1251
e128ab2c
DM
1252This zero-width assertion executes any embedded Perl code. It always
1253succeeds, and its return value is set as C<$^R>.
19799a22 1254
e128ab2c
DM
1255In literal patterns, the code is parsed at the same time as the
1256surrounding code. While within the pattern, control is passed temporarily
1257back to the perl parser, until the logically-balancing closing brace is
1258encountered. This is similar to the way that an array index expression in
1259a literal string is handled, for example
77ea4f6d 1260
e128ab2c
DM
1261 "abc$array[ 1 + f('[') + g()]def"
1262
1263In particular, braces do not need to be balanced:
1264
576fa024 1265 s/abc(?{ f('{'); })/def/
e128ab2c
DM
1266
1267Even in a pattern that is interpolated and compiled at run-time, literal
1268code blocks will be compiled once, at perl compile time; the following
1269prints "ABCD":
1270
1271 print "D";
1272 my $qr = qr/(?{ BEGIN { print "A" } })/;
1273 my $foo = "foo";
1274 /$foo$qr(?{ BEGIN { print "B" } })/;
1275 BEGIN { print "C" }
1276
1277In patterns where the text of the code is derived from run-time
1278information rather than appearing literally in a source code /pattern/,
1279the code is compiled at the same time that the pattern is compiled, and
5771dda0 1280for reasons of security, C<use re 'eval'> must be in scope. This is to
e128ab2c
DM
1281stop user-supplied patterns containing code snippets from being
1282executable.
1283
5771dda0 1284In situations where you need to enable this with C<use re 'eval'>, you should
e128ab2c
DM
1285also have taint checking enabled. Better yet, use the carefully
1286constrained evaluation within a Safe compartment. See L<perlsec> for
1287details about both these mechanisms.
1288
1289From the viewpoint of parsing, lexical variable scope and closures,
1290
1291 /AAA(?{ BBB })CCC/
1292
1293behaves approximately like
1294
1295 /AAA/ && do { BBB } && /CCC/
1296
1297Similarly,
1298
1299 qr/AAA(?{ BBB })CCC/
1300
1301behaves approximately like
77ea4f6d 1302
e128ab2c
DM
1303 sub { /AAA/ && do { BBB } && /CCC/ }
1304
1305In particular:
1306
1307 { my $i = 1; $r = qr/(?{ print $i })/ }
1308 my $i = 2;
1309 /$r/; # prints "1"
1310
1311Inside a C<(?{...})> block, C<$_> refers to the string the regular
754091cb 1312expression is matching against. You can also use C<pos()> to know what is
fa11829f 1313the current position of matching within this string.
754091cb 1314
e128ab2c
DM
1315The code block introduces a new scope from the perspective of lexical
1316variable declarations, but B<not> from the perspective of C<local> and
1317similar localizing behaviours. So later code blocks within the same
1318pattern will still see the values which were localized in earlier blocks.
1319These accumulated localizations are undone either at the end of a
1320successful match, or if the assertion is backtracked (compare
1321L<"Backtracking">). For example,
b9ac3b5b
GS
1322
1323 $_ = 'a' x 8;
5d458dd8 1324 m<
d1fbf752 1325 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1326 (
5d458dd8 1327 a
b9ac3b5b 1328 (?{
d1fbf752
KW
1329 local $cnt = $cnt + 1; # Update $cnt,
1330 # backtracking-safe.
b9ac3b5b 1331 })
5d458dd8 1332 )*
b9ac3b5b 1333 aaaa
d1fbf752
KW
1334 (?{ $res = $cnt }) # On success copy to
1335 # non-localized location.
b9ac3b5b
GS
1336 >x;
1337
e128ab2c
DM
1338will initially increment C<$cnt> up to 8; then during backtracking, its
1339value will be unwound back to 4, which is the value assigned to C<$res>.
1340At the end of the regex execution, $cnt will be wound back to its initial
1341value of 0.
1342
1343This assertion may be used as the condition in a
b9ac3b5b 1344
e128ab2c
DM
1345 (?(condition)yes-pattern|no-pattern)
1346
1347switch. If I<not> used in this way, the result of evaluation of C<code>
1348is put into the special variable C<$^R>. This happens immediately, so
1349C<$^R> can be used from other C<(?{ code })> assertions inside the same
1350regular expression.
b9ac3b5b 1351
19799a22
GS
1352The assignment to C<$^R> above is properly localized, so the old
1353value of C<$^R> is restored if the assertion is backtracked; compare
1354L<"Backtracking">.
b9ac3b5b 1355
e128ab2c
DM
1356Note that the special variable C<$^N> is particularly useful with code
1357blocks to capture the results of submatches in variables without having to
1358keep track of the number of nested parentheses. For example:
1359
1360 $_ = "The brown fox jumps over the lazy dog";
1361 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1362 print "color = $color, animal = $animal\n";
1363
8988a1bb 1364
14455d6c 1365=item C<(??{ code })>
d74e8afc
ITB
1366X<(??{})>
1367X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1368
19799a22 1369B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1370experimental, and may be changed without notice. Code executed that
1371has side effects may not perform identically from version to version
1372due to the effect of future optimisations in the regex engine.
0f5d15d6 1373
e128ab2c
DM
1374This is a "postponed" regular subexpression. It behaves in I<exactly> the
1375same way as a C<(?{ code })> code block as described above, except that
1376its return value, rather than being assigned to C<$^R>, is treated as a
1377pattern, compiled if it's a string (or used as-is if its a qr// object),
1378then matched as if it were inserted instead of this construct.
6bda09f9 1379
e128ab2c
DM
1380During the matching of this sub-pattern, it has its own set of
1381captures which are valid during the sub-match, but are discarded once
1382control returns to the main pattern. For example, the following matches,
1383with the inner pattern capturing "B" and matching "BB", while the outer
1384pattern captures "A";
1385
1386 my $inner = '(.)\1';
1387 "ABBA" =~ /^(.)(??{ $inner })\1/;
1388 print $1; # prints "A";
6bda09f9 1389
e128ab2c
DM
1390Note that this means that there is no way for the inner pattern to refer
1391to a capture group defined outside. (The code block itself can use C<$1>,
1392etc., to refer to the enclosing pattern's capture groups.) Thus, although
0f5d15d6 1393
e128ab2c
DM
1394 ('a' x 100)=~/(??{'(.)' x 100})/
1395
1396I<will> match, it will I<not> set $1 on exit.
