This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlexperiment: mark regexp backtracking verbs as accepted
[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
13b0f67d
DM
94In Perl 5.20 and higher this is ignored. Due to a new copy-on-write
95mechanism, ${^PREMATCH}, ${^MATCH}, and ${^POSTMATCH} will be available
96after the match regardless of the modifier.
97
e2e6bec7
DN
98=item g and c
99X</g> X</c>
100
101Global matching, and keep the Current position after failed matching.
102Unlike i, m, s and x, these two flags affect the way the regex is used
103rather than the regex itself. See
104L<perlretut/"Using regular expressions in Perl"> for further explanation
105of the g and c modifiers.
106
b6fa137b
FC
107=item a, d, l and u
108X</a> X</d> X</l> X</u>
109
516074bb
KW
110These modifiers, all new in 5.14, affect which character-set semantics
111(Unicode, etc.) are used, as described below in
ed7efc79 112L</Character set modifiers>.
b6fa137b 113
55497cff 114=back
a0d0e21e 115
516074bb
KW
116Regular expression modifiers are usually written in documentation
117as e.g., "the C</x> modifier", even though the delimiter
b6fa137b 118in question might not really be a slash. The modifiers C</imsxadlup>
ab7bb42d 119may also be embedded within the regular expression itself using
ed7efc79 120the C<(?...)> construct, see L</Extended Patterns> below.
b6fa137b 121
ed7efc79
KW
122=head3 /x
123
b6fa137b 124C</x> tells
7b059540 125the regular expression parser to ignore most whitespace that is neither
55497cff 126backslashed nor within a character class. You can use this to break up
4633a7c4 127your regular expression into (slightly) more readable parts. The C<#>
54310121 128character is also treated as a metacharacter introducing a comment,
55497cff 129just as in ordinary Perl code. This also means that if you want real
14218588 130whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a 131class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
132escape them (using backslashes or C<\Q...\E>) or encode them using octal,
133hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
134making Perl's regular expressions more readable. Note that you have to
135be careful not to include the pattern delimiter in the comment--perl has
136no way of knowing you did not intend to close the pattern early. See
137the C-comment deletion code in L<perlop>. Also note that anything inside
7651b971 138a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
0b928c2f 139space interpretation within a single multi-character construct. For
7651b971 140example in C<\x{...}>, regardless of the C</x> modifier, there can be no
9bb1f947 141spaces. Same for a L<quantifier|/Quantifiers> such as C<{3}> or
000947ad
KW
142C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<(>,
143C<?>, and C<:>. Within any delimiters for such a
f9e949fd
KW
144construct, allowed spaces are not affected by C</x>, and depend on the
145construct. For example, C<\x{...}> can't have spaces because hexadecimal
146numbers don't have spaces in them. But, Unicode properties can have spaces, so
0b928c2f 147in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
9bb1f947 148L<perluniprops/Properties accessible through \p{} and \P{}>.
d74e8afc 149X</x>
a0d0e21e 150
ed7efc79
KW
151=head3 Character set modifiers
152
153C</d>, C</u>, C</a>, and C</l>, available starting in 5.14, are called
154the character set modifiers; they affect the character set semantics
155used for the regular expression.
156
808432af
KW
157The C</d>, C</u>, and C</l> modifiers are not likely to be of much use
158to you, and so you need not worry about them very much. They exist for
159Perl's internal use, so that complex regular expression data structures
160can be automatically serialized and later exactly reconstituted,
161including all their nuances. But, since Perl can't keep a secret, and
162there may be rare instances where they are useful, they are documented
163here.
ed7efc79 164
808432af
KW
165The C</a> modifier, on the other hand, may be useful. Its purpose is to
166allow code that is to work mostly on ASCII data to not have to concern
167itself with Unicode.
ca9560b2 168
808432af
KW
169Briefly, C</l> sets the character set to that of whatever B<L>ocale is in
170effect at the time of the execution of the pattern match.
ca9560b2 171
808432af
KW
172C</u> sets the character set to B<U>nicode.
173
174C</a> also sets the character set to Unicode, BUT adds several
175restrictions for B<A>SCII-safe matching.
176
177C</d> is the old, problematic, pre-5.14 B<D>efault character set
178behavior. Its only use is to force that old behavior.
179
180At any given time, exactly one of these modifiers is in effect. Their
181existence allows Perl to keep the originally compiled behavior of a
182regular expression, regardless of what rules are in effect when it is
183actually executed. And if it is interpolated into a larger regex, the
184original's rules continue to apply to it, and only it.
185
186The C</l> and C</u> modifiers are automatically selected for
187regular expressions compiled within the scope of various pragmas,
188and we recommend that in general, you use those pragmas instead of
189specifying these modifiers explicitly. For one thing, the modifiers
190affect only pattern matching, and do not extend to even any replacement
191done, whereas using the pragmas give consistent results for all
192appropriate operations within their scopes. For example,
193
194 s/foo/\Ubar/il
195
196will match "foo" using the locale's rules for case-insensitive matching,
197but the C</l> does not affect how the C<\U> operates. Most likely you
198want both of them to use locale rules. To do this, instead compile the
199regular expression within the scope of C<use locale>. This both
200implicitly adds the C</l> and applies locale rules to the C<\U>. The
201lesson is to C<use locale> and not C</l> explicitly.
202
203Similarly, it would be better to use C<use feature 'unicode_strings'>
204instead of,
205
206 s/foo/\Lbar/iu
207
208to get Unicode rules, as the C<\L> in the former (but not necessarily
209the latter) would also use Unicode rules.
210
211More detail on each of the modifiers follows. Most likely you don't
212need to know this detail for C</l>, C</u>, and C</d>, and can skip ahead
213to L<E<sol>a|/E<sol>a (and E<sol>aa)>.
ca9560b2 214
ed7efc79
KW
215=head4 /l
216
217means to use the current locale's rules (see L<perllocale>) when pattern
218matching. For example, C<\w> will match the "word" characters of that
219locale, and C<"/i"> case-insensitive matching will match according to
220the locale's case folding rules. The locale used will be the one in
221effect at the time of execution of the pattern match. This may not be
222the same as the compilation-time locale, and can differ from one match
223to another if there is an intervening call of the
b6fa137b 224L<setlocale() function|perllocale/The setlocale function>.
ed7efc79
KW
225
226Perl only supports single-byte locales. This means that code points
227above 255 are treated as Unicode no matter what locale is in effect.
228Under Unicode rules, there are a few case-insensitive matches that cross
229the 255/256 boundary. These are disallowed under C</l>. For example,
516074bb
KW
2300xFF (on ASCII platforms) does not caselessly match the character at
2310x178, C<LATIN CAPITAL LETTER Y WITH DIAERESIS>, because 0xFF may not be
232C<LATIN SMALL LETTER Y WITH DIAERESIS> in the current locale, and Perl
233has no way of knowing if that character even exists in the locale, much
234less what code point it is.
ed7efc79
KW
235
236This modifier may be specified to be the default by C<use locale>, but
237see L</Which character set modifier is in effect?>.
b6fa137b
FC
238X</l>
239
ed7efc79
KW
240=head4 /u
241
242means to use Unicode rules when pattern matching. On ASCII platforms,
243this means that the code points between 128 and 255 take on their
808432af
KW
244Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's).
245(Otherwise Perl considers their meanings to be undefined.) Thus,
246under this modifier, the ASCII platform effectively becomes a Unicode
247platform; and hence, for example, C<\w> will match any of the more than
248100_000 word characters in Unicode.
ed7efc79
KW
249
250Unlike most locales, which are specific to a language and country pair,
516074bb
KW
251Unicode classifies all the characters that are letters I<somewhere> in
252the world as
ed7efc79
KW
253C<\w>. For example, your locale might not think that C<LATIN SMALL
254LETTER ETH> is a letter (unless you happen to speak Icelandic), but
255Unicode does. Similarly, all the characters that are decimal digits
256somewhere in the world will match C<\d>; this is hundreds, not 10,
257possible matches. And some of those digits look like some of the 10
258ASCII digits, but mean a different number, so a human could easily think
259a number is a different quantity than it really is. For example,
260C<BENGALI DIGIT FOUR> (U+09EA) looks very much like an
261C<ASCII DIGIT EIGHT> (U+0038). And, C<\d+>, may match strings of digits
262that are a mixture from different writing systems, creating a security
67592e11 263issue. L<Unicode::UCD/num()> can be used to sort
516074bb
KW
264this out. Or the C</a> modifier can be used to force C<\d> to match
265just the ASCII 0 through 9.
ed7efc79 266
516074bb
KW
267Also, under this modifier, case-insensitive matching works on the full
268set of Unicode
ed7efc79
KW
269characters. The C<KELVIN SIGN>, for example matches the letters "k" and
270"K"; and C<LATIN SMALL LIGATURE FF> matches the sequence "ff", which,
271if you're not prepared, might make it look like a hexadecimal constant,
272presenting another potential security issue. See
273L<http://unicode.org/reports/tr36> for a detailed discussion of Unicode
274security issues.
275
ed7efc79 276This modifier may be specified to be the default by C<use feature
66cbab2c
KW
277'unicode_strings>, C<use locale ':not_characters'>, or
278C<L<use 5.012|perlfunc/use VERSION>> (or higher),
808432af 279but see L</Which character set modifier is in effect?>.
b6fa137b
FC
280X</u>
281
ed7efc79
KW
282=head4 /d
283
284This modifier means to use the "Default" native rules of the platform
285except when there is cause to use Unicode rules instead, as follows:
286
287=over 4
288
289=item 1
290
291the target string is encoded in UTF-8; or
292
293=item 2
294
295the pattern is encoded in UTF-8; or
296
297=item 3
298
299the pattern explicitly mentions a code point that is above 255 (say by
300C<\x{100}>); or
301
302=item 4
b6fa137b 303
ed7efc79
KW
304the pattern uses a Unicode name (C<\N{...}>); or
305
306=item 5
307
9d1a5160
KW
308the pattern uses a Unicode property (C<\p{...}>); or
309
310=item 6
311
312the pattern uses L</C<(?[ ])>>
ed7efc79
KW
313
314=back
315
316Another mnemonic for this modifier is "Depends", as the rules actually
317used depend on various things, and as a result you can get unexpected
808432af
KW
318results. See L<perlunicode/The "Unicode Bug">. The Unicode Bug has
319become rather infamous, leading to yet another (printable) name for this
320modifier, "Dodgy".
ed7efc79 321
4b9734bf
KW
322Unless the pattern or string are encoded in UTF-8, only ASCII characters
323can match positively.
ed7efc79
KW
324
325Here are some examples of how that works on an ASCII platform:
326
327 $str = "\xDF"; # $str is not in UTF-8 format.
328 $str =~ /^\w/; # No match, as $str isn't in UTF-8 format.
329 $str .= "\x{0e0b}"; # Now $str is in UTF-8 format.
330 $str =~ /^\w/; # Match! $str is now in UTF-8 format.
331 chop $str;
332 $str =~ /^\w/; # Still a match! $str remains in UTF-8 format.
333
808432af
KW
334This modifier is automatically selected by default when none of the
335others are, so yet another name for it is "Default".
336
337Because of the unexpected behaviors associated with this modifier, you
338probably should only use it to maintain weird backward compatibilities.
339
340=head4 /a (and /aa)
341
342This modifier stands for ASCII-restrict (or ASCII-safe). This modifier,
343unlike the others, may be doubled-up to increase its effect.
344
345When it appears singly, it causes the sequences C<\d>, C<\s>, C<\w>, and
346the Posix character classes to match only in the ASCII range. They thus
347revert to their pre-5.6, pre-Unicode meanings. Under C</a>, C<\d>
348always means precisely the digits C<"0"> to C<"9">; C<\s> means the five
d28d8023
KW
349characters C<[ \f\n\r\t]>, and starting in Perl v5.18, experimentally,
350the vertical tab; C<\w> means the 63 characters
808432af
KW
351C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
352C<[[:print:]]> match only the appropriate ASCII-range characters.
