This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlre: fix typo
[perl5.git] / pod / perlre.pod
CommitLineData
a0d0e21e 1=head1 NAME
d74e8afc 2X<regular expression> X<regex> X<regexp>
a0d0e21e
LW
3
4perlre - Perl regular expressions
5
6=head1 DESCRIPTION
7
5d458dd8 8This page describes the syntax of regular expressions in Perl.
91e0c79e 9
cc46d5f2 10If you haven't used regular expressions before, a quick-start
91e0c79e
MJD
11introduction is available in L<perlrequick>, and a longer tutorial
12introduction is available in L<perlretut>.
13
14For reference on how regular expressions are used in matching
15operations, plus various examples of the same, see discussions of
16C<m//>, C<s///>, C<qr//> and C<??> in L<perlop/"Regexp Quote-Like
17Operators">.
cb1a09d0 18
0d017f4d
WL
19
20=head2 Modifiers
21
19799a22 22Matching operations can have various modifiers. Modifiers
5a964f20 23that relate to the interpretation of the regular expression inside
19799a22 24are listed below. Modifiers that alter the way a regular expression
5d458dd8 25is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
1e66bd83 26L<perlop/"Gory details of parsing quoted constructs">.
a0d0e21e 27
55497cff
PP
28=over 4
29
54310121 30=item m
d74e8afc 31X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff
PP
32
33Treat string as multiple lines. That is, change "^" and "$" from matching
1e4aa4a5
AP
34the start or end of line only at the left and right ends of the string to
35matching them anywhere within the string.
55497cff 36
54310121 37=item s
d74e8afc
ITB
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
55497cff
PP
40
41Treat string as single line. That is, change "." to match any character
19799a22 42whatsoever, even a newline, which normally it would not match.
55497cff 43
34d67d80 44Used together, as C</ms>, they let the "." match any character whatsoever,
fb55449c 45while still allowing "^" and "$" to match, respectively, just after
19799a22 46and just before newlines within the string.
7b8d334a 47
87e95b7f
YO
48=item i
49X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
50X<regular expression, case-insensitive>
51
52Do case-insensitive pattern matching.
53
5027a30b
KW
54If locale matching rules are in effect, the case map is taken from the
55current
17580e7a 56locale for code points less than 255, and from Unicode rules for larger
ed7efc79
KW
57code points. However, matches that would cross the Unicode
58rules/non-Unicode rules boundary (ords 255/256) will not succeed. See
59L<perllocale>.
60
61There are a number of Unicode characters that match multiple characters
62under C</i>. For example, C<LATIN SMALL LIGATURE FI>
63should match the sequence C<fi>. Perl is not
64currently able to do this when the multiple characters are in the pattern and
65are split between groupings, or when one or more are quantified. Thus
66
67 "\N{LATIN SMALL LIGATURE FI}" =~ /fi/i; # Matches
68 "\N{LATIN SMALL LIGATURE FI}" =~ /[fi][fi]/i; # Doesn't match!
69 "\N{LATIN SMALL LIGATURE FI}" =~ /fi*/i; # Doesn't match!
70
71 # The below doesn't match, and it isn't clear what $1 and $2 would
72 # be even if it did!!
73 "\N{LATIN SMALL LIGATURE FI}" =~ /(f)(i)/i; # Doesn't match!
74
9d53c457
KW
75Perl doesn't match multiple characters in a bracketed
76character class unless the character that maps to them is explicitly
77mentioned, and it doesn't match them at all if the character class is
78inverted, which otherwise could be highly confusing. See
79L<perlrecharclass/Bracketed Character Classes>, and
1f59b283
KW
80L<perlrecharclass/Negation>.
81
54310121 82=item x
d74e8afc 83X</x>
55497cff
PP
84
85Extend your pattern's legibility by permitting whitespace and comments.
ed7efc79 86Details in L</"/x">
55497cff 87
87e95b7f
YO
88=item p
89X</p> X<regex, preserve> X<regexp, preserve>
90
632a1772 91Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
87e95b7f
YO
92${^POSTMATCH} are available for use after matching.