19799a22
GS
1397
1398The following pattern matches a parenthesized group:
0f5d15d6 1399
d1fbf752
KW
1400 $re = qr{
1401 \(
1402 (?:
1403 (?> [^()]+ ) # Non-parens without backtracking
1404 |
1405 (??{ $re }) # Group with matching parens
1406 )*
1407 \)
1408 }x;
0f5d15d6 1409
93f313ef
KW
1410See also
1411L<C<(?I<PARNO>)>|/(?PARNO) (?-PARNO) (?+PARNO) (?R) (?0)>
1412for a different, more efficient way to accomplish
6bda09f9
YO
1413the same task.
1414
e128ab2c
DM
1415Executing a postponed regular expression 50 times without consuming any
1416input string will result in a fatal error. The maximum depth is compiled
1417into perl, so changing it requires a custom build.
6bda09f9 1418
93f313ef 1419=item C<(?I<PARNO>)> C<(?-I<PARNO>)> C<(?+I<PARNO>)> C<(?R)> C<(?0)>
542fa716 1420X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1421X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1422X<regex, relative recursion>
6bda09f9 1423
e128ab2c
DM
1424Similar to C<(??{ code })> except that it does not involve executing any
1425code or potentially compiling a returned pattern string; instead it treats
1426the part of the current pattern contained within a specified capture group
1427as an independent pattern that must match at the current position.
1428Capture groups contained by the pattern will have the value as determined
1429by the outermost recursion.
6bda09f9 1430
93f313ef 1431I<PARNO> is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1432the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1433the beginning of the whole pattern. C<(?0)> is an alternate syntax for
93f313ef 1434C<(?R)>. If I<PARNO> is preceded by a plus or minus sign then it is assumed
c27a5cfe 1435to be relative, with negative numbers indicating preceding capture groups
542fa716 1436and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1437declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1438Note that the counting for relative recursion differs from that of
c27a5cfe 1439relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1440included.
6bda09f9 1441
81714fb9 1442The following pattern matches a function foo() which may contain
f145b7e9 1443balanced parentheses as the argument.
6bda09f9 1444
d1fbf752 1445 $re = qr{ ( # paren group 1 (full function)
81714fb9 1446 foo
d1fbf752 1447 ( # paren group 2 (parens)
6bda09f9 1448 \(
d1fbf752 1449 ( # paren group 3 (contents of parens)
6bda09f9 1450 (?:
d1fbf752 1451 (?> [^()]+ ) # Non-parens without backtracking
6bda09f9 1452 |
d1fbf752 1453 (?2) # Recurse to start of paren group 2
6bda09f9
YO
1454 )*
1455 )
1456 \)
1457 )
1458 )
1459 }x;
1460
1461If the pattern was used as follows
1462
1463 'foo(bar(baz)+baz(bop))'=~/$re/
1464 and print "\$1 = $1\n",
1465 "\$2 = $2\n",
1466 "\$3 = $3\n";
1467
1468the output produced should be the following:
1469
1470 $1 = foo(bar(baz)+baz(bop))
1471 $2 = (bar(baz)+baz(bop))
81714fb9 1472 $3 = bar(baz)+baz(bop)
6bda09f9 1473
c27a5cfe 1474If there is no corresponding capture group defined, then it is a
61528107 1475fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1476string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1477into perl, so changing it requires a custom build.
1478
542fa716
YO
1479The following shows how using negative indexing can make it
1480easier to embed recursive patterns inside of a C<qr//> construct
1481for later use:
1482
1483 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
c77257ed 1484 if (/foo $parens \s+ \+ \s+ bar $parens/x) {
542fa716
YO
1485 # do something here...
1486 }
1487
81714fb9 1488B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1489PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1490a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1491as atomic. Also, modifiers are resolved at compile time, so constructs
1492like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1493be processed.
6bda09f9 1494
894be9b7
YO
1495=item C<(?&NAME)>
1496X<(?&NAME)>
1497
93f313ef 1498Recurse to a named subpattern. Identical to C<(?I<PARNO>)> except that the
0d017f4d 1499parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1500the same name, then it recurses to the leftmost.
1501
1502It is an error to refer to a name that is not declared somewhere in the
1503pattern.
1504
1f1031fe
YO
1505B<NOTE:> In order to make things easier for programmers with experience
1506with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1507may be used instead of C<< (?&NAME) >>.
1f1031fe 1508
e2e6a0f1
YO
1509=item C<(?(condition)yes-pattern|no-pattern)>
1510X<(?()>
286f584a 1511
e2e6a0f1 1512=item C<(?(condition)yes-pattern)>
286f584a 1513
41ef34de
ML
1514Conditional expression. Matches C<yes-pattern> if C<condition> yields
1515a true value, matches C<no-pattern> otherwise. A missing pattern always
1516matches.
1517
25e26d77 1518C<(condition)> should be one of: 1) an integer in
e2e6a0f1 1519parentheses (which is valid if the corresponding pair of parentheses
25e26d77 1520matched); 2) a look-ahead/look-behind/evaluate zero-width assertion; 3) a
c27a5cfe 1521name in angle brackets or single quotes (which is valid if a group
25e26d77 1522with the given name matched); or 4) the special symbol (R) (true when
e2e6a0f1
YO
1523evaluated inside of recursion or eval). Additionally the R may be
1524followed by a number, (which will be true when evaluated when recursing
1525inside of the appropriate group), or by C<&NAME>, in which case it will
1526be true only when evaluated during recursion in the named group.
1527
1528Here's a summary of the possible predicates:
1529
1530=over 4
1531
1532=item (1) (2) ...