353
354This modifier is useful for people who only incidentally use Unicode,
355and who do not wish to be burdened with its complexities and security
356concerns.
357
358With C</a>, one can write C<\d> with confidence that it will only match
359ASCII characters, and should the need arise to match beyond ASCII, you
360can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>). There are
361similar C<\p{...}> constructs that can match beyond ASCII both white
362space (see L<perlrecharclass/Whitespace>), and Posix classes (see
363L<perlrecharclass/POSIX Character Classes>). Thus, this modifier
364doesn't mean you can't use Unicode, it means that to get Unicode
365matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
366signals Unicode.
367
368As you would expect, this modifier causes, for example, C<\D> to mean
369the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
370C<\D>, C<\S>, and C<\W>. C<\b> still means to match at the boundary
371between C<\w> and C<\W>, using the C</a> definitions of them (similarly
372for C<\B>).
373
374Otherwise, C</a> behaves like the C</u> modifier, in that
375case-insensitive matching uses Unicode semantics; for example, "k" will
376match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
377points in the Latin1 range, above ASCII will have Unicode rules when it
378comes to case-insensitive matching.
379
380To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
381specify the "a" twice, for example C</aai> or C</aia>. (The first
382occurrence of "a" restricts the C<\d>, etc., and the second occurrence
383adds the C</i> restrictions.) But, note that code points outside the
384ASCII range will use Unicode rules for C</i> matching, so the modifier
385doesn't really restrict things to just ASCII; it just forbids the
386intermixing of ASCII and non-ASCII.
387
388To summarize, this modifier provides protection for applications that
389don't wish to be exposed to all of Unicode. Specifying it twice
390gives added protection.
391
392This modifier may be specified to be the default by C<use re '/a'>
393or C<use re '/aa'>. If you do so, you may actually have occasion to use
31dc26d6 394the C</u> modifier explicitly if there are a few regular expressions
808432af
KW
395where you do want full Unicode rules (but even here, it's best if
396everything were under feature C<"unicode_strings">, along with the
397C<use re '/aa'>). Also see L</Which character set modifier is in
398effect?>.
399X</a>
400X</aa>
401
ed7efc79
KW
402=head4 Which character set modifier is in effect?
403
404Which of these modifiers is in effect at any given point in a regular
808432af
KW
405expression depends on a fairly complex set of interactions. These have
406been designed so that in general you don't have to worry about it, but
407this section gives the gory details. As
ed7efc79
KW
408explained below in L</Extended Patterns> it is possible to explicitly
409specify modifiers that apply only to portions of a regular expression.
410The innermost always has priority over any outer ones, and one applying
6368643f
KW
411to the whole expression has priority over any of the default settings that are
412described in the remainder of this section.
ed7efc79 413
916cec3f 414The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
ed7efc79
KW
415default modifiers (including these) for regular expressions compiled
416within its scope. This pragma has precedence over the other pragmas
516074bb 417listed below that also change the defaults.
ed7efc79
KW
418
419Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
66cbab2c 420and C<L<use feature 'unicode_strings|feature>>, or
ed7efc79
KW
421C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
422C</u> when not in the same scope as either C<L<use locale|perllocale>>
66cbab2c
KW
423or C<L<use bytes|bytes>>.
424(C<L<use locale ':not_characters'|perllocale/Unicode and UTF-8>> also
425sets the default to C</u>, overriding any plain C<use locale>.)
426Unlike the mechanisms mentioned above, these
6368643f
KW
427affect operations besides regular expressions pattern matching, and so
428give more consistent results with other operators, including using
429C<\U>, C<\l>, etc. in substitution replacements.
ed7efc79
KW
430
431If none of the above apply, for backwards compatibility reasons, the
432C</d> modifier is the one in effect by default. As this can lead to
433unexpected results, it is best to specify which other rule set should be
434used.
435
436=head4 Character set modifier behavior prior to Perl 5.14
437
438Prior to 5.14, there were no explicit modifiers, but C</l> was implied
439for regexes compiled within the scope of C<use locale>, and C</d> was
440implied otherwise. However, interpolating a regex into a larger regex
441would ignore the original compilation in favor of whatever was in effect
442at the time of the second compilation. There were a number of
443inconsistencies (bugs) with the C</d> modifier, where Unicode rules
444would be used when inappropriate, and vice versa. C<\p{}> did not imply
445Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
b6fa137b 446
a0d0e21e
LW
447=head2 Regular Expressions
448
04838cea
RGS
449=head3 Metacharacters
450
384f06ae 451The patterns used in Perl pattern matching evolved from those supplied in
14218588 452the Version 8 regex routines. (The routines are derived
19799a22
GS
453(distantly) from Henry Spencer's freely redistributable reimplementation
454of the V8 routines.) See L<Version 8 Regular Expressions> for
455details.
a0d0e21e
LW
456
457In particular the following metacharacters have their standard I<egrep>-ish
458meanings:
d74e8afc
ITB
459X<metacharacter>
460X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
461
a0d0e21e 462
f793d64a
KW
463 \ Quote the next metacharacter
464 ^ Match the beginning of the line
465 . Match any character (except newline)
466 $ Match the end of the line (or before newline at the end)
467 | Alternation
468 () Grouping
469 [] Bracketed Character class
a0d0e21e 470
14218588
GS
471By default, the "^" character is guaranteed to match only the
472beginning of the string, the "$" character only the end (or before the
473newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
474assumption that the string contains only one line. Embedded newlines
475will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 476string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
477newline within the string (except if the newline is the last character in
478the string), and "$" will match before any newline. At the
a0d0e21e
LW
479cost of a little more overhead, you can do this by using the /m modifier
480on the pattern match operator. (Older programs did this by setting C<$*>,
db7cd43a 481but this option was removed in perl 5.10.)
d74e8afc 482X<^> X<$> X</m>
a0d0e21e 483
14218588 484To simplify multi-line substitutions, the "." character never matches a
55497cff 485newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 486the string is a single line--even if it isn't.
d74e8afc 487X<.> X</s>
a0d0e21e 488
04838cea
RGS
489=head3 Quantifiers
490
a0d0e21e 491The following standard quantifiers are recognized:
d74e8afc 492X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 493
f793d64a
KW
494 * Match 0 or more times
495 + Match 1 or more times
496 ? Match 1 or 0 times
497 {n} Match exactly n times
498 {n,} Match at least n times
499 {n,m} Match at least n but not more than m times
a0d0e21e 500
0b928c2f 501(If a curly bracket occurs in any other context and does not form part of
4d68ffa0
KW
502a backslashed sequence like C<\x{...}>, it is treated as a regular
503character. In particular, the lower quantifier bound is not optional,
504and a typo in a quantifier silently causes it to be treated as the
505literal characters. For example,
9af81bfe
KW
506
507 /o{4,3}/
508
509looks like a quantifier that matches 0 times, since 4 is greater than 3,
510but it really means to match the sequence of six characters
4d68ffa0
KW
511S<C<"o { 4 , 3 }">>. It is planned to eventually require literal uses
512of curly brackets to be escaped, say by preceding them with a backslash
513or enclosing them within square brackets, (C<"\{"> or C<"[{]">). This
514change will allow for future syntax extensions (like making the lower
515bound of a quantifier optional), and better error checking. In the
516meantime, you should get in the habit of escaping all instances where
517you mean a literal "{".)
9af81bfe
KW
518
519The "*" quantifier is equivalent to C<{0,}>, the "+"
527e91da 520quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 521to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
522This is usually 32766 on the most common platforms. The actual limit can
523be seen in the error message generated by code such as this:
524
820475bd 525 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 526
54310121
PP
527By default, a quantified subpattern is "greedy", that is, it will match as
528many times as possible (given a particular starting location) while still
529allowing the rest of the pattern to match. If you want it to match the
530minimum number of times possible, follow the quantifier with a "?". Note
531that the meanings don't change, just the "greediness":
0d017f4d 532X<metacharacter> X<greedy> X<greediness>
d74e8afc 533X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 534
f793d64a
KW
535 *? Match 0 or more times, not greedily
536 +? Match 1 or more times, not greedily
537 ?? Match 0 or 1 time, not greedily
0b928c2f 538 {n}? Match exactly n times, not greedily (redundant)
f793d64a
KW
539 {n,}? Match at least n times, not greedily
540 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 541
5f3789aa 542Normally when a quantified subpattern does not allow the rest of the
b9b4dddf 543overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 544sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
545as well.
546
f793d64a
KW
547 *+ Match 0 or more times and give nothing back
548 ++ Match 1 or more times and give nothing back
549 ?+ Match 0 or 1 time and give nothing back
550 {n}+ Match exactly n times and give nothing back (redundant)
551 {n,}+ Match at least n times and give nothing back
552 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
553
554For instance,
555
556 'aaaa' =~ /a++a/
557
558will never match, as the C<a++> will gobble up all the C<a>'s in the
559string and won't leave any for the remaining part of the pattern. This
560feature can be extremely useful to give perl hints about where it
561shouldn't backtrack. For instance, the typical "match a double-quoted
562string" problem can be most efficiently performed when written as:
563
564 /"(?:[^"\\]++|\\.)*+"/
565
0d017f4d 566as we know that if the final quote does not match, backtracking will not
0b928c2f
FC
567help. See the independent subexpression
568L</C<< (?>pattern) >>> for more details;
b9b4dddf
YO
569possessive quantifiers are just syntactic sugar for that construct. For
570instance the above example could also be written as follows:
571
572 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
573
5f3789aa
YO
574Note that the possessive quantifier modifier can not be be combined
575with the non-greedy modifier. This is because it would make no sense.
576Consider the follow equivalency table:
577
578 Illegal Legal
579 ------------ ------
580 X??+ X{0}
581 X+?+ X{1}
582 X{min,max}?+ X{min}
583
04838cea
RGS
584=head3 Escape sequences
585
0b928c2f 586Because patterns are processed as double-quoted strings, the following
a0d0e21e
LW
587also work:
588
f793d64a
KW
589 \t tab (HT, TAB)
590 \n newline (LF, NL)
591 \r return (CR)
592 \f form feed (FF)
593 \a alarm (bell) (BEL)
594 \e escape (think troff) (ESC)
f793d64a 595 \cK control char (example: VT)
dc0d9c48 596 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 597 \N{name} named Unicode character or character sequence
f793d64a 598 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 599 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
600 \l lowercase next char (think vi)
601 \u uppercase next char (think vi)
602 \L lowercase till \E (think vi)
603 \U uppercase till \E (think vi)
604 \Q quote (disable) pattern metacharacters till \E
605 \E end either case modification or quoted section, think vi
a0d0e21e 606
9bb1f947 607Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 608
e1d1eefb 609=head3 Character Classes and other Special Escapes
04838cea 610
a0d0e21e 611In addition, Perl defines the following:
d0b16107 612X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 613
f793d64a
KW
614 Sequence Note Description
615 [...] [1] Match a character according to the rules of the
616 bracketed character class defined by the "...".
617 Example: [a-z] matches "a" or "b" or "c" ... or "z"
618 [[:...:]] [2] Match a character according to the rules of the POSIX
619 character class "..." within the outer bracketed
620 character class. Example: [[:upper:]] matches any
621 uppercase character.