93
d78f32f6
FC
94In Perl 5.18 and higher this is ignored. ${^PREMATCH}, ${^MATCH}, and
95${^POSTMATCH} will be available after the match regardless of the modifier.
96
e2e6bec7
DN
97=item g and c
98X</g> X</c>
99
100Global matching, and keep the Current position after failed matching.
101Unlike i, m, s and x, these two flags affect the way the regex is used
102rather than the regex itself. See
103L<perlretut/"Using regular expressions in Perl"> for further explanation
104of the g and c modifiers.
105
b6fa137b
FC
106=item a, d, l and u
107X</a> X</d> X</l> X</u>
108
516074bb
KW
109These modifiers, all new in 5.14, affect which character-set semantics
110(Unicode, etc.) are used, as described below in
ed7efc79 111L</Character set modifiers>.
b6fa137b 112
55497cff 113=back
a0d0e21e 114
516074bb
KW
115Regular expression modifiers are usually written in documentation
116as e.g., "the C</x> modifier", even though the delimiter
b6fa137b 117in question might not really be a slash. The modifiers C</imsxadlup>
ab7bb42d 118may also be embedded within the regular expression itself using
ed7efc79 119the C<(?...)> construct, see L</Extended Patterns> below.
b6fa137b 120
ed7efc79
KW
121=head3 /x
122
b6fa137b 123C</x> tells
7b059540 124the regular expression parser to ignore most whitespace that is neither
55497cff 125backslashed nor within a character class. You can use this to break up
4633a7c4 126your regular expression into (slightly) more readable parts. The C<#>
54310121 127character is also treated as a metacharacter introducing a comment,
55497cff 128just as in ordinary Perl code. This also means that if you want real
14218588 129whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a 130class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
131escape them (using backslashes or C<\Q...\E>) or encode them using octal,
132hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
133making Perl's regular expressions more readable. Note that you have to
134be careful not to include the pattern delimiter in the comment--perl has
135no way of knowing you did not intend to close the pattern early. See
136the C-comment deletion code in L<perlop>. Also note that anything inside
7651b971 137a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
0b928c2f 138space interpretation within a single multi-character construct. For
7651b971 139example in C<\x{...}>, regardless of the C</x> modifier, there can be no
9bb1f947 140spaces. Same for a L<quantifier|/Quantifiers> such as C<{3}> or
f9e949fd
KW
141C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<?> and C<:>,
142but can between the C<(> and C<?>. Within any delimiters for such a
143construct, allowed spaces are not affected by C</x>, and depend on the
144construct. For example, C<\x{...}> can't have spaces because hexadecimal
145numbers don't have spaces in them. But, Unicode properties can have spaces, so
0b928c2f 146in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
9bb1f947 147L<perluniprops/Properties accessible through \p{} and \P{}>.
d74e8afc 148X</x>
a0d0e21e 149
ed7efc79
KW
150=head3 Character set modifiers
151
152C</d>, C</u>, C</a>, and C</l>, available starting in 5.14, are called
153the character set modifiers; they affect the character set semantics
154used for the regular expression.
155
808432af
KW
156The C</d>, C</u>, and C</l> modifiers are not likely to be of much use
157to you, and so you need not worry about them very much. They exist for
158Perl's internal use, so that complex regular expression data structures
159can be automatically serialized and later exactly reconstituted,
160including all their nuances. But, since Perl can't keep a secret, and
161there may be rare instances where they are useful, they are documented
162here.
ed7efc79 163
808432af
KW
164The C</a> modifier, on the other hand, may be useful. Its purpose is to
165allow code that is to work mostly on ASCII data to not have to concern
166itself with Unicode.
ca9560b2 167
808432af
KW
168Briefly, C</l> sets the character set to that of whatever B<L>ocale is in
169effect at the time of the execution of the pattern match.
ca9560b2 170
808432af
KW
171C</u> sets the character set to B<U>nicode.
172
173C</a> also sets the character set to Unicode, BUT adds several
174restrictions for B<A>SCII-safe matching.