1533
c27a5cfe 1534Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1535
1536=item (<NAME>) ('NAME')
1537
c27a5cfe 1538Checks if a group with the given name has matched something.
e2e6a0f1 1539
f01cd190
FC
1540=item (?=...) (?!...) (?<=...) (?<!...)
1541
1542Checks whether the pattern matches (or does not match, for the '!'
1543variants).
1544
e2e6a0f1
YO
1545=item (?{ CODE })
1546
f01cd190 1547Treats the return value of the code block as the condition.
e2e6a0f1
YO
1548
1549=item (R)
1550
1551Checks if the expression has been evaluated inside of recursion.
1552
1553=item (R1) (R2) ...
1554
1555Checks if the expression has been evaluated while executing directly
1556inside of the n-th capture group. This check is the regex equivalent of
1557
1558 if ((caller(0))[3] eq 'subname') { ... }
1559
1560In other words, it does not check the full recursion stack.
1561
1562=item (R&NAME)
1563
1564Similar to C<(R1)>, this predicate checks to see if we're executing
1565directly inside of the leftmost group with a given name (this is the same
1566logic used by C<(?&NAME)> to disambiguate). It does not check the full
1567stack, but only the name of the innermost active recursion.
1568
1569=item (DEFINE)
1570
1571In this case, the yes-pattern is never directly executed, and no
1572no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1573See below for details.
1574
1575=back
1576
1577For example:
1578
1579 m{ ( \( )?
1580 [^()]+
1581 (?(1) \) )
1582 }x
1583
1584matches a chunk of non-parentheses, possibly included in parentheses
1585themselves.
1586
0b928c2f
FC
1587A special form is the C<(DEFINE)> predicate, which never executes its
1588yes-pattern directly, and does not allow a no-pattern. This allows one to
1589define subpatterns which will be executed only by the recursion mechanism.
e2e6a0f1
YO
1590This way, you can define a set of regular expression rules that can be
1591bundled into any pattern you choose.
1592
1593It is recommended that for this usage you put the DEFINE block at the
1594end of the pattern, and that you name any subpatterns defined within it.
1595
1596Also, it's worth noting that patterns defined this way probably will
1597not be as efficient, as the optimiser is not very clever about
1598handling them.
1599
1600An example of how this might be used is as follows:
1601
2bf803e2 1602 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1603 (?(DEFINE)
2bf803e2
YO
1604 (?<NAME_PAT>....)
1605 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1606 )/x
1607
c27a5cfe
KW
1608Note that capture groups matched inside of recursion are not accessible
1609after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1610necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1611C<$+{NAME}> would be.
286f584a 1612
51a1303c
BF
1613Finally, keep in mind that subpatterns created inside a DEFINE block
1614count towards the absolute and relative number of captures, so this:
1615
1616 my @captures = "a" =~ /(.) # First capture
1617 (?(DEFINE)
1618 (?<EXAMPLE> 1 ) # Second capture
1619 )/x;
1620 say scalar @captures;
1621
1622Will output 2, not 1. This is particularly important if you intend to
1623compile the definitions with the C<qr//> operator, and later
1624interpolate them in another pattern.
1625
c47ff5f1 1626=item C<< (?>pattern) >>
6bda09f9 1627X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1628
19799a22
GS
1629An "independent" subexpression, one which matches the substring
1630that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1631position, and it matches I<nothing other than this substring>. This
19799a22
GS
1632construct is useful for optimizations of what would otherwise be
1633"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1634It may also be useful in places where the "grab all you can, and do not
1635give anything back" semantic is desirable.
19799a22 1636
c47ff5f1 1637For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1638(anchored at the beginning of string, as above) will match I<all>
1639characters C<a> at the beginning of string, leaving no C<a> for
1640C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1641since the match of the subgroup C<a*> is influenced by the following
1642group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1643C<a*ab> will match fewer characters than a standalone C<a*>, since
1644this makes the tail match.
1645
0b928c2f
FC
1646C<< (?>pattern) >> does not disable backtracking altogether once it has
1647matched. It is still possible to backtrack past the construct, but not
1648into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1649
c47ff5f1 1650An effect similar to C<< (?>pattern) >> may be achieved by writing
0b928c2f
FC
1651C<(?=(pattern))\g{-1}>. This matches the same substring as a standalone
1652C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
c47ff5f1 1653makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1654(The difference between these two constructs is that the second one
1655uses a capturing group, thus shifting ordinals of backreferences
1656in the rest of a regular expression.)
1657
1658Consider this pattern:
c277df42 1659
871b0233 1660 m{ \(
e2e6a0f1 1661 (
f793d64a 1662 [^()]+ # x+
e2e6a0f1 1663 |
871b0233
IZ
1664 \( [^()]* \)
1665 )+
e2e6a0f1 1666 \)
871b0233 1667 }x
5a964f20 1668
19799a22
GS
1669That will efficiently match a nonempty group with matching parentheses
1670two levels deep or less. However, if there is no such group, it
1671will take virtually forever on a long string. That's because there
1672are so many different ways to split a long string into several
1673substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1674to a subpattern of the above pattern. Consider how the pattern
1675above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1676seconds, but that each extra letter doubles this time. This
1677exponential performance will make it appear that your program has
14218588 1678hung. However, a tiny change to this pattern
5a964f20 1679
e2e6a0f1
YO
1680 m{ \(
1681 (
f793d64a 1682 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1683 |
871b0233
IZ
1684 \( [^()]* \)
1685 )+
e2e6a0f1 1686 \)
871b0233 1687 }x
c277df42 1688
c47ff5f1 1689which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1690this yourself would be a productive exercise), but finishes in a fourth
1691the time when used on a similar string with 1000000 C<a>s. Be aware,
0b928c2f
FC
1692however, that, when this construct is followed by a
1693quantifier, it currently triggers a warning message under
9f1b1f2d 1694the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1695C<"matches null string many times in regex">.
c277df42 1696
c47ff5f1 1697On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1698effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1699This was only 4 times slower on a string with 1000000 C<a>s.