572224ce 622 (?[...]) [8] Extended bracketed character class
d35dd6c6
KW
623 \w [3] Match a "word" character (alphanumeric plus "_", plus
624 other connector punctuation chars plus Unicode
0b928c2f 625 marks)
f793d64a
KW
626 \W [3] Match a non-"word" character
627 \s [3] Match a whitespace character
628 \S [3] Match a non-whitespace character
629 \d [3] Match a decimal digit character
630 \D [3] Match a non-digit character
631 \pP [3] Match P, named property. Use \p{Prop} for longer names
632 \PP [3] Match non-P
633 \X [4] Match Unicode "eXtended grapheme cluster"
634 \C Match a single C-language char (octet) even if that is
635 part of a larger UTF-8 character. Thus it breaks up
636 characters into their UTF-8 bytes, so you may end up
637 with malformed pieces of UTF-8. Unsupported in
638 lookbehind.
c27a5cfe 639 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
640 '1' may actually be any positive integer.
641 \g1 [5] Backreference to a specific or previous group,
642 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 643 previous group and may optionally be wrapped in
f793d64a
KW
644 curly brackets for safer parsing.
645 \g{name} [5] Named backreference
646 \k<name> [5] Named backreference
647 \K [6] Keep the stuff left of the \K, don't include it in $&
2171640d 648 \N [7] Any character but \n. Not affected by /s modifier
f793d64a
KW
649 \v [3] Vertical whitespace
650 \V [3] Not vertical whitespace
651 \h [3] Horizontal whitespace
652 \H [3] Not horizontal whitespace
653 \R [4] Linebreak
e1d1eefb 654
9bb1f947
KW
655=over 4
656
657=item [1]
658
659See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 660
9bb1f947 661=item [2]
b8c5462f 662
9bb1f947 663See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 664
9bb1f947 665=item [3]
5496314a 666
9bb1f947 667See L<perlrecharclass/Backslash sequences> for details.
5496314a 668
9bb1f947 669=item [4]
5496314a 670
9bb1f947 671See L<perlrebackslash/Misc> for details.
d0b16107 672
9bb1f947 673=item [5]
b8c5462f 674
c27a5cfe 675See L</Capture groups> below for details.
93733859 676
9bb1f947 677=item [6]
b8c5462f 678
9bb1f947
KW
679See L</Extended Patterns> below for details.
680
681=item [7]
682
683Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
684character or character sequence whose name is C<NAME>; and similarly
685when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
686code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947 687
572224ce
KW
688=item [8]
689
690See L<perlrecharclass/Extended Bracketed Character Classes> for details.
691
9bb1f947 692=back
d0b16107 693
04838cea
RGS
694=head3 Assertions
695
a0d0e21e 696Perl defines the following zero-width assertions:
d74e8afc
ITB
697X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
698X<regexp, zero-width assertion>
699X<regular expression, zero-width assertion>
700X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 701
9bb1f947
KW
702 \b Match a word boundary
703 \B Match except at a word boundary
704 \A Match only at beginning of string
705 \Z Match only at end of string, or before newline at the end
706 \z Match only at end of string
707 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 708 of prior m//g)
a0d0e21e 709
14218588 710A word boundary (C<\b>) is a spot between two characters
19799a22
GS
711that has a C<\w> on one side of it and a C<\W> on the other side
712of it (in either order), counting the imaginary characters off the
713beginning and end of the string as matching a C<\W>. (Within
714character classes C<\b> represents backspace rather than a word
715boundary, just as it normally does in any double-quoted string.)
716The C<\A> and C<\Z> are just like "^" and "$", except that they
717won't match multiple times when the C</m> modifier is used, while
718"^" and "$" will match at every internal line boundary. To match
719the actual end of the string and not ignore an optional trailing
720newline, use C<\z>.
d74e8afc 721X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
722
723The C<\G> assertion can be used to chain global matches (using
724C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
725It is also useful when writing C<lex>-like scanners, when you have
726several patterns that you want to match against consequent substrings
0b928c2f 727of your string; see the previous reference. The actual location
19799a22 728where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d 729an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
0b928c2f
FC
730matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
731is modified somewhat, in that contents to the left of C<\G> are
58e23c8d
YO
732not counted when determining the length of the match. Thus the following
733will not match forever:
d74e8afc 734X<\G>
c47ff5f1 735
e761bb84
CO
736 my $string = 'ABC';
737 pos($string) = 1;
738 while ($string =~ /(.\G)/g) {
739 print $1;
740 }
58e23c8d
YO
741
742It will print 'A' and then terminate, as it considers the match to
743be zero-width, and thus will not match at the same position twice in a
744row.
745
746It is worth noting that C<\G> improperly used can result in an infinite
747loop. Take care when using patterns that include C<\G> in an alternation.
748
c27a5cfe 749=head3 Capture groups
04838cea 750
c27a5cfe
KW
751The bracketing construct C<( ... )> creates capture groups (also referred to as
752capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
753the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
754for the second, and so on.
755This is called a I<backreference>.
d74e8afc 756X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 757X<regex, capture group> X<regexp, capture group>
d74e8afc 758X<regular expression, capture buffer> X<backreference>
c27a5cfe 759X<regular expression, capture group> X<backreference>
1f1031fe 760X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
761X<named capture buffer> X<regular expression, named capture buffer>
762X<named capture group> X<regular expression, named capture group>
763X<%+> X<$+{name}> X<< \k<name> >>
764There is no limit to the number of captured substrings that you may use.
765Groups are numbered with the leftmost open parenthesis being number 1, etc. If
766a group did not match, the associated backreference won't match either. (This
767can happen if the group is optional, or in a different branch of an
768alternation.)
769You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
770this form, described below.
771
772You can also refer to capture groups relatively, by using a negative number, so
773that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
774group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
775example:
5624f11d
YO
776
777 /
c27a5cfe
KW
778 (Y) # group 1
779 ( # group 2
780 (X) # group 3
781 \g{-1} # backref to group 3
782 \g{-3} # backref to group 1
5624f11d
YO
783 )
784 /x
785
d8b950dc
KW
786would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
787interpolate regexes into larger regexes and not have to worry about the
788capture groups being renumbered.
789
790You can dispense with numbers altogether and create named capture groups.
791The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
792reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
793also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
794I<name> must not begin with a number, nor contain hyphens.
795When different groups within the same pattern have the same name, any reference
796to that name assumes the leftmost defined group. Named groups count in
797absolute and relative numbering, and so can also be referred to by those
798numbers.
799(It's possible to do things with named capture groups that would otherwise
800require C<(??{})>.)
801
802Capture group contents are dynamically scoped and available to you outside the
803pattern until the end of the enclosing block or until the next successful
804match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
805You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
806etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
807
808Braces are required in referring to named capture groups, but are optional for
809absolute or relative numbered ones. Braces are safer when creating a regex by
810concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
811contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
812is probably not what you intended.
813
814The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
815there were no named nor relative numbered capture groups. Absolute numbered
0b928c2f
FC
816groups were referred to using C<\1>,
817C<\2>, etc., and this notation is still
d8b950dc
KW
818accepted (and likely always will be). But it leads to some ambiguities if
819there are more than 9 capture groups, as C<\10> could mean either the tenth
820capture group, or the character whose ordinal in octal is 010 (a backspace in
821ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
822only if at least 10 left parentheses have opened before it. Likewise C<\11> is
823a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
824And so on. C<\1> through C<\9> are always interpreted as backreferences.
825There are several examples below that illustrate these perils. You can avoid
826the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
827and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
828digits padded with leading zeros, since a leading zero implies an octal
829constant.
d8b950dc
KW
830
831The C<\I<digit>> notation also works in certain circumstances outside
ed7efc79 832the pattern. See L</Warning on \1 Instead of $1> below for details.
81714fb9 833
14218588 834Examples:
a0d0e21e
LW
835
836 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
837
d8b950dc 838 /(.)\g1/ # find first doubled char
81714fb9
YO
839 and print "'$1' is the first doubled character\n";
840
841 /(?<char>.)\k<char>/ # ... a different way
842 and print "'$+{char}' is the first doubled character\n";
843
d8b950dc 844 /(?'char'.)\g1/ # ... mix and match
81714fb9 845 and print "'$1' is the first doubled character\n";
c47ff5f1 846
14218588 847 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
848 $hours = $1;
849 $minutes = $2;
850 $seconds = $3;
a0d0e21e 851 }
c47ff5f1 852
9d860678
KW
853 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
854 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
855 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
856 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
857
858 $a = '(.)\1'; # Creates problems when concatenated.
859 $b = '(.)\g{1}'; # Avoids the problems.
860 "aa" =~ /${a}/; # True
861 "aa" =~ /${b}/; # True
862 "aa0" =~ /${a}0/; # False!
863 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
864 "aa\x08" =~ /${a}0/; # True!
865 "aa\x08" =~ /${b}0/; # False
9d860678 866
14218588
GS
867Several special variables also refer back to portions of the previous
868match. C<$+> returns whatever the last bracket match matched.
869C<$&> returns the entire matched string. (At one point C<$0> did
870also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
871everything before the matched string. C<$'> returns everything
872after the matched string. And C<$^N> contains whatever was matched by
873the most-recently closed group (submatch). C<$^N> can be used in
874extended patterns (see below), for example to assign a submatch to a
81714fb9 875variable.
d74e8afc 876X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 877
d8b950dc
KW
878These special variables, like the C<%+> hash and the numbered match variables
879(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
880until the end of the enclosing block or until the next successful
881match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
882X<$+> X<$^N> X<$&> X<$`> X<$'>
883X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
884
0d017f4d 885B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 886which makes it easier to write code that tests for a series of more
665e98b9
JH
887specific cases and remembers the best match.
888
13b0f67d
DM
889B<WARNING>: If your code is to run on Perl 5.16 or earlier,
890beware that once Perl sees that you need one of C<$&>, C<$`>, or
14218588 891C<$'> anywhere in the program, it has to provide them for every
13b0f67d
DM
892pattern match. This may substantially slow your program.
893
894Perl uses the same mechanism to produce C<$1>, C<$2>, etc, so you also
895pay a price for each pattern that contains capturing parentheses.
896(To avoid this cost while retaining the grouping behaviour, use the
14218588
GS
897extended regular expression C<(?: ... )> instead.) But if you never
898use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
899parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
900if you can, but if you can't (and some algorithms really appreciate
901them), once you've used them once, use them at will, because you've
13b0f67d 902already paid the price.
d74e8afc 903X<$&> X<$`> X<$'>
68dc0745 904
13b0f67d
DM
905Perl 5.16 introduced a slightly more efficient mechanism that notes
906separately whether each of C<$`>, C<$&>, and C<$'> have been seen, and
907thus may only need to copy part of the string. Perl 5.20 introduced a
908much more efficient copy-on-write mechanism which eliminates any slowdown.
909
910As another workaround for this problem, Perl 5.10.0 introduced C<${^PREMATCH}>,
cde0cee5
YO
911C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
912and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 913successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
914The use of these variables incurs no global performance penalty, unlike
915their punctuation char equivalents, however at the trade-off that you
13b0f67d
DM
916have to tell perl when you want to use them. As of Perl 5.20, these three
917variables are equivalent to C<$`>, C<$&> and C<$'>, and C</p> is ignored.
87e95b7f 918X</p> X<p modifier>
cde0cee5 919
9d727203
KW
920=head2 Quoting metacharacters
921
19799a22
GS
922Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
923C<\w>, C<\n>. Unlike some other regular expression languages, there
924are no backslashed symbols that aren't alphanumeric. So anything
0f264506 925that looks like \\, \(, \), \[, \], \{, or \} is always
19799a22
GS
926interpreted as a literal character, not a metacharacter. This was
927once used in a common idiom to disable or quote the special meanings
928of regular expression metacharacters in a string that you want to
36bbe248 929use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
930
931 $pattern =~ s/(\W)/\\$1/g;
932
f1cbbd6e 933(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
934Today it is more common to use the quotemeta() function or the C<\Q>
935metaquoting escape sequence to disable all metacharacters' special
936meanings like this:
a0d0e21e
LW
937
938 /$unquoted\Q$quoted\E$unquoted/
939
9da458fc
IZ
940Beware that if you put literal backslashes (those not inside
941interpolated variables) between C<\Q> and C<\E>, double-quotish
942backslash interpolation may lead to confusing results. If you
943I<need> to use literal backslashes within C<\Q...\E>,
944consult L<perlop/"Gory details of parsing quoted constructs">.