175
176C</d> is the old, problematic, pre-5.14 B<D>efault character set
177behavior. Its only use is to force that old behavior.
178
179At any given time, exactly one of these modifiers is in effect. Their
180existence allows Perl to keep the originally compiled behavior of a
181regular expression, regardless of what rules are in effect when it is
182actually executed. And if it is interpolated into a larger regex, the
183original's rules continue to apply to it, and only it.
184
185The C</l> and C</u> modifiers are automatically selected for
186regular expressions compiled within the scope of various pragmas,
187and we recommend that in general, you use those pragmas instead of
188specifying these modifiers explicitly. For one thing, the modifiers
189affect only pattern matching, and do not extend to even any replacement
190done, whereas using the pragmas give consistent results for all
191appropriate operations within their scopes. For example,
192
193 s/foo/\Ubar/il
194
195will match "foo" using the locale's rules for case-insensitive matching,
196but the C</l> does not affect how the C<\U> operates. Most likely you
197want both of them to use locale rules. To do this, instead compile the
198regular expression within the scope of C<use locale>. This both
199implicitly adds the C</l> and applies locale rules to the C<\U>. The
200lesson is to C<use locale> and not C</l> explicitly.
201
202Similarly, it would be better to use C<use feature 'unicode_strings'>
203instead of,
204
205 s/foo/\Lbar/iu
206
207to get Unicode rules, as the C<\L> in the former (but not necessarily
208the latter) would also use Unicode rules.
209
210More detail on each of the modifiers follows. Most likely you don't
211need to know this detail for C</l>, C</u>, and C</d>, and can skip ahead
212to L<E<sol>a|/E<sol>a (and E<sol>aa)>.
ca9560b2 213
ed7efc79
KW
214=head4 /l
215
216means to use the current locale's rules (see L<perllocale>) when pattern
217matching. For example, C<\w> will match the "word" characters of that
218locale, and C<"/i"> case-insensitive matching will match according to
219the locale's case folding rules. The locale used will be the one in
220effect at the time of execution of the pattern match. This may not be
221the same as the compilation-time locale, and can differ from one match
222to another if there is an intervening call of the
b6fa137b 223L<setlocale() function|perllocale/The setlocale function>.
ed7efc79
KW
224
225Perl only supports single-byte locales. This means that code points
226above 255 are treated as Unicode no matter what locale is in effect.
227Under Unicode rules, there are a few case-insensitive matches that cross
228the 255/256 boundary. These are disallowed under C</l>. For example,
516074bb
KW
2290xFF (on ASCII platforms) does not caselessly match the character at
2300x178, C<LATIN CAPITAL LETTER Y WITH DIAERESIS>, because 0xFF may not be
231C<LATIN SMALL LETTER Y WITH DIAERESIS> in the current locale, and Perl
232has no way of knowing if that character even exists in the locale, much
233less what code point it is.
ed7efc79
KW
234
235This modifier may be specified to be the default by C<use locale>, but
236see L</Which character set modifier is in effect?>.
b6fa137b
FC
237X</l>
238
ed7efc79
KW
239=head4 /u
240
241means to use Unicode rules when pattern matching. On ASCII platforms,
242this means that the code points between 128 and 255 take on their
808432af
KW
243Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's).
244(Otherwise Perl considers their meanings to be undefined.) Thus,
245under this modifier, the ASCII platform effectively becomes a Unicode
246platform; and hence, for example, C<\w> will match any of the more than
247100_000 word characters in Unicode.
ed7efc79
KW
248
249Unlike most locales, which are specific to a language and country pair,
516074bb
KW
250Unicode classifies all the characters that are letters I<somewhere> in
251the world as
ed7efc79
KW
252C<\w>. For example, your locale might not think that C<LATIN SMALL
253LETTER ETH> is a letter (unless you happen to speak Icelandic), but
254Unicode does. Similarly, all the characters that are decimal digits
255somewhere in the world will match C<\d>; this is hundreds, not 10,
256possible matches. And some of those digits look like some of the 10
257ASCII digits, but mean a different number, so a human could easily think
258a number is a different quantity than it really is. For example,
259C<BENGALI DIGIT FOUR> (U+09EA) looks very much like an
260C<ASCII DIGIT EIGHT> (U+0038). And, C<\d+>, may match strings of digits
261that are a mixture from different writing systems, creating a security
67592e11 262issue. L<Unicode::UCD/num()> can be used to sort
516074bb
KW
263this out. Or the C</a> modifier can be used to force C<\d> to match
264just the ASCII 0 through 9.