1700
9da458fc
IZ
1701The "grab all you can, and do not give anything back" semantic is desirable
1702in many situations where on the first sight a simple C<()*> looks like
1703the correct solution. Suppose we parse text with comments being delimited
1704by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1705its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1706the comment delimiter, because it may "give up" some whitespace if
1707the remainder of the pattern can be made to match that way. The correct
1708answer is either one of these:
1709
1710 (?>#[ \t]*)
1711 #[ \t]*(?![ \t])
1712
1713For example, to grab non-empty comments into $1, one should use either
1714one of these:
1715
1716 / (?> \# [ \t]* ) ( .+ ) /x;
1717 / \# [ \t]* ( [^ \t] .* ) /x;
1718
1719Which one you pick depends on which of these expressions better reflects
1720the above specification of comments.
1721
6bda09f9
YO
1722In some literature this construct is called "atomic matching" or
1723"possessive matching".
1724
b9b4dddf
YO
1725Possessive quantifiers are equivalent to putting the item they are applied
1726to inside of one of these constructs. The following equivalences apply:
1727
1728 Quantifier Form Bracketing Form
1729 --------------- ---------------
1730 PAT*+ (?>PAT*)
1731 PAT++ (?>PAT+)
1732 PAT?+ (?>PAT?)
1733 PAT{min,max}+ (?>PAT{min,max})
1734
9d1a5160 1735=item C<(?[ ])>
f4f5fe57 1736
572224ce 1737See L<perlrecharclass/Extended Bracketed Character Classes>.
9d1a5160 1738
e2e6a0f1
YO
1739=back
1740
1741=head2 Special Backtracking Control Verbs
1742
1743B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1744removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1745be noted to avoid problems during upgrades.
1746
1747These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1748otherwise stated the ARG argument is optional; in some cases, it is
1749forbidden.
1750
1751Any pattern containing a special backtracking verb that allows an argument
e1020413 1752has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1753C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1754rules apply:
e2e6a0f1 1755
5d458dd8
YO
1756On failure, the C<$REGERROR> variable will be set to the ARG value of the
1757verb pattern, if the verb was involved in the failure of the match. If the
1758ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1759name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1760none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1761
5d458dd8
YO
1762On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1763the C<$REGMARK> variable will be set to the name of the last
1764C<(*MARK:NAME)> pattern executed. See the explanation for the
1765C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1766
5d458dd8 1767B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1768and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1769readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1770Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1771
1772If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1773argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1774
70ca8714 1775=over 3
e2e6a0f1
YO
1776
1777=item Verbs that take an argument
1778
1779=over 4
1780
5d458dd8 1781=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1782X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1783
5d458dd8
YO
1784This zero-width pattern prunes the backtracking tree at the current point
1785when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1786where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1787A may backtrack as necessary to match. Once it is reached, matching
1788continues in B, which may also backtrack as necessary; however, should B
1789not match, then no further backtracking will take place, and the pattern
1790will fail outright at the current starting position.
54612592
YO
1791
1792The following example counts all the possible matching strings in a
1793pattern (without actually matching any of them).
1794
e2e6a0f1 1795 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1796 print "Count=$count\n";
1797
1798which produces:
1799
1800 aaab
1801 aaa
1802 aa
1803 a
1804 aab
1805 aa
1806 a
1807 ab
1808 a
1809 Count=9
1810
5d458dd8 1811If we add a C<(*PRUNE)> before the count like the following
54612592 1812
5d458dd8 1813 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1814 print "Count=$count\n";
1815
0b928c2f 1816we prevent backtracking and find the count of the longest matching string
353c6505 1817at each matching starting point like so:
54612592
YO
1818
1819 aaab
1820 aab
1821 ab
1822 Count=3
1823
5d458dd8 1824Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1825
5d458dd8
YO
1826See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1827control backtracking. In some cases, the use of C<(*PRUNE)> can be
1828replaced with a C<< (?>pattern) >> with no functional difference; however,
1829C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1830C<< (?>pattern) >> alone.
54612592 1831
5d458dd8
YO
1832=item C<(*SKIP)> C<(*SKIP:NAME)>
1833X<(*SKIP)>
e2e6a0f1 1834
5d458dd8 1835This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1836failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1837to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1838of this pattern. This effectively means that the regex engine "skips" forward
1839to this position on failure and tries to match again, (assuming that
1840there is sufficient room to match).
1841
1842The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1843C<(*MARK:NAME)> was encountered while matching, then it is that position
1844which is used as the "skip point". If no C<(*MARK)> of that name was
1845encountered, then the C<(*SKIP)> operator has no effect. When used
1846without a name the "skip point" is where the match point was when
1847executing the (*SKIP) pattern.
1848
0b928c2f 1849Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1850is twice as long:
1851
d1fbf752
KW
1852 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1853 print "Count=$count\n";
24b23f37
YO
1854
1855outputs
1856
1857 aaab
1858 aaab
1859 Count=2
1860
5d458dd8 1861Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1862executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1863C<(*SKIP)> was executed.
1864
5d458dd8 1865=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1866X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1867
1868This zero-width pattern can be used to mark the point reached in a string
1869when a certain part of the pattern has been successfully matched. This
1870mark may be given a name. A later C<(*SKIP)> pattern will then skip
1871forward to that point if backtracked into on failure. Any number of
b4222fa9 1872C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1873
1874In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1875can be used to "label" a pattern branch, so that after matching, the
1876program can determine which branches of the pattern were involved in the
1877match.
1878
1879When a match is successful, the C<$REGMARK> variable will be set to the
1880name of the most recently executed C<(*MARK:NAME)> that was involved
1881in the match.
1882
1883This can be used to determine which branch of a pattern was matched
c27a5cfe 1884without using a separate capture group for each branch, which in turn
5d458dd8
YO
1885can result in a performance improvement, as perl cannot optimize
1886C</(?:(x)|(y)|(z))/> as efficiently as something like
1887C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1888
1889When a match has failed, and unless another verb has been involved in
1890failing the match and has provided its own name to use, the C<$REGERROR>
1891variable will be set to the name of the most recently executed
1892C<(*MARK:NAME)>.