945
736fe711
KW
946C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
947
19799a22
GS
948=head2 Extended Patterns
949
14218588 950Perl also defines a consistent extension syntax for features not
0b928c2f
FC
951found in standard tools like B<awk> and
952B<lex>. The syntax for most of these is a
14218588
GS
953pair of parentheses with a question mark as the first thing within
954the parentheses. The character after the question mark indicates
955the extension.
19799a22 956
14218588
GS
957The stability of these extensions varies widely. Some have been
958part of the core language for many years. Others are experimental
959and may change without warning or be completely removed. Check
960the documentation on an individual feature to verify its current
961status.
19799a22 962
14218588
GS
963A question mark was chosen for this and for the minimal-matching
964construct because 1) question marks are rare in older regular
965expressions, and 2) whenever you see one, you should stop and
0b928c2f 966"question" exactly what is going on. That's psychology....
a0d0e21e 967
70ca8714 968=over 4
a0d0e21e 969
cc6b7395 970=item C<(?#text)>
d74e8afc 971X<(?#)>
a0d0e21e 972
14218588 973A comment. The text is ignored. If the C</x> modifier enables
19799a22 974whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
975the comment as soon as it sees a C<)>, so there is no way to put a literal
976C<)> in the comment.
a0d0e21e 977
cfaf538b 978=item C<(?adlupimsx-imsx)>
fb85c044 979
cfaf538b 980=item C<(?^alupimsx)>
fb85c044 981X<(?)> X<(?^)>
19799a22 982
0b6d1084
JH
983One or more embedded pattern-match modifiers, to be turned on (or
984turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
985the remainder of the enclosing pattern group (if any).
986
fb85c044 987This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 988configuration file, taken from an argument, or specified in a table
0b928c2f
FC
989somewhere. Consider the case where some patterns want to be
990case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 991include C<(?i)> at the front of the pattern. For example:
19799a22
GS
992
993 $pattern = "foobar";
5d458dd8 994 if ( /$pattern/i ) { }
19799a22
GS
995
996 # more flexible:
997
998 $pattern = "(?i)foobar";
5d458dd8 999 if ( /$pattern/ ) { }
19799a22 1000
0b6d1084 1001These modifiers are restored at the end of the enclosing group. For example,
19799a22 1002
d8b950dc 1003 ( (?i) blah ) \s+ \g1
19799a22 1004
0d017f4d
WL
1005will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
1006repetition of the previous word, assuming the C</x> modifier, and no C</i>
1007modifier outside this group.
19799a22 1008
8eb5594e 1009These modifiers do not carry over into named subpatterns called in the
dd72e27b 1010enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
8eb5594e
DR
1011change the case-sensitivity of the "NAME" pattern.
1012
dc925305
KW
1013Any of these modifiers can be set to apply globally to all regular
1014expressions compiled within the scope of a C<use re>. See
a0bbd6ff 1015L<re/"'/flags' mode">.
dc925305 1016
9de15fec
KW
1017Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1018after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
1019C<"d">) may follow the caret to override it.
1020But a minus sign is not legal with it.
1021
dc925305 1022Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 1023that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 1024C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
1025others, and a maximum of one (or two C<a>'s) may appear in the
1026construct. Thus, for
0b928c2f 1027example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 1028C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1029
1030Note also that the C<p> modifier is special in that its presence
1031anywhere in a pattern has a global effect.
cde0cee5 1032
5a964f20 1033=item C<(?:pattern)>
d74e8afc 1034X<(?:)>
a0d0e21e 1035
cfaf538b 1036=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1037
cfaf538b 1038=item C<(?^aluimsx:pattern)>
fb85c044
KW
1039X<(?^:)>
1040
5a964f20
TC
1041This is for clustering, not capturing; it groups subexpressions like
1042"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1043
5a964f20 1044 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1045
1046is like
1047
5a964f20 1048 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1049
19799a22
GS
1050but doesn't spit out extra fields. It's also cheaper not to capture
1051characters if you don't need to.
a0d0e21e 1052
19799a22 1053Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1054C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1055
1056 /(?s-i:more.*than).*million/i
1057
14218588 1058is equivalent to the more verbose
ca9dfc88
IZ
1059
1060 /(?:(?s-i)more.*than).*million/i
1061
fb85c044 1062Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1063after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1064flags (except C<"d">) may follow the caret, so
fb85c044
KW
1065
1066 (?^x:foo)
1067
1068is equivalent to
1069
1070 (?x-ims:foo)
1071
1072The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1073surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1074modified by any flags specified.
1075
1076The caret allows for simpler stringification of compiled regular
1077expressions. These look like
1078
1079 (?^:pattern)
1080
1081with any non-default flags appearing between the caret and the colon.
1082A test that looks at such stringification thus doesn't need to have the
1083system default flags hard-coded in it, just the caret. If new flags are
1084added to Perl, the meaning of the caret's expansion will change to include
1085the default for those flags, so the test will still work, unchanged.
1086
1087Specifying a negative flag after the caret is an error, as the flag is
1088redundant.
1089
1090Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1091to match at the beginning.
1092
594d7033
YO
1093=item C<(?|pattern)>
1094X<(?|)> X<Branch reset>
1095
1096This is the "branch reset" pattern, which has the special property
c27a5cfe 1097that the capture groups are numbered from the same starting point
99d59c4d 1098in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1099
c27a5cfe 1100Capture groups are numbered from left to right, but inside this
693596a8 1101construct the numbering is restarted for each branch.
4deaaa80 1102
c27a5cfe 1103The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1104following this construct will be numbered as though the construct
1105contained only one branch, that being the one with the most capture
c27a5cfe 1106groups in it.
4deaaa80 1107
0b928c2f 1108This construct is useful when you want to capture one of a
4deaaa80
PJ
1109number of alternative matches.
1110
1111Consider the following pattern. The numbers underneath show in
c27a5cfe 1112which group the captured content will be stored.
594d7033
YO
1113
1114
1115 # before ---------------branch-reset----------- after
1116 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1117 # 1 2 2 3 2 3 4
1118
ab106183
A
1119Be careful when using the branch reset pattern in combination with
1120named captures. Named captures are implemented as being aliases to
c27a5cfe 1121numbered groups holding the captures, and that interferes with the
ab106183
A
1122implementation of the branch reset pattern. If you are using named
1123captures in a branch reset pattern, it's best to use the same names,
1124in the same order, in each of the alternations:
1125
1126 /(?| (?<a> x ) (?<b> y )
1127 | (?<a> z ) (?<b> w )) /x
1128
1129Not doing so may lead to surprises:
1130
1131 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1132 say $+ {a}; # Prints '12'
1133 say $+ {b}; # *Also* prints '12'.
1134
c27a5cfe
KW
1135The problem here is that both the group named C<< a >> and the group
1136named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1137
ee9b8eae
YO
1138=item Look-Around Assertions
1139X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1140
0b928c2f 1141Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1142pattern without including it in C<$&>. Positive assertions match when
1143their subpattern matches, negative assertions match when their subpattern
1144fails. Look-behind matches text up to the current match position,
1145look-ahead matches text following the current match position.
1146
1147=over 4
1148
5a964f20 1149=item C<(?=pattern)>
d74e8afc 1150X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1151
19799a22 1152A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1153matches a word followed by a tab, without including the tab in C<$&>.
1154
5a964f20 1155=item C<(?!pattern)>
d74e8afc 1156X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1157
19799a22 1158A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1159matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1160however that look-ahead and look-behind are NOT the same thing. You cannot
1161use this for look-behind.
7b8d334a 1162
5a964f20 1163If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1164will not do what you want. That's because the C<(?!foo)> is just saying that
1165the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1166match. Use look-behind instead (see below).
c277df42 1167
ee9b8eae
YO
1168=item C<(?<=pattern)> C<\K>
1169X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1170
c47ff5f1 1171A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1172matches a word that follows a tab, without including the tab in C<$&>.
1173Works only for fixed-width look-behind.
c277df42 1174
ee9b8eae
YO
1175There is a special form of this construct, called C<\K>, which causes the
1176regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1177not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1178look-behind. The use of C<\K> inside of another look-around assertion
1179is allowed, but the behaviour is currently not well defined.
1180
c62285ac 1181For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1182equivalent C<< (?<=...) >> construct, and it is especially useful in
1183situations where you want to efficiently remove something following
1184something else in a string. For instance
1185
1186 s/(foo)bar/$1/g;
1187
1188can be rewritten as the much more efficient
1189
1190 s/foo\Kbar//g;
1191
5a964f20 1192=item C<(?<!pattern)>
d74e8afc 1193X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1194
19799a22
GS
1195A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1196matches any occurrence of "foo" that does not follow "bar". Works
1197only for fixed-width look-behind.
c277df42 1198
ee9b8eae
YO
1199=back
1200
81714fb9
YO
1201=item C<(?'NAME'pattern)>
1202
1203=item C<< (?<NAME>pattern) >>
1204X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1205
c27a5cfe 1206A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1207parentheses C<()> but for the additional fact that the group
1208can be referred to by name in various regular expression
1209constructs (like C<\g{NAME}>) and can be accessed by name
1210after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1211for more details on the C<%+> and C<%-> hashes.
81714fb9 1212
c27a5cfe
KW
1213If multiple distinct capture groups have the same name then the
1214$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1215
0d017f4d 1216The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1217
1218B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1219function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1220numbered sequentially regardless of being named or not. Thus in the
1221pattern
1222
1223 /(x)(?<foo>y)(z)/
1224
1225$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1226the opposite which is what a .NET regex hacker might expect.
1227
1f1031fe
YO
1228Currently NAME is restricted to simple identifiers only.
1229In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1230its Unicode extension (see L<utf8>),
1231though it isn't extended by the locale (see L<perllocale>).
81714fb9 1232
1f1031fe 1233B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1234with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1235may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1236support the use of single quotes as a delimiter for the name.
81714fb9 1237
1f1031fe
YO
1238=item C<< \k<NAME> >>
1239
1240=item C<< \k'NAME' >>
81714fb9
YO
1241
1242Named backreference. Similar to numeric backreferences, except that
1243the group is designated by name and not number. If multiple groups
1244have the same name then it refers to the leftmost defined group in
1245the current match.
1246
0d017f4d 1247It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1248earlier in the pattern.
1249
1250Both forms are equivalent.
1251
1f1031fe 1252B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1253with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1254may be used instead of C<< \k<NAME> >>.
1f1031fe 1255
cc6b7395 1256=item C<(?{ code })>
d74e8afc 1257X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1258
19799a22 1259B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1260experimental, and may be changed without notice. Code executed that
1261has side effects may not perform identically from version to version
e128ab2c
DM
1262due to the effect of future optimisations in the regex engine. The
1263implementation of this feature was radically overhauled for the 5.18.0
1264release, and its behaviour in earlier versions of perl was much buggier,
1265especially in relation to parsing, lexical vars, scoping, recursion and
1266reentrancy.
c277df42 1267
e128ab2c
DM
1268This zero-width assertion executes any embedded Perl code. It always
1269succeeds, and its return value is set as C<$^R>.