ed7efc79 265
516074bb
KW
266Also, under this modifier, case-insensitive matching works on the full
267set of Unicode
ed7efc79
KW
268characters. The C<KELVIN SIGN>, for example matches the letters "k" and
269"K"; and C<LATIN SMALL LIGATURE FF> matches the sequence "ff", which,
270if you're not prepared, might make it look like a hexadecimal constant,
271presenting another potential security issue. See
272L<http://unicode.org/reports/tr36> for a detailed discussion of Unicode
273security issues.
274
6368643f
KW
275On the EBCDIC platforms that Perl handles, the native character set is
276equivalent to Latin-1. Thus this modifier changes behavior only when
ed7efc79
KW
277the C<"/i"> modifier is also specified, and it turns out it affects only
278two characters, giving them full Unicode semantics: the C<MICRO SIGN>
6368643f 279will match the Greek capital and small letters C<MU>, otherwise not; and
ed7efc79
KW
280the C<LATIN CAPITAL LETTER SHARP S> will match any of C<SS>, C<Ss>,
281C<sS>, and C<ss>, otherwise not.
282
283This modifier may be specified to be the default by C<use feature
66cbab2c
KW
284'unicode_strings>, C<use locale ':not_characters'>, or
285C<L<use 5.012|perlfunc/use VERSION>> (or higher),
808432af 286but see L</Which character set modifier is in effect?>.
b6fa137b
FC
287X</u>
288
ed7efc79
KW
289=head4 /d
290
291This modifier means to use the "Default" native rules of the platform
292except when there is cause to use Unicode rules instead, as follows:
293
294=over 4
295
296=item 1
297
298the target string is encoded in UTF-8; or
299
300=item 2
301
302the pattern is encoded in UTF-8; or
303
304=item 3
305
306the pattern explicitly mentions a code point that is above 255 (say by
307C<\x{100}>); or
308
309=item 4
b6fa137b 310
ed7efc79
KW
311the pattern uses a Unicode name (C<\N{...}>); or
312
313=item 5
314
9d1a5160
KW
315the pattern uses a Unicode property (C<\p{...}>); or
316
317=item 6
318
319the pattern uses L</C<(?[ ])>>
ed7efc79
KW
320
321=back
322
323Another mnemonic for this modifier is "Depends", as the rules actually
324used depend on various things, and as a result you can get unexpected
808432af
KW
325results. See L<perlunicode/The "Unicode Bug">. The Unicode Bug has
326become rather infamous, leading to yet another (printable) name for this
327modifier, "Dodgy".
ed7efc79
KW
328
329On ASCII platforms, the native rules are ASCII, and on EBCDIC platforms
330(at least the ones that Perl handles), they are Latin-1.
331
332Here are some examples of how that works on an ASCII platform:
333
334 $str = "\xDF"; # $str is not in UTF-8 format.
335 $str =~ /^\w/; # No match, as $str isn't in UTF-8 format.
336 $str .= "\x{0e0b}"; # Now $str is in UTF-8 format.
337 $str =~ /^\w/; # Match! $str is now in UTF-8 format.
338 chop $str;
339 $str =~ /^\w/; # Still a match! $str remains in UTF-8 format.
340
808432af
KW
341This modifier is automatically selected by default when none of the
342others are, so yet another name for it is "Default".
343
344Because of the unexpected behaviors associated with this modifier, you
345probably should only use it to maintain weird backward compatibilities.
346
347=head4 /a (and /aa)
348
349This modifier stands for ASCII-restrict (or ASCII-safe). This modifier,
350unlike the others, may be doubled-up to increase its effect.