1893
42ac7c82 1894See L</(*SKIP)> for more details.
5d458dd8 1895
b62d2d15
YO
1896As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1897
5d458dd8
YO
1898=item C<(*THEN)> C<(*THEN:NAME)>
1899
ac9d8485 1900This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1901C<(*PRUNE)>, this verb always matches, and when backtracked into on
1902failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
1903innermost enclosing group (capturing or otherwise) that has alternations.
1904The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1905count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
1906
1907Its name comes from the observation that this operation combined with the
1908alternation operator (C<|>) can be used to create what is essentially a
1909pattern-based if/then/else block:
1910
1911 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1912
1913Note that if this operator is used and NOT inside of an alternation then
1914it acts exactly like the C<(*PRUNE)> operator.
1915
1916 / A (*PRUNE) B /
1917
1918is the same as
1919
1920 / A (*THEN) B /
1921
1922but
1923
25e26d77 1924 / ( A (*THEN) B | C ) /
5d458dd8
YO
1925
1926is not the same as
1927
25e26d77 1928 / ( A (*PRUNE) B | C ) /
5d458dd8
YO
1929
1930as after matching the A but failing on the B the C<(*THEN)> verb will
1931backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1932
cbeadc21
JV
1933=back
1934
1935=item Verbs without an argument
1936
1937=over 4
1938
e2e6a0f1
YO
1939=item C<(*COMMIT)>
1940X<(*COMMIT)>
24b23f37 1941
241e7389 1942This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1943zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1944into on failure it causes the match to fail outright. No further attempts
1945to find a valid match by advancing the start pointer will occur again.
1946For example,
24b23f37 1947
d1fbf752
KW
1948 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
1949 print "Count=$count\n";
24b23f37
YO
1950
1951outputs
1952
1953 aaab
1954 Count=1
1955
e2e6a0f1
YO
1956In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1957does not match, the regex engine will not try any further matching on the
1958rest of the string.
c277df42 1959
e2e6a0f1
YO
1960=item C<(*FAIL)> C<(*F)>
1961X<(*FAIL)> X<(*F)>
9af228c6 1962
e2e6a0f1
YO
1963This pattern matches nothing and always fails. It can be used to force the
1964engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1965fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1966
e2e6a0f1 1967It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1968
e2e6a0f1
YO
1969=item C<(*ACCEPT)>
1970X<(*ACCEPT)>
9af228c6 1971
e2e6a0f1
YO
1972B<WARNING:> This feature is highly experimental. It is not recommended
1973for production code.
9af228c6 1974
e2e6a0f1
YO
1975This pattern matches nothing and causes the end of successful matching at
1976the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1977whether there is actually more to match in the string. When inside of a
0d017f4d 1978nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1979via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1980
c27a5cfe 1981If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1982marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1983For instance:
9af228c6 1984
e2e6a0f1 1985 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1986
e2e6a0f1 1987will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 1988be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 1989string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1990
1991=back
c277df42 1992
a0d0e21e
LW
1993=back
1994
c07a80fd 1995=head2 Backtracking
d74e8afc 1996X<backtrack> X<backtracking>
c07a80fd 1997
35a734be
IZ
1998NOTE: This section presents an abstract approximation of regular
1999expression behavior. For a more rigorous (and complicated) view of
2000the rules involved in selecting a match among possible alternatives,
0d017f4d 2001see L<Combining RE Pieces>.
35a734be 2002
c277df42 2003A fundamental feature of regular expression matching involves the
5a964f20 2004notion called I<backtracking>, which is currently used (when needed)
0d017f4d 2005by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
2006C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
2007internally, but the general principle outlined here is valid.
c07a80fd
PP
2008
2009For a regular expression to match, the I<entire> regular expression must
2010match, not just part of it. So if the beginning of a pattern containing a
2011quantifier succeeds in a way that causes later parts in the pattern to
2012fail, the matching engine backs up and recalculates the beginning
2013part--that's why it's called backtracking.
2014
2015Here is an example of backtracking: Let's say you want to find the
2016word following "foo" in the string "Food is on the foo table.":
2017
2018 $_ = "Food is on the foo table.";
2019 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 2020 print "$2 follows $1.\n";
c07a80fd
PP
2021 }
2022
2023When the match runs, the first part of the regular expression (C<\b(foo)>)
2024finds a possible match right at the beginning of the string, and loads up
2025$1 with "Foo". However, as soon as the matching engine sees that there's
2026no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 2027mistake and starts over again one character after where it had the
c07a80fd
PP
2028tentative match. This time it goes all the way until the next occurrence
2029of "foo". The complete regular expression matches this time, and you get
2030the expected output of "table follows foo."
2031
2032Sometimes minimal matching can help a lot. Imagine you'd like to match
2033everything between "foo" and "bar". Initially, you write something
2034like this:
2035
2036 $_ = "The food is under the bar in the barn.";
2037 if ( /foo(.*)bar/ ) {
f793d64a 2038 print "got <$1>\n";
c07a80fd
PP
2039 }
2040
2041Which perhaps unexpectedly yields:
2042
2043 got <d is under the bar in the >
2044
2045That's because C<.*> was greedy, so you get everything between the
14218588 2046I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
2047to use minimal matching to make sure you get the text between a "foo"
2048and the first "bar" thereafter.
2049
2050 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2051 got <d is under the >
2052
0d017f4d 2053Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2054of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2055So you write this:
2056
2057 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2058 if ( /(.*)(\d*)/ ) { # Wrong!
2059 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2060 }
2061
2062That won't work at all, because C<.*> was greedy and gobbled up the
2063whole string. As C<\d*> can match on an empty string the complete
2064regular expression matched successfully.