19799a22 1270
e128ab2c
DM
1271In literal patterns, the code is parsed at the same time as the
1272surrounding code. While within the pattern, control is passed temporarily
1273back to the perl parser, until the logically-balancing closing brace is
1274encountered. This is similar to the way that an array index expression in
1275a literal string is handled, for example
77ea4f6d 1276
e128ab2c
DM
1277 "abc$array[ 1 + f('[') + g()]def"
1278
1279In particular, braces do not need to be balanced:
1280
576fa024 1281 s/abc(?{ f('{'); })/def/
e128ab2c
DM
1282
1283Even in a pattern that is interpolated and compiled at run-time, literal
1284code blocks will be compiled once, at perl compile time; the following
1285prints "ABCD":
1286
1287 print "D";
1288 my $qr = qr/(?{ BEGIN { print "A" } })/;
1289 my $foo = "foo";
1290 /$foo$qr(?{ BEGIN { print "B" } })/;
1291 BEGIN { print "C" }
1292
1293In patterns where the text of the code is derived from run-time
1294information rather than appearing literally in a source code /pattern/,
1295the code is compiled at the same time that the pattern is compiled, and
5771dda0 1296for reasons of security, C<use re 'eval'> must be in scope. This is to
e128ab2c
DM
1297stop user-supplied patterns containing code snippets from being
1298executable.
1299
5771dda0 1300In situations where you need to enable this with C<use re 'eval'>, you should
e128ab2c
DM
1301also have taint checking enabled. Better yet, use the carefully
1302constrained evaluation within a Safe compartment. See L<perlsec> for
1303details about both these mechanisms.
1304
1305From the viewpoint of parsing, lexical variable scope and closures,
1306
1307 /AAA(?{ BBB })CCC/
1308
1309behaves approximately like
1310
1311 /AAA/ && do { BBB } && /CCC/
1312
1313Similarly,
1314
1315 qr/AAA(?{ BBB })CCC/
1316
1317behaves approximately like
77ea4f6d 1318
e128ab2c
DM
1319 sub { /AAA/ && do { BBB } && /CCC/ }
1320
1321In particular:
1322
1323 { my $i = 1; $r = qr/(?{ print $i })/ }
1324 my $i = 2;
1325 /$r/; # prints "1"
1326
1327Inside a C<(?{...})> block, C<$_> refers to the string the regular
754091cb 1328expression is matching against. You can also use C<pos()> to know what is
fa11829f 1329the current position of matching within this string.
754091cb 1330
e128ab2c
DM
1331The code block introduces a new scope from the perspective of lexical
1332variable declarations, but B<not> from the perspective of C<local> and
1333similar localizing behaviours. So later code blocks within the same
1334pattern will still see the values which were localized in earlier blocks.
1335These accumulated localizations are undone either at the end of a
1336successful match, or if the assertion is backtracked (compare
1337L<"Backtracking">). For example,
b9ac3b5b
GS
1338
1339 $_ = 'a' x 8;
5d458dd8 1340 m<
d1fbf752 1341 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1342 (
5d458dd8 1343 a
b9ac3b5b 1344 (?{
d1fbf752
KW
1345 local $cnt = $cnt + 1; # Update $cnt,
1346 # backtracking-safe.
b9ac3b5b 1347 })
5d458dd8 1348 )*
b9ac3b5b 1349 aaaa
d1fbf752
KW
1350 (?{ $res = $cnt }) # On success copy to
1351 # non-localized location.
b9ac3b5b
GS
1352 >x;
1353
e128ab2c
DM
1354will initially increment C<$cnt> up to 8; then during backtracking, its
1355value will be unwound back to 4, which is the value assigned to C<$res>.
1356At the end of the regex execution, $cnt will be wound back to its initial
1357value of 0.
1358
1359This assertion may be used as the condition in a
b9ac3b5b 1360
e128ab2c
DM
1361 (?(condition)yes-pattern|no-pattern)
1362
1363switch. If I<not> used in this way, the result of evaluation of C<code>
1364is put into the special variable C<$^R>. This happens immediately, so
1365C<$^R> can be used from other C<(?{ code })> assertions inside the same
1366regular expression.
b9ac3b5b 1367
19799a22
GS
1368The assignment to C<$^R> above is properly localized, so the old
1369value of C<$^R> is restored if the assertion is backtracked; compare
1370L<"Backtracking">.
b9ac3b5b 1371
e128ab2c
DM
1372Note that the special variable C<$^N> is particularly useful with code
1373blocks to capture the results of submatches in variables without having to
1374keep track of the number of nested parentheses. For example:
1375
1376 $_ = "The brown fox jumps over the lazy dog";
1377 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1378 print "color = $color, animal = $animal\n";
1379
8988a1bb 1380
14455d6c 1381=item C<(??{ code })>
d74e8afc
ITB
1382X<(??{})>
1383X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1384
19799a22 1385B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1386experimental, and may be changed without notice. Code executed that
1387has side effects may not perform identically from version to version
1388due to the effect of future optimisations in the regex engine.
0f5d15d6 1389
e128ab2c
DM
1390This is a "postponed" regular subexpression. It behaves in I<exactly> the
1391same way as a C<(?{ code })> code block as described above, except that
1392its return value, rather than being assigned to C<$^R>, is treated as a
1393pattern, compiled if it's a string (or used as-is if its a qr// object),
1394then matched as if it were inserted instead of this construct.
6bda09f9 1395
e128ab2c
DM
1396During the matching of this sub-pattern, it has its own set of
1397captures which are valid during the sub-match, but are discarded once
1398control returns to the main pattern. For example, the following matches,
1399with the inner pattern capturing "B" and matching "BB", while the outer
1400pattern captures "A";
1401
1402 my $inner = '(.)\1';
1403 "ABBA" =~ /^(.)(??{ $inner })\1/;
1404 print $1; # prints "A";
6bda09f9 1405
e128ab2c
DM
1406Note that this means that there is no way for the inner pattern to refer
1407to a capture group defined outside. (The code block itself can use C<$1>,
1408etc., to refer to the enclosing pattern's capture groups.) Thus, although
0f5d15d6 1409
e128ab2c
DM
1410 ('a' x 100)=~/(??{'(.)' x 100})/
1411
1412I<will> match, it will I<not> set $1 on exit.
19799a22
GS
1413
1414The following pattern matches a parenthesized group:
0f5d15d6 1415
d1fbf752
KW
1416 $re = qr{
1417 \(
1418 (?:
1419 (?> [^()]+ ) # Non-parens without backtracking
1420 |
1421 (??{ $re }) # Group with matching parens
1422 )*
1423 \)
1424 }x;
0f5d15d6 1425
93f313ef
KW
1426See also
1427L<C<(?I<PARNO>)>|/(?PARNO) (?-PARNO) (?+PARNO) (?R) (?0)>
1428for a different, more efficient way to accomplish
6bda09f9
YO
1429the same task.
1430
e128ab2c
DM
1431Executing a postponed regular expression 50 times without consuming any
1432input string will result in a fatal error. The maximum depth is compiled
1433into perl, so changing it requires a custom build.
6bda09f9 1434
93f313ef 1435=item C<(?I<PARNO>)> C<(?-I<PARNO>)> C<(?+I<PARNO>)> C<(?R)> C<(?0)>
542fa716 1436X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1437X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1438X<regex, relative recursion>
6bda09f9 1439
e128ab2c
DM
1440Similar to C<(??{ code })> except that it does not involve executing any
1441code or potentially compiling a returned pattern string; instead it treats
1442the part of the current pattern contained within a specified capture group
1443as an independent pattern that must match at the current position.
1444Capture groups contained by the pattern will have the value as determined
1445by the outermost recursion.
6bda09f9 1446
93f313ef 1447I<PARNO> is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1448the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1449the beginning of the whole pattern. C<(?0)> is an alternate syntax for
93f313ef 1450C<(?R)>. If I<PARNO> is preceded by a plus or minus sign then it is assumed
c27a5cfe 1451to be relative, with negative numbers indicating preceding capture groups
542fa716 1452and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1453declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1454Note that the counting for relative recursion differs from that of
c27a5cfe 1455relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1456included.
6bda09f9 1457
81714fb9 1458The following pattern matches a function foo() which may contain
f145b7e9 1459balanced parentheses as the argument.
6bda09f9 1460
d1fbf752 1461 $re = qr{ ( # paren group 1 (full function)
81714fb9 1462 foo
d1fbf752 1463 ( # paren group 2 (parens)
6bda09f9 1464 \(
d1fbf752 1465 ( # paren group 3 (contents of parens)
6bda09f9 1466 (?:
d1fbf752 1467 (?> [^()]+ ) # Non-parens without backtracking
6bda09f9 1468 |
d1fbf752 1469 (?2) # Recurse to start of paren group 2
6bda09f9
YO
1470 )*
1471 )
1472 \)
1473 )
1474 )
1475 }x;
1476
1477If the pattern was used as follows
1478
1479 'foo(bar(baz)+baz(bop))'=~/$re/
1480 and print "\$1 = $1\n",
1481 "\$2 = $2\n",
1482 "\$3 = $3\n";
1483
1484the output produced should be the following:
1485
1486 $1 = foo(bar(baz)+baz(bop))
1487 $2 = (bar(baz)+baz(bop))
81714fb9 1488 $3 = bar(baz)+baz(bop)
6bda09f9 1489
c27a5cfe 1490If there is no corresponding capture group defined, then it is a
61528107 1491fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1492string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1493into perl, so changing it requires a custom build.
1494
542fa716
YO
1495The following shows how using negative indexing can make it
1496easier to embed recursive patterns inside of a C<qr//> construct
1497for later use:
1498
1499 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
c77257ed 1500 if (/foo $parens \s+ \+ \s+ bar $parens/x) {
542fa716
YO
1501 # do something here...
1502 }
1503
81714fb9 1504B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1505PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1506a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1507as atomic. Also, modifiers are resolved at compile time, so constructs
1508like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1509be processed.
6bda09f9 1510
894be9b7
YO
1511=item C<(?&NAME)>
1512X<(?&NAME)>
1513
93f313ef 1514Recurse to a named subpattern. Identical to C<(?I<PARNO>)> except that the
0d017f4d 1515parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1516the same name, then it recurses to the leftmost.
1517
1518It is an error to refer to a name that is not declared somewhere in the
1519pattern.
1520
1f1031fe
YO
1521B<NOTE:> In order to make things easier for programmers with experience
1522with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1523may be used instead of C<< (?&NAME) >>.
1f1031fe 1524
e2e6a0f1
YO
1525=item C<(?(condition)yes-pattern|no-pattern)>
1526X<(?()>
286f584a 1527
e2e6a0f1 1528=item C<(?(condition)yes-pattern)>
286f584a 1529
41ef34de
ML
1530Conditional expression. Matches C<yes-pattern> if C<condition> yields
1531a true value, matches C<no-pattern> otherwise. A missing pattern always
1532matches.
1533
25e26d77 1534C<(condition)> should be one of: 1) an integer in
e2e6a0f1 1535parentheses (which is valid if the corresponding pair of parentheses
25e26d77 1536matched); 2) a look-ahead/look-behind/evaluate zero-width assertion; 3) a
c27a5cfe 1537name in angle brackets or single quotes (which is valid if a group
25e26d77 1538with the given name matched); or 4) the special symbol (R) (true when
e2e6a0f1
YO
1539evaluated inside of recursion or eval). Additionally the R may be
1540followed by a number, (which will be true when evaluated when recursing
1541inside of the appropriate group), or by C<&NAME>, in which case it will
1542be true only when evaluated during recursion in the named group.
1543
1544Here's a summary of the possible predicates:
1545
1546=over 4
1547
1548=item (1) (2) ...
1549
c27a5cfe 1550Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1551
1552=item (<NAME>) ('NAME')
1553
c27a5cfe 1554Checks if a group with the given name has matched something.
e2e6a0f1 1555
f01cd190
FC
1556=item (?=...) (?!...) (?<=...) (?<!...)
1557
1558Checks whether the pattern matches (or does not match, for the '!'
1559variants).