2065
8e1088bc 2066 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2067
2068Here are some variants, most of which don't work:
2069
2070 $_ = "I have 2 numbers: 53147";
2071 @pats = qw{
f793d64a
KW
2072 (.*)(\d*)
2073 (.*)(\d+)
2074 (.*?)(\d*)
2075 (.*?)(\d+)
2076 (.*)(\d+)$
2077 (.*?)(\d+)$
2078 (.*)\b(\d+)$
2079 (.*\D)(\d+)$
c07a80fd
PP
2080 };
2081
2082 for $pat (@pats) {
f793d64a
KW
2083 printf "%-12s ", $pat;
2084 if ( /$pat/ ) {
2085 print "<$1> <$2>\n";
2086 } else {
2087 print "FAIL\n";
2088 }
c07a80fd
PP
2089 }
2090
2091That will print out:
2092
2093 (.*)(\d*) <I have 2 numbers: 53147> <>
2094 (.*)(\d+) <I have 2 numbers: 5314> <7>
2095 (.*?)(\d*) <> <>
2096 (.*?)(\d+) <I have > <2>
2097 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2098 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2099 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2100 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2101
2102As you see, this can be a bit tricky. It's important to realize that a
2103regular expression is merely a set of assertions that gives a definition
2104of success. There may be 0, 1, or several different ways that the
2105definition might succeed against a particular string. And if there are
5a964f20
TC
2106multiple ways it might succeed, you need to understand backtracking to
2107know which variety of success you will achieve.
c07a80fd 2108
19799a22 2109When using look-ahead assertions and negations, this can all get even
8b19b778 2110trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2111followed by "123". You might try to write that as
2112
871b0233 2113 $_ = "ABC123";
f793d64a
KW
2114 if ( /^\D*(?!123)/ ) { # Wrong!
2115 print "Yup, no 123 in $_\n";
871b0233 2116 }
c07a80fd
PP
2117
2118But that isn't going to match; at least, not the way you're hoping. It
2119claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2120why that pattern matches, contrary to popular expectations:
c07a80fd 2121
4358a253
SS
2122 $x = 'ABC123';
2123 $y = 'ABC445';
c07a80fd 2124
4358a253
SS
2125 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2126 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2127
4358a253
SS
2128 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2129 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2130
2131This prints
2132
2133 2: got ABC
2134 3: got AB
2135 4: got ABC
2136
5f05dabc 2137You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2138general purpose version of test 1. The important difference between
2139them is that test 3 contains a quantifier (C<\D*>) and so can use
2140backtracking, whereas test 1 will not. What's happening is
2141that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2142non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2143let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2144fail.
14218588 2145
c07a80fd 2146The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2147try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2148a quantifier (C<\D*>) has been used in the regular expression, the
2149search engine can backtrack and retry the match differently
54310121 2150in the hope of matching the complete regular expression.
c07a80fd 2151
5a964f20
TC
2152The pattern really, I<really> wants to succeed, so it uses the
2153standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2154time. Now there's indeed something following "AB" that is not
14218588 2155"123". It's "C123", which suffices.
c07a80fd 2156
14218588
GS
2157We can deal with this by using both an assertion and a negation.
2158We'll say that the first part in $1 must be followed both by a digit
2159and by something that's not "123". Remember that the look-aheads
2160are zero-width expressions--they only look, but don't consume any
2161of the string in their match. So rewriting this way produces what
c07a80fd
PP
2162you'd expect; that is, case 5 will fail, but case 6 succeeds:
2163
4358a253
SS
2164 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2165 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2166
2167 6: got ABC
2168
5a964f20 2169In other words, the two zero-width assertions next to each other work as though
19799a22 2170they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2171matches only if you're at the beginning of the line AND the end of the
2172line simultaneously. The deeper underlying truth is that juxtaposition in
2173regular expressions always means AND, except when you write an explicit OR
2174using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2175although the attempted matches are made at different positions because "a"
2176is not a zero-width assertion, but a one-width assertion.
2177
0d017f4d 2178B<WARNING>: Particularly complicated regular expressions can take
14218588 2179exponential time to solve because of the immense number of possible
0d017f4d 2180ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2181internal optimizations done by the regular expression engine, this will
2182take a painfully long time to run:
c07a80fd 2183
e1901655
IZ
2184 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2185
2186And if you used C<*>'s in the internal groups instead of limiting them
2187to 0 through 5 matches, then it would take forever--or until you ran
2188out of stack space. Moreover, these internal optimizations are not
2189always applicable. For example, if you put C<{0,5}> instead of C<*>
2190on the external group, no current optimization is applicable, and the
2191match takes a long time to finish.
c07a80fd 2192
9da458fc
IZ
2193A powerful tool for optimizing such beasts is what is known as an
2194"independent group",
96090e4f 2195which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2196zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2197the tail match, since they are in "logical" context: only
14218588 2198whether they match is considered relevant. For an example
9da458fc 2199where side-effects of look-ahead I<might> have influenced the
96090e4f 2200following match, see L</C<< (?>pattern) >>>.
c277df42 2201
a0d0e21e 2202=head2 Version 8 Regular Expressions
d74e8afc 2203X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2204
5a964f20 2205In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2206routines, here are the pattern-matching rules not described above.
2207
54310121 2208Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2209with a special meaning described here or above. You can cause
5a964f20 2210characters that normally function as metacharacters to be interpreted
5f05dabc 2211literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2212character; "\\" matches a "\"). This escape mechanism is also required
2213for the character used as the pattern delimiter.
2214
2215A series of characters matches that series of characters in the target
0b928c2f 2216string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2217string.
a0d0e21e
LW
2218
2219You can specify a character class, by enclosing a list of characters
5d458dd8 2220in C<[]>, which will match any character from the list. If the
a0d0e21e 2221first character after the "[" is "^", the class matches any character not
14218588 2222in the list. Within a list, the "-" character specifies a
5a964f20 2223range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2224inclusive. If you want either "-" or "]" itself to be a member of a
2225class, put it at the start of the list (possibly after a "^"), or
2226escape it with a backslash. "-" is also taken literally when it is
2227at the end of the list, just before the closing "]". (The
84850974
DD
2228following all specify the same class of three characters: C<[-az]>,
2229C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2230specifies a class containing twenty-six characters, even on EBCDIC-based
2231character sets.) Also, if you try to use the character
2232classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2233a range, the "-" is understood literally.
a0d0e21e 2234
8ada0baa
JH
2235Note also that the whole range idea is rather unportable between
2236character sets--and even within character sets they may cause results
2237you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2238that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2239[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2240spell out the character sets in full.
2241
54310121 2242Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2243used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2244"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2245of three octal digits, matches the character whose coded character set value
5d458dd8 2246is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2247matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2248matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2249matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2250
2251You can specify a series of alternatives for a pattern using "|" to
2252separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2253or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2254first alternative includes everything from the last pattern delimiter
0b928c2f 2255("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2256the last alternative contains everything from the last "|" to the next
0b928c2f 2257closing pattern delimiter. That's why it's common practice to include
14218588 2258alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2259start and end.
2260
5a964f20 2261Alternatives are tried from left to right, so the first
a3cb178b
GS
2262alternative found for which the entire expression matches, is the one that
2263is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2264example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2265part will match, as that is the first alternative tried, and it successfully
2266matches the target string. (This might not seem important, but it is
2267important when you are capturing matched text using parentheses.)
2268
5a964f20 2269Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2270so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2271
14218588
GS
2272Within a pattern, you may designate subpatterns for later reference
2273by enclosing them in parentheses, and you may refer back to the
2274I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2275\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2276of their opening parenthesis. A backreference matches whatever
2277actually matched the subpattern in the string being examined, not
d8b950dc 2278the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2279match "0x1234 0x4321", but not "0x1234 01234", because subpattern
22801 matched "0x", even though the rule C<0|0x> could potentially match
2281the leading 0 in the second number.
cb1a09d0 2282
0d017f4d 2283=head2 Warning on \1 Instead of $1
cb1a09d0 2284
5a964f20 2285Some people get too used to writing things like:
cb1a09d0
AD
2286
2287 $pattern =~ s/(\W)/\\\1/g;
2288
3ff1c45a
KW
2289This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2290shocking the
cb1a09d0 2291B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2292PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2293the usual double-quoted string means a control-A. The customary Unix
2294meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2295of doing that, you get yourself into trouble if you then add an C</e>
2296modifier.
2297
f793d64a 2298 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2299
2300Or if you try to do
2301
2302 s/(\d+)/\1000/;
2303
2304You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2305C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2306with the operation of matching a backreference. Certainly they mean two
2307different things on the I<left> side of the C<s///>.
9fa51da4 2308
0d017f4d 2309=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2310
19799a22 2311B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2312
2313Regular expressions provide a terse and powerful programming language. As
2314with most other power tools, power comes together with the ability
2315to wreak havoc.
2316
2317A common abuse of this power stems from the ability to make infinite
628afcb5 2318loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2319
2320 'foo' =~ m{ ( o? )* }x;
2321
0d017f4d 2322The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2323in the string is not moved by the match, C<o?> would match again and again
527e91da 2324because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2325is with the looping modifier C<//g>:
2326
2327 @matches = ( 'foo' =~ m{ o? }xg );
2328
2329or
2330
2331 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2332
2333or the loop implied by split().
2334
2335However, long experience has shown that many programming tasks may
14218588
GS
2336be significantly simplified by using repeated subexpressions that
2337may match zero-length substrings. Here's a simple example being:
c84d73f1 2338
d1fbf752 2339 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2340 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2341
9da458fc 2342Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2343the infinite loop>. The rules for this are different for lower-level
527e91da 2344loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2345ones like the C</g> modifier or split() operator.
2346
19799a22
GS
2347The lower-level loops are I<interrupted> (that is, the loop is
2348broken) when Perl detects that a repeated expression matched a
2349zero-length substring. Thus
c84d73f1
IZ
2350
2351 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2352
5d458dd8 2353is made equivalent to
c84d73f1 2354
0b928c2f
FC
2355 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2356
2357For example, this program
2358
2359 #!perl -l
2360 "aaaaab" =~ /
2361 (?:
2362 a # non-zero
2363 | # or
2364 (?{print "hello"}) # print hello whenever this
2365 # branch is tried
2366 (?=(b)) # zero-width assertion
2367 )* # any number of times
2368 /x;
2369 print $&;
2370 print $1;
c84d73f1 2371
0b928c2f
FC
2372prints
2373
2374 hello
2375 aaaaa
2376 b
2377
2378Notice that "hello" is only printed once, as when Perl sees that the sixth
2379iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2380the C<*>.
2381
2382The higher-level loops preserve an additional state between iterations:
5d458dd8 2383whether the last match was zero-length. To break the loop, the following
c84d73f1 2384match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2385This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2386and so the I<second best> match is chosen if the I<best> match is of
2387zero length.
2388
19799a22 2389For example:
c84d73f1
IZ
2390
2391 $_ = 'bar';
2392 s/\w??/<$&>/g;
2393
20fb949f 2394results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2395match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2396best> match is what is matched by C<\w>. Thus zero-length matches
2397alternate with one-character-long matches.
2398
5d458dd8 2399Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2400position one notch further in the string.
2401
19799a22 2402The additional state of being I<matched with zero-length> is associated with
c84d73f1 2403the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2404Zero-length matches at the end of the previous match are ignored
2405during C<split>.
c84d73f1 2406
0d017f4d 2407=head2 Combining RE Pieces
35a734be
IZ
2408
2409Each of the elementary pieces of regular expressions which were described
2410before (such as C<ab> or C<\Z>) could match at most one substring
2411at the given position of the input string. However, in a typical regular
2412expression these elementary pieces are combined into more complicated
0b928c2f 2413patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2414(in these examples C<S> and C<T> are regular subexpressions).
2415
2416Such combinations can include alternatives, leading to a problem of choice:
2417if we match a regular expression C<a|ab> against C<"abc">, will it match
2418substring C<"a"> or C<"ab">? One way to describe which substring is
2419actually matched is the concept of backtracking (see L<"Backtracking">).
2420However, this description is too low-level and makes you think
2421in terms of a particular implementation.