1560
e2e6a0f1
YO
1561=item (?{ CODE })
1562
f01cd190 1563Treats the return value of the code block as the condition.
e2e6a0f1
YO
1564
1565=item (R)
1566
1567Checks if the expression has been evaluated inside of recursion.
1568
1569=item (R1) (R2) ...
1570
1571Checks if the expression has been evaluated while executing directly
1572inside of the n-th capture group. This check is the regex equivalent of
1573
1574 if ((caller(0))[3] eq 'subname') { ... }
1575
1576In other words, it does not check the full recursion stack.
1577
1578=item (R&NAME)
1579
1580Similar to C<(R1)>, this predicate checks to see if we're executing
1581directly inside of the leftmost group with a given name (this is the same
1582logic used by C<(?&NAME)> to disambiguate). It does not check the full
1583stack, but only the name of the innermost active recursion.
1584
1585=item (DEFINE)
1586
1587In this case, the yes-pattern is never directly executed, and no
1588no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1589See below for details.
1590
1591=back
1592
1593For example:
1594
1595 m{ ( \( )?
1596 [^()]+
1597 (?(1) \) )
1598 }x
1599
1600matches a chunk of non-parentheses, possibly included in parentheses
1601themselves.
1602
0b928c2f
FC
1603A special form is the C<(DEFINE)> predicate, which never executes its
1604yes-pattern directly, and does not allow a no-pattern. This allows one to
1605define subpatterns which will be executed only by the recursion mechanism.
e2e6a0f1
YO
1606This way, you can define a set of regular expression rules that can be
1607bundled into any pattern you choose.
1608
1609It is recommended that for this usage you put the DEFINE block at the
1610end of the pattern, and that you name any subpatterns defined within it.
1611
1612Also, it's worth noting that patterns defined this way probably will
31dc26d6 1613not be as efficient, as the optimizer is not very clever about
e2e6a0f1
YO
1614handling them.
1615
1616An example of how this might be used is as follows:
1617
2bf803e2 1618 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1619 (?(DEFINE)
2bf803e2
YO
1620 (?<NAME_PAT>....)
1621 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1622 )/x
1623
c27a5cfe
KW
1624Note that capture groups matched inside of recursion are not accessible
1625after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1626necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1627C<$+{NAME}> would be.
286f584a 1628
51a1303c
BF
1629Finally, keep in mind that subpatterns created inside a DEFINE block
1630count towards the absolute and relative number of captures, so this:
1631
1632 my @captures = "a" =~ /(.) # First capture
1633 (?(DEFINE)
1634 (?<EXAMPLE> 1 ) # Second capture
1635 )/x;
1636 say scalar @captures;
1637
1638Will output 2, not 1. This is particularly important if you intend to
1639compile the definitions with the C<qr//> operator, and later
1640interpolate them in another pattern.
1641
c47ff5f1 1642=item C<< (?>pattern) >>
6bda09f9 1643X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1644
19799a22
GS
1645An "independent" subexpression, one which matches the substring
1646that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1647position, and it matches I<nothing other than this substring>. This
19799a22
GS
1648construct is useful for optimizations of what would otherwise be
1649"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1650It may also be useful in places where the "grab all you can, and do not
1651give anything back" semantic is desirable.
19799a22 1652
c47ff5f1 1653For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1654(anchored at the beginning of string, as above) will match I<all>
1655characters C<a> at the beginning of string, leaving no C<a> for
1656C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1657since the match of the subgroup C<a*> is influenced by the following
1658group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1659C<a*ab> will match fewer characters than a standalone C<a*>, since
1660this makes the tail match.
1661
0b928c2f
FC
1662C<< (?>pattern) >> does not disable backtracking altogether once it has
1663matched. It is still possible to backtrack past the construct, but not
1664into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1665
c47ff5f1 1666An effect similar to C<< (?>pattern) >> may be achieved by writing
0b928c2f
FC
1667C<(?=(pattern))\g{-1}>. This matches the same substring as a standalone
1668C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
c47ff5f1 1669makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1670(The difference between these two constructs is that the second one
1671uses a capturing group, thus shifting ordinals of backreferences
1672in the rest of a regular expression.)
1673
1674Consider this pattern:
c277df42 1675
871b0233 1676 m{ \(
e2e6a0f1 1677 (
f793d64a 1678 [^()]+ # x+
e2e6a0f1 1679 |
871b0233
IZ
1680 \( [^()]* \)
1681 )+
e2e6a0f1 1682 \)
871b0233 1683 }x
5a964f20 1684
19799a22
GS
1685That will efficiently match a nonempty group with matching parentheses
1686two levels deep or less. However, if there is no such group, it
1687will take virtually forever on a long string. That's because there
1688are so many different ways to split a long string into several
1689substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1690to a subpattern of the above pattern. Consider how the pattern
1691above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1692seconds, but that each extra letter doubles this time. This
1693exponential performance will make it appear that your program has
14218588 1694hung. However, a tiny change to this pattern
5a964f20 1695
e2e6a0f1
YO
1696 m{ \(
1697 (
f793d64a 1698 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1699 |
871b0233
IZ
1700 \( [^()]* \)
1701 )+
e2e6a0f1 1702 \)
871b0233 1703 }x
c277df42 1704
c47ff5f1 1705which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1706this yourself would be a productive exercise), but finishes in a fourth
1707the time when used on a similar string with 1000000 C<a>s. Be aware,
0b928c2f
FC
1708however, that, when this construct is followed by a
1709quantifier, it currently triggers a warning message under
9f1b1f2d 1710the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1711C<"matches null string many times in regex">.
c277df42 1712
c47ff5f1 1713On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1714effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1715This was only 4 times slower on a string with 1000000 C<a>s.
1716
9da458fc
IZ
1717The "grab all you can, and do not give anything back" semantic is desirable
1718in many situations where on the first sight a simple C<()*> looks like
1719the correct solution. Suppose we parse text with comments being delimited
1720by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1721its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1722the comment delimiter, because it may "give up" some whitespace if
1723the remainder of the pattern can be made to match that way. The correct
1724answer is either one of these:
1725
1726 (?>#[ \t]*)
1727 #[ \t]*(?![ \t])
1728
1729For example, to grab non-empty comments into $1, one should use either
1730one of these:
1731
1732 / (?> \# [ \t]* ) ( .+ ) /x;
1733 / \# [ \t]* ( [^ \t] .* ) /x;
1734
1735Which one you pick depends on which of these expressions better reflects
1736the above specification of comments.
1737
6bda09f9
YO
1738In some literature this construct is called "atomic matching" or
1739"possessive matching".
1740
b9b4dddf
YO
1741Possessive quantifiers are equivalent to putting the item they are applied
1742to inside of one of these constructs. The following equivalences apply:
1743
1744 Quantifier Form Bracketing Form
1745 --------------- ---------------
1746 PAT*+ (?>PAT*)
1747 PAT++ (?>PAT+)
1748 PAT?+ (?>PAT?)
1749 PAT{min,max}+ (?>PAT{min,max})
1750
9d1a5160 1751=item C<(?[ ])>
f4f5fe57 1752
572224ce 1753See L<perlrecharclass/Extended Bracketed Character Classes>.
9d1a5160 1754
e2e6a0f1
YO
1755=back
1756
1757=head2 Special Backtracking Control Verbs
1758
e2e6a0f1
YO
1759These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1760otherwise stated the ARG argument is optional; in some cases, it is
1761forbidden.
1762
1763Any pattern containing a special backtracking verb that allows an argument
e1020413 1764has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1765C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1766rules apply:
e2e6a0f1 1767
5d458dd8
YO
1768On failure, the C<$REGERROR> variable will be set to the ARG value of the
1769verb pattern, if the verb was involved in the failure of the match. If the
1770ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1771name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1772none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1773
5d458dd8
YO
1774On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1775the C<$REGMARK> variable will be set to the name of the last
1776C<(*MARK:NAME)> pattern executed. See the explanation for the
1777C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1778
5d458dd8 1779B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1780and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1781readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1782Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1783
1784If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1785argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1786
70ca8714 1787=over 3
e2e6a0f1
YO
1788
1789=item Verbs that take an argument
1790
1791=over 4
1792
5d458dd8 1793=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1794X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1795
5d458dd8
YO
1796This zero-width pattern prunes the backtracking tree at the current point
1797when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1798where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1799A may backtrack as necessary to match. Once it is reached, matching
1800continues in B, which may also backtrack as necessary; however, should B
1801not match, then no further backtracking will take place, and the pattern
1802will fail outright at the current starting position.
54612592
YO
1803
1804The following example counts all the possible matching strings in a
1805pattern (without actually matching any of them).
1806
e2e6a0f1 1807 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1808 print "Count=$count\n";
1809
1810which produces:
1811
1812 aaab
1813 aaa
1814 aa
1815 a
1816 aab
1817 aa
1818 a
1819 ab
1820 a
1821 Count=9
1822
5d458dd8 1823If we add a C<(*PRUNE)> before the count like the following
54612592 1824
5d458dd8 1825 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1826 print "Count=$count\n";
1827
0b928c2f 1828we prevent backtracking and find the count of the longest matching string
353c6505 1829at each matching starting point like so:
54612592
YO
1830
1831 aaab
1832 aab
1833 ab
1834 Count=3
1835
5d458dd8 1836Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1837
5d458dd8
YO
1838See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1839control backtracking. In some cases, the use of C<(*PRUNE)> can be
1840replaced with a C<< (?>pattern) >> with no functional difference; however,
1841C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1842C<< (?>pattern) >> alone.
54612592 1843
5d458dd8
YO
1844=item C<(*SKIP)> C<(*SKIP:NAME)>
1845X<(*SKIP)>
e2e6a0f1 1846
5d458dd8 1847This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1848failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1849to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1850of this pattern. This effectively means that the regex engine "skips" forward
1851to this position on failure and tries to match again, (assuming that
1852there is sufficient room to match).
1853
1854The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1855C<(*MARK:NAME)> was encountered while matching, then it is that position
1856which is used as the "skip point". If no C<(*MARK)> of that name was
1857encountered, then the C<(*SKIP)> operator has no effect. When used
1858without a name the "skip point" is where the match point was when
1859executing the (*SKIP) pattern.
1860
0b928c2f 1861Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1862is twice as long:
1863
d1fbf752
KW
1864 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1865 print "Count=$count\n";
24b23f37
YO
1866
1867outputs
1868
1869 aaab
1870 aaab
1871 Count=2
1872
5d458dd8 1873Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1874executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1875C<(*SKIP)> was executed.
1876
5d458dd8 1877=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1878X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1879
1880This zero-width pattern can be used to mark the point reached in a string
1881when a certain part of the pattern has been successfully matched. This
1882mark may be given a name. A later C<(*SKIP)> pattern will then skip
1883forward to that point if backtracked into on failure. Any number of
b4222fa9 1884C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1885
1886In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1887can be used to "label" a pattern branch, so that after matching, the
1888program can determine which branches of the pattern were involved in the
1889match.
1890
1891When a match is successful, the C<$REGMARK> variable will be set to the
1892name of the most recently executed C<(*MARK:NAME)> that was involved
1893in the match.
1894
1895This can be used to determine which branch of a pattern was matched
c27a5cfe 1896without using a separate capture group for each branch, which in turn
5d458dd8
YO
1897can result in a performance improvement, as perl cannot optimize
1898C</(?:(x)|(y)|(z))/> as efficiently as something like
1899C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1900
1901When a match has failed, and unless another verb has been involved in
1902failing the match and has provided its own name to use, the C<$REGERROR>
1903variable will be set to the name of the most recently executed
1904C<(*MARK:NAME)>.
1905
42ac7c82 1906See L</(*SKIP)> for more details.