2422
2423Another description starts with notions of "better"/"worse". All the
2424substrings which may be matched by the given regular expression can be
2425sorted from the "best" match to the "worst" match, and it is the "best"
2426match which is chosen. This substitutes the question of "what is chosen?"
2427by the question of "which matches are better, and which are worse?".
2428
2429Again, for elementary pieces there is no such question, since at most
2430one match at a given position is possible. This section describes the
2431notion of better/worse for combining operators. In the description
2432below C<S> and C<T> are regular subexpressions.
2433
13a2d996 2434=over 4
35a734be
IZ
2435
2436=item C<ST>
2437
2438Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2439substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2440which can be matched by C<T>.
35a734be 2441
0b928c2f 2442If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2443match than C<A'B'>.
2444
2445If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2446C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2447
2448=item C<S|T>
2449
2450When C<S> can match, it is a better match than when only C<T> can match.
2451
2452Ordering of two matches for C<S> is the same as for C<S>. Similar for
2453two matches for C<T>.
2454
2455=item C<S{REPEAT_COUNT}>
2456
2457Matches as C<SSS...S> (repeated as many times as necessary).
2458
2459=item C<S{min,max}>
2460
2461Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2462
2463=item C<S{min,max}?>
2464
2465Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2466
2467=item C<S?>, C<S*>, C<S+>
2468
2469Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2470
2471=item C<S??>, C<S*?>, C<S+?>
2472
2473Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2474
c47ff5f1 2475=item C<< (?>S) >>
35a734be
IZ
2476
2477Matches the best match for C<S> and only that.
2478
2479=item C<(?=S)>, C<(?<=S)>
2480
2481Only the best match for C<S> is considered. (This is important only if
2482C<S> has capturing parentheses, and backreferences are used somewhere
2483else in the whole regular expression.)
2484
2485=item C<(?!S)>, C<(?<!S)>
2486
2487For this grouping operator there is no need to describe the ordering, since
2488only whether or not C<S> can match is important.
2489
93f313ef 2490=item C<(??{ EXPR })>, C<(?I<PARNO>)>
35a734be
IZ
2491
2492The ordering is the same as for the regular expression which is
93f313ef 2493the result of EXPR, or the pattern contained by capture group I<PARNO>.
35a734be
IZ
2494
2495=item C<(?(condition)yes-pattern|no-pattern)>
2496
2497Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2498already determined. The ordering of the matches is the same as for the
2499chosen subexpression.
2500
2501=back
2502
2503The above recipes describe the ordering of matches I<at a given position>.
2504One more rule is needed to understand how a match is determined for the
2505whole regular expression: a match at an earlier position is always better
2506than a match at a later position.
2507
0d017f4d 2508=head2 Creating Custom RE Engines
c84d73f1 2509
0b928c2f
FC
2510As of Perl 5.10.0, one can create custom regular expression engines. This
2511is not for the faint of heart, as they have to plug in at the C level. See
2512L<perlreapi> for more details.
2513
2514As an alternative, overloaded constants (see L<overload>) provide a simple
2515way to extend the functionality of the RE engine, by substituting one
2516pattern for another.
c84d73f1
IZ
2517
2518Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2519matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2520characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2521at these positions, so we want to have each C<\Y|> in the place of the
2522more complicated version. We can create a module C<customre> to do
2523this:
2524
2525 package customre;
2526 use overload;
2527
2528 sub import {
2529 shift;
2530 die "No argument to customre::import allowed" if @_;
2531 overload::constant 'qr' => \&convert;
2532 }
2533
2534 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2535
580a9fe1
RGS
2536 # We must also take care of not escaping the legitimate \\Y|
2537 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2538 my %rules = ( '\\' => '\\\\',
f793d64a 2539 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2540 sub convert {
2541 my $re = shift;
5d458dd8 2542 $re =~ s{
c84d73f1
IZ
2543 \\ ( \\ | Y . )
2544 }
5d458dd8 2545 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2546 return $re;
2547 }
2548
2549Now C<use customre> enables the new escape in constant regular
2550expressions, i.e., those without any runtime variable interpolations.
2551As documented in L<overload>, this conversion will work only over
2552literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2553part of this regular expression needs to be converted explicitly
2554(but only if the special meaning of C<\Y|> should be enabled inside $re):
2555
2556 use customre;
2557 $re = <>;
2558 chomp $re;
2559 $re = customre::convert $re;
2560 /\Y|$re\Y|/;
2561
0b928c2f 2562=head2 PCRE/Python Support
1f1031fe 2563
0b928c2f 2564As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2565to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2566Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2567
2568=over 4
2569
ae5648b3 2570=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2571
c27a5cfe 2572Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2573
2574=item C<< (?P=NAME) >>
2575
c27a5cfe 2576Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2577
2578=item C<< (?P>NAME) >>
2579
c27a5cfe 2580Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2581
ee9b8eae 2582=back
1f1031fe 2583
19799a22
GS
2584=head1 BUGS
2585
88c9975e
KW
2586Many regular expression constructs don't work on EBCDIC platforms.
2587
ed7efc79
KW
2588There are a number of issues with regard to case-insensitive matching
2589in Unicode rules. See C<i> under L</Modifiers> above.
2590
9da458fc
IZ
2591This document varies from difficult to understand to completely
2592and utterly opaque. The wandering prose riddled with jargon is
2593hard to fathom in several places.
2594
2595This document needs a rewrite that separates the tutorial content
2596from the reference content.
19799a22
GS
2597
2598=head1 SEE ALSO
9fa51da4 2599
91e0c79e
MJD
2600L<perlrequick>.
2601
2602L<perlretut>.
2603
9b599b2a
GS
2604L<perlop/"Regexp Quote-Like Operators">.
2605
1e66bd83
PP
2606L<perlop/"Gory details of parsing quoted constructs">.
2607
14218588
GS
2608L<perlfaq6>.
2609
9b599b2a
GS
2610L<perlfunc/pos>.
2611
2612L<perllocale>.
2613
fb55449c
JH
2614L<perlebcdic>.
2615
14218588
GS
2616I<Mastering Regular Expressions> by Jeffrey Friedl, published
2617by O'Reilly and Associates.