5d458dd8 1907
b62d2d15
YO
1908As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1909
5d458dd8
YO
1910=item C<(*THEN)> C<(*THEN:NAME)>
1911
ac9d8485 1912This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1913C<(*PRUNE)>, this verb always matches, and when backtracked into on
1914failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
1915innermost enclosing group (capturing or otherwise) that has alternations.
1916The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1917count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
1918
1919Its name comes from the observation that this operation combined with the
1920alternation operator (C<|>) can be used to create what is essentially a
1921pattern-based if/then/else block:
1922
1923 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1924
1925Note that if this operator is used and NOT inside of an alternation then
1926it acts exactly like the C<(*PRUNE)> operator.
1927
1928 / A (*PRUNE) B /
1929
1930is the same as
1931
1932 / A (*THEN) B /
1933
1934but
1935
25e26d77 1936 / ( A (*THEN) B | C ) /
5d458dd8
YO
1937
1938is not the same as
1939
25e26d77 1940 / ( A (*PRUNE) B | C ) /
5d458dd8
YO
1941
1942as after matching the A but failing on the B the C<(*THEN)> verb will
1943backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1944
cbeadc21
JV
1945=back
1946
1947=item Verbs without an argument
1948
1949=over 4
1950
e2e6a0f1
YO
1951=item C<(*COMMIT)>
1952X<(*COMMIT)>
24b23f37 1953
241e7389 1954This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1955zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1956into on failure it causes the match to fail outright. No further attempts
1957to find a valid match by advancing the start pointer will occur again.
1958For example,
24b23f37 1959
d1fbf752
KW
1960 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
1961 print "Count=$count\n";
24b23f37
YO
1962
1963outputs
1964
1965 aaab
1966 Count=1
1967
e2e6a0f1
YO
1968In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1969does not match, the regex engine will not try any further matching on the
1970rest of the string.
c277df42 1971
e2e6a0f1
YO
1972=item C<(*FAIL)> C<(*F)>
1973X<(*FAIL)> X<(*F)>
9af228c6 1974
e2e6a0f1
YO
1975This pattern matches nothing and always fails. It can be used to force the
1976engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1977fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1978
e2e6a0f1 1979It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1980
e2e6a0f1
YO
1981=item C<(*ACCEPT)>
1982X<(*ACCEPT)>
9af228c6 1983
e2e6a0f1
YO
1984This pattern matches nothing and causes the end of successful matching at
1985the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1986whether there is actually more to match in the string. When inside of a
0d017f4d 1987nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1988via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1989
c27a5cfe 1990If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1991marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1992For instance:
9af228c6 1993
e2e6a0f1 1994 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1995
e2e6a0f1 1996will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 1997be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 1998string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1999
2000=back
c277df42 2001
a0d0e21e
LW
2002=back
2003
c07a80fd 2004=head2 Backtracking
d74e8afc 2005X<backtrack> X<backtracking>
c07a80fd 2006
35a734be
IZ
2007NOTE: This section presents an abstract approximation of regular
2008expression behavior. For a more rigorous (and complicated) view of
2009the rules involved in selecting a match among possible alternatives,
0d017f4d 2010see L<Combining RE Pieces>.
35a734be 2011
c277df42 2012A fundamental feature of regular expression matching involves the
5a964f20 2013notion called I<backtracking>, which is currently used (when needed)
0d017f4d 2014by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
2015C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
2016internally, but the general principle outlined here is valid.
c07a80fd
PP
2017
2018For a regular expression to match, the I<entire> regular expression must
2019match, not just part of it. So if the beginning of a pattern containing a
2020quantifier succeeds in a way that causes later parts in the pattern to
2021fail, the matching engine backs up and recalculates the beginning
2022part--that's why it's called backtracking.
2023
2024Here is an example of backtracking: Let's say you want to find the
2025word following "foo" in the string "Food is on the foo table.":
2026
2027 $_ = "Food is on the foo table.";
2028 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 2029 print "$2 follows $1.\n";
c07a80fd
PP
2030 }
2031
2032When the match runs, the first part of the regular expression (C<\b(foo)>)
2033finds a possible match right at the beginning of the string, and loads up
2034$1 with "Foo". However, as soon as the matching engine sees that there's
2035no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 2036mistake and starts over again one character after where it had the
c07a80fd
PP
2037tentative match. This time it goes all the way until the next occurrence
2038of "foo". The complete regular expression matches this time, and you get
2039the expected output of "table follows foo."
2040
2041Sometimes minimal matching can help a lot. Imagine you'd like to match
2042everything between "foo" and "bar". Initially, you write something
2043like this:
2044
2045 $_ = "The food is under the bar in the barn.";
2046 if ( /foo(.*)bar/ ) {
f793d64a 2047 print "got <$1>\n";
c07a80fd
PP
2048 }
2049
2050Which perhaps unexpectedly yields:
2051
2052 got <d is under the bar in the >
2053
2054That's because C<.*> was greedy, so you get everything between the
14218588 2055I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
2056to use minimal matching to make sure you get the text between a "foo"
2057and the first "bar" thereafter.
2058
2059 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2060 got <d is under the >
2061
0d017f4d 2062Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2063of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2064So you write this:
2065
2066 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2067 if ( /(.*)(\d*)/ ) { # Wrong!
2068 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2069 }
2070
2071That won't work at all, because C<.*> was greedy and gobbled up the
2072whole string. As C<\d*> can match on an empty string the complete
2073regular expression matched successfully.
2074
8e1088bc 2075 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2076
2077Here are some variants, most of which don't work:
2078
2079 $_ = "I have 2 numbers: 53147";
2080 @pats = qw{
f793d64a
KW
2081 (.*)(\d*)
2082 (.*)(\d+)
2083 (.*?)(\d*)
2084 (.*?)(\d+)
2085 (.*)(\d+)$
2086 (.*?)(\d+)$
2087 (.*)\b(\d+)$
2088 (.*\D)(\d+)$
c07a80fd
PP
2089 };
2090
2091 for $pat (@pats) {
f793d64a
KW
2092 printf "%-12s ", $pat;
2093 if ( /$pat/ ) {
2094 print "<$1> <$2>\n";
2095 } else {
2096 print "FAIL\n";
2097 }
c07a80fd
PP
2098 }
2099
2100That will print out:
2101
2102 (.*)(\d*) <I have 2 numbers: 53147> <>
2103 (.*)(\d+) <I have 2 numbers: 5314> <7>
2104 (.*?)(\d*) <> <>
2105 (.*?)(\d+) <I have > <2>
2106 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2107 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2108 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2109 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2110
2111As you see, this can be a bit tricky. It's important to realize that a
2112regular expression is merely a set of assertions that gives a definition
2113of success. There may be 0, 1, or several different ways that the
2114definition might succeed against a particular string. And if there are
5a964f20
TC
2115multiple ways it might succeed, you need to understand backtracking to
2116know which variety of success you will achieve.
c07a80fd 2117
19799a22 2118When using look-ahead assertions and negations, this can all get even
8b19b778 2119trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2120followed by "123". You might try to write that as
2121
871b0233 2122 $_ = "ABC123";
f793d64a
KW
2123 if ( /^\D*(?!123)/ ) { # Wrong!
2124 print "Yup, no 123 in $_\n";
871b0233 2125 }
c07a80fd
PP
2126
2127But that isn't going to match; at least, not the way you're hoping. It
2128claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2129why that pattern matches, contrary to popular expectations:
c07a80fd 2130
4358a253
SS
2131 $x = 'ABC123';
2132 $y = 'ABC445';
c07a80fd 2133
4358a253
SS
2134 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2135 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2136
4358a253
SS
2137 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2138 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2139
2140This prints
2141
2142 2: got ABC
2143 3: got AB
2144 4: got ABC
2145
5f05dabc 2146You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2147general purpose version of test 1. The important difference between
2148them is that test 3 contains a quantifier (C<\D*>) and so can use
2149backtracking, whereas test 1 will not. What's happening is
2150that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2151non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2152let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2153fail.
14218588 2154
c07a80fd 2155The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2156try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2157a quantifier (C<\D*>) has been used in the regular expression, the
2158search engine can backtrack and retry the match differently
54310121 2159in the hope of matching the complete regular expression.
c07a80fd 2160
5a964f20
TC
2161The pattern really, I<really> wants to succeed, so it uses the
2162standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2163time. Now there's indeed something following "AB" that is not
14218588 2164"123". It's "C123", which suffices.
c07a80fd 2165
14218588
GS
2166We can deal with this by using both an assertion and a negation.
2167We'll say that the first part in $1 must be followed both by a digit
2168and by something that's not "123". Remember that the look-aheads
2169are zero-width expressions--they only look, but don't consume any
2170of the string in their match. So rewriting this way produces what
c07a80fd
PP
2171you'd expect; that is, case 5 will fail, but case 6 succeeds:
2172
4358a253
SS
2173 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2174 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2175
2176 6: got ABC
2177
5a964f20 2178In other words, the two zero-width assertions next to each other work as though
19799a22 2179they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2180matches only if you're at the beginning of the line AND the end of the
2181line simultaneously. The deeper underlying truth is that juxtaposition in
2182regular expressions always means AND, except when you write an explicit OR
2183using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2184although the attempted matches are made at different positions because "a"
2185is not a zero-width assertion, but a one-width assertion.
2186
0d017f4d 2187B<WARNING>: Particularly complicated regular expressions can take
14218588 2188exponential time to solve because of the immense number of possible
0d017f4d 2189ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2190internal optimizations done by the regular expression engine, this will
2191take a painfully long time to run:
c07a80fd 2192
e1901655
IZ
2193 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2194
2195And if you used C<*>'s in the internal groups instead of limiting them
2196to 0 through 5 matches, then it would take forever--or until you ran
2197out of stack space. Moreover, these internal optimizations are not
2198always applicable. For example, if you put C<{0,5}> instead of C<*>
2199on the external group, no current optimization is applicable, and the
2200match takes a long time to finish.
c07a80fd 2201
9da458fc
IZ
2202A powerful tool for optimizing such beasts is what is known as an
2203"independent group",
96090e4f 2204which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2205zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2206the tail match, since they are in "logical" context: only
14218588 2207whether they match is considered relevant. For an example
9da458fc 2208where side-effects of look-ahead I<might> have influenced the
96090e4f 2209following match, see L</C<< (?>pattern) >>>.
c277df42 2210
a0d0e21e 2211=head2 Version 8 Regular Expressions
d74e8afc 2212X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2213
5a964f20 2214In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2215routines, here are the pattern-matching rules not described above.
2216
54310121 2217Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2218with a special meaning described here or above. You can cause
5a964f20 2219characters that normally function as metacharacters to be interpreted
5f05dabc 2220literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2221character; "\\" matches a "\"). This escape mechanism is also required
2222for the character used as the pattern delimiter.
2223
2224A series of characters matches that series of characters in the target
0b928c2f 2225string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2226string.
a0d0e21e
LW
2227
2228You can specify a character class, by enclosing a list of characters
5d458dd8 2229in C<[]>, which will match any character from the list. If the
a0d0e21e 2230first character after the "[" is "^", the class matches any character not
14218588 2231in the list. Within a list, the "-" character specifies a
5a964f20 2232range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2233inclusive. If you want either "-" or "]" itself to be a member of a
2234class, put it at the start of the list (possibly after a "^"), or
2235escape it with a backslash. "-" is also taken literally when it is
2236at the end of the list, just before the closing "]". (The
84850974
DD
2237following all specify the same class of three characters: C<[-az]>,
2238C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2239specifies a class containing twenty-six characters, even on EBCDIC-based
2240character sets.) Also, if you try to use the character
2241classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2242a range, the "-" is understood literally.
a0d0e21e 2243
8ada0baa
JH
2244Note also that the whole range idea is rather unportable between
2245character sets--and even within character sets they may cause results
2246you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2247that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2248[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2249spell out the character sets in full.
2250
54310121 2251Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2252used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2253"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2254of three octal digits, matches the character whose coded character set value
5d458dd8 2255is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2256matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2257matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2258matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2259
2260You can specify a series of alternatives for a pattern using "|" to
2261separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2262or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2263first alternative includes everything from the last pattern delimiter
0b928c2f 2264("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2265the last alternative contains everything from the last "|" to the next
0b928c2f 2266closing pattern delimiter. That's why it's common practice to include
14218588 2267alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2268start and end.
2269
5a964f20 2270Alternatives are tried from left to right, so the first
a3cb178b
GS
2271alternative found for which the entire expression matches, is the one that
2272is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2273example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2274part will match, as that is the first alternative tried, and it successfully
2275matches the target string. (This might not seem important, but it is
2276important when you are capturing matched text using parentheses.)
2277
5a964f20 2278Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2279so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2280
14218588
GS
2281Within a pattern, you may designate subpatterns for later reference
2282by enclosing them in parentheses, and you may refer back to the
2283I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2284\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2285of their opening parenthesis. A backreference matches whatever
2286actually matched the subpattern in the string being examined, not
d8b950dc 2287the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2288match "0x1234 0x4321", but not "0x1234 01234", because subpattern
22891 matched "0x", even though the rule C<0|0x> could potentially match
2290the leading 0 in the second number.
cb1a09d0 2291
0d017f4d 2292=head2 Warning on \1 Instead of $1
cb1a09d0 2293
5a964f20 2294Some people get too used to writing things like:
cb1a09d0
AD
2295
2296 $pattern =~ s/(\W)/\\\1/g;
2297
3ff1c45a
KW
2298This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2299shocking the
cb1a09d0 2300B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2301PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2302the usual double-quoted string means a control-A. The customary Unix
2303meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2304of doing that, you get yourself into trouble if you then add an C</e>
2305modifier.
2306
f793d64a 2307 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2308
2309Or if you try to do
2310
2311 s/(\d+)/\1000/;
2312
2313You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2314C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2315with the operation of matching a backreference. Certainly they mean two
2316different things on the I<left> side of the C<s///>.
9fa51da4 2317
0d017f4d 2318=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2319
19799a22 2320B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2321
2322Regular expressions provide a terse and powerful programming language. As
2323with most other power tools, power comes together with the ability
2324to wreak havoc.
2325
2326A common abuse of this power stems from the ability to make infinite
628afcb5 2327loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2328
2329 'foo' =~ m{ ( o? )* }x;
2330
0d017f4d 2331The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2332in the string is not moved by the match, C<o?> would match again and again
527e91da 2333because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2334is with the looping modifier C<//g>:
2335
2336 @matches = ( 'foo' =~ m{ o? }xg );
2337
2338or
2339
2340 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2341
2342or the loop implied by split().
2343
2344However, long experience has shown that many programming tasks may
14218588
GS
2345be significantly simplified by using repeated subexpressions that
2346may match zero-length substrings. Here's a simple example being:
c84d73f1 2347
d1fbf752 2348 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2349 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2350
9da458fc 2351Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2352the infinite loop>. The rules for this are different for lower-level
527e91da 2353loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2354ones like the C</g> modifier or split() operator.
2355
19799a22
GS
2356The lower-level loops are I<interrupted> (that is, the loop is
2357broken) when Perl detects that a repeated expression matched a
2358zero-length substring. Thus
c84d73f1
IZ
2359
2360 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2361
5d458dd8 2362is made equivalent to
c84d73f1 2363
0b928c2f
FC
2364 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2365
2366For example, this program
2367
2368 #!perl -l
2369 "aaaaab" =~ /
2370 (?:
2371 a # non-zero
2372 | # or
2373 (?{print "hello"}) # print hello whenever this
2374 # branch is tried
2375 (?=(b)) # zero-width assertion
2376 )* # any number of times
2377 /x;
2378 print $&;
2379 print $1;
c84d73f1 2380
0b928c2f
FC
2381prints
2382
2383 hello
2384 aaaaa
2385 b
2386
2387Notice that "hello" is only printed once, as when Perl sees that the sixth
2388iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2389the C<*>.
2390
2391The higher-level loops preserve an additional state between iterations:
5d458dd8 2392whether the last match was zero-length. To break the loop, the following
c84d73f1 2393match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2394This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2395and so the I<second best> match is chosen if the I<best> match is of
2396zero length.
2397
19799a22 2398For example:
c84d73f1
IZ
2399
2400 $_ = 'bar';
2401 s/\w??/<$&>/g;
2402
20fb949f 2403results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2404match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2405best> match is what is matched by C<\w>. Thus zero-length matches
2406alternate with one-character-long matches.
2407
5d458dd8 2408Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2409position one notch further in the string.
2410
19799a22 2411The additional state of being I<matched with zero-length> is associated with
c84d73f1 2412the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2413Zero-length matches at the end of the previous match are ignored
2414during C<split>.
c84d73f1 2415
0d017f4d 2416=head2 Combining RE Pieces
35a734be
IZ
2417
2418Each of the elementary pieces of regular expressions which were described
2419before (such as C<ab> or C<\Z>) could match at most one substring
2420at the given position of the input string. However, in a typical regular
2421expression these elementary pieces are combined into more complicated
0b928c2f 2422patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2423(in these examples C<S> and C<T> are regular subexpressions).
2424
2425Such combinations can include alternatives, leading to a problem of choice:
2426if we match a regular expression C<a|ab> against C<"abc">, will it match
2427substring C<"a"> or C<"ab">? One way to describe which substring is
2428actually matched is the concept of backtracking (see L<"Backtracking">).
2429However, this description is too low-level and makes you think
2430in terms of a particular implementation.
2431
2432Another description starts with notions of "better"/"worse". All the
2433substrings which may be matched by the given regular expression can be
2434sorted from the "best" match to the "worst" match, and it is the "best"
2435match which is chosen. This substitutes the question of "what is chosen?"
2436by the question of "which matches are better, and which are worse?".
2437
2438Again, for elementary pieces there is no such question, since at most
2439one match at a given position is possible. This section describes the
2440notion of better/worse for combining operators. In the description
2441below C<S> and C<T> are regular subexpressions.
2442
13a2d996 2443=over 4
35a734be
IZ
2444
2445=item C<ST>
2446
2447Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2448substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2449which can be matched by C<T>.
35a734be 2450
0b928c2f 2451If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2452match than C<A'B'>.
2453
2454If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2455C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2456
2457=item C<S|T>
2458
2459When C<S> can match, it is a better match than when only C<T> can match.
2460
2461Ordering of two matches for C<S> is the same as for C<S>. Similar for
2462two matches for C<T>.
2463
2464=item C<S{REPEAT_COUNT}>
2465
2466Matches as C<SSS...S> (repeated as many times as necessary).
2467
2468=item C<S{min,max}>
2469
2470Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2471
2472=item C<S{min,max}?>
2473
2474Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2475
2476=item C<S?>, C<S*>, C<S+>
2477
2478Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2479
2480=item C<S??>, C<S*?>, C<S+?>
2481
2482Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2483
c47ff5f1 2484=item C<< (?>S) >>
35a734be
IZ
2485
2486Matches the best match for C<S> and only that.
2487
2488=item C<(?=S)>, C<(?<=S)>
2489
2490Only the best match for C<S> is considered. (This is important only if
2491C<S> has capturing parentheses, and backreferences are used somewhere
2492else in the whole regular expression.)
2493
2494=item C<(?!S)>, C<(?<!S)>
2495
2496For this grouping operator there is no need to describe the ordering, since
2497only whether or not C<S> can match is important.
2498
93f313ef 2499=item C<(??{ EXPR })>, C<(?I<PARNO>)>
35a734be
IZ
2500
2501The ordering is the same as for the regular expression which is
93f313ef 2502the result of EXPR, or the pattern contained by capture group I<PARNO>.
35a734be
IZ
2503
2504=item C<(?(condition)yes-pattern|no-pattern)>
2505
2506Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2507already determined. The ordering of the matches is the same as for the
2508chosen subexpression.
2509
2510=back
2511
2512The above recipes describe the ordering of matches I<at a given position>.
2513One more rule is needed to understand how a match is determined for the
2514whole regular expression: a match at an earlier position is always better
2515than a match at a later position.
2516
0d017f4d 2517=head2 Creating Custom RE Engines
c84d73f1 2518
0b928c2f
FC
2519As of Perl 5.10.0, one can create custom regular expression engines. This
2520is not for the faint of heart, as they have to plug in at the C level. See
2521L<perlreapi> for more details.
2522
2523As an alternative, overloaded constants (see L<overload>) provide a simple
2524way to extend the functionality of the RE engine, by substituting one
2525pattern for another.
c84d73f1
IZ
2526
2527Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2528matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2529characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2530at these positions, so we want to have each C<\Y|> in the place of the
2531more complicated version. We can create a module C<customre> to do
2532this:
2533
2534 package customre;
2535 use overload;
2536
2537 sub import {
2538 shift;
2539 die "No argument to customre::import allowed" if @_;
2540 overload::constant 'qr' => \&convert;
2541 }
2542
2543 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2544
580a9fe1
RGS
2545 # We must also take care of not escaping the legitimate \\Y|
2546 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2547 my %rules = ( '\\' => '\\\\',
f793d64a 2548 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2549 sub convert {
2550 my $re = shift;
5d458dd8 2551 $re =~ s{
c84d73f1
IZ
2552 \\ ( \\ | Y . )
2553 }
5d458dd8 2554 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2555 return $re;
2556 }
2557
2558Now C<use customre> enables the new escape in constant regular
2559expressions, i.e., those without any runtime variable interpolations.
2560As documented in L<overload>, this conversion will work only over
2561literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2562part of this regular expression needs to be converted explicitly
2563(but only if the special meaning of C<\Y|> should be enabled inside $re):
2564
2565 use customre;
2566 $re = <>;
2567 chomp $re;
2568 $re = customre::convert $re;
2569 /\Y|$re\Y|/;
2570
0b928c2f 2571=head2 PCRE/Python Support
1f1031fe 2572
0b928c2f 2573As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2574to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2575Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2576
2577=over 4
2578
ae5648b3 2579=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2580
c27a5cfe 2581Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2582
2583=item C<< (?P=NAME) >>
2584
c27a5cfe 2585Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2586
2587=item C<< (?P>NAME) >>
2588
c27a5cfe 2589Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2590
ee9b8eae 2591=back
1f1031fe 2592
19799a22
GS
2593=head1 BUGS
2594
88c9975e
KW
2595Many regular expression constructs don't work on EBCDIC platforms.
2596
ed7efc79
KW
2597There are a number of issues with regard to case-insensitive matching
2598in Unicode rules. See C<i> under L</Modifiers> above.
2599
9da458fc
IZ
2600This document varies from difficult to understand to completely
2601and utterly opaque. The wandering prose riddled with jargon is
2602hard to fathom in several places.
2603
2604This document needs a rewrite that separates the tutorial content
2605from the reference content.
19799a22
GS
2606
2607=head1 SEE ALSO
9fa51da4 2608
91e0c79e
MJD
2609L<perlrequick>.
2610
2611L<perlretut>.
2612
9b599b2a
GS
2613L<perlop/"Regexp Quote-Like Operators">.
2614
1e66bd83
PP
2615L<perlop/"Gory details of parsing quoted constructs">.
2616
14218588
GS
2617L<perlfaq6>.
2618
9b599b2a
GS
2619L<perlfunc/pos>.
2620
2621L<perllocale>.
2622
fb55449c
JH
2623L<perlebcdic>.
2624
14218588
GS
2625I<Mastering Regular Expressions> by Jeffrey Friedl, published
2626by O'Reilly and Associates.