351
352When it appears singly, it causes the sequences C<\d>, C<\s>, C<\w>, and
353the Posix character classes to match only in the ASCII range. They thus
354revert to their pre-5.6, pre-Unicode meanings. Under C</a>, C<\d>
355always means precisely the digits C<"0"> to C<"9">; C<\s> means the five
356characters C<[ \f\n\r\t]>; C<\w> means the 63 characters
357C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
358C<[[:print:]]> match only the appropriate ASCII-range characters.
359
360This modifier is useful for people who only incidentally use Unicode,
361and who do not wish to be burdened with its complexities and security
362concerns.
363
364With C</a>, one can write C<\d> with confidence that it will only match
365ASCII characters, and should the need arise to match beyond ASCII, you
366can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>). There are
367similar C<\p{...}> constructs that can match beyond ASCII both white
368space (see L<perlrecharclass/Whitespace>), and Posix classes (see
369L<perlrecharclass/POSIX Character Classes>). Thus, this modifier
370doesn't mean you can't use Unicode, it means that to get Unicode
371matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
372signals Unicode.
373
374As you would expect, this modifier causes, for example, C<\D> to mean
375the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
376C<\D>, C<\S>, and C<\W>. C<\b> still means to match at the boundary
377between C<\w> and C<\W>, using the C</a> definitions of them (similarly
378for C<\B>).
379
380Otherwise, C</a> behaves like the C</u> modifier, in that
381case-insensitive matching uses Unicode semantics; for example, "k" will
382match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
383points in the Latin1 range, above ASCII will have Unicode rules when it
384comes to case-insensitive matching.
385
386To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
387specify the "a" twice, for example C</aai> or C</aia>. (The first
388occurrence of "a" restricts the C<\d>, etc., and the second occurrence
389adds the C</i> restrictions.) But, note that code points outside the
390ASCII range will use Unicode rules for C</i> matching, so the modifier
391doesn't really restrict things to just ASCII; it just forbids the
392intermixing of ASCII and non-ASCII.
393
394To summarize, this modifier provides protection for applications that
395don't wish to be exposed to all of Unicode. Specifying it twice
396gives added protection.
397
398This modifier may be specified to be the default by C<use re '/a'>
399or C<use re '/aa'>. If you do so, you may actually have occasion to use
400the C</u> modifier explictly if there are a few regular expressions
401where you do want full Unicode rules (but even here, it's best if
402everything were under feature C<"unicode_strings">, along with the
403C<use re '/aa'>). Also see L</Which character set modifier is in
404effect?>.
405X</a>
406X</aa>
407
ed7efc79
KW
408=head4 Which character set modifier is in effect?
409
410Which of these modifiers is in effect at any given point in a regular
808432af
KW
411expression depends on a fairly complex set of interactions. These have
412been designed so that in general you don't have to worry about it, but
413this section gives the gory details. As
ed7efc79
KW
414explained below in L</Extended Patterns> it is possible to explicitly
415specify modifiers that apply only to portions of a regular expression.
416The innermost always has priority over any outer ones, and one applying
6368643f
KW
417to the whole expression has priority over any of the default settings that are
418described in the remainder of this section.
ed7efc79 419
916cec3f 420The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
ed7efc79
KW
421default modifiers (including these) for regular expressions compiled
422within its scope. This pragma has precedence over the other pragmas
516074bb 423listed below that also change the defaults.
ed7efc79
KW
424
425Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
66cbab2c 426and C<L<use feature 'unicode_strings|feature>>, or
ed7efc79
KW
427C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
428C</u> when not in the same scope as either C<L<use locale|perllocale>>
66cbab2c
KW
429or C<L<use bytes|bytes>>.
430(C<L<use locale ':not_characters'|perllocale/Unicode and UTF-8>> also
431sets the default to C</u>, overriding any plain C<use locale>.)
432Unlike the mechanisms mentioned above, these
6368643f
KW
433affect operations besides regular expressions pattern matching, and so
434give more consistent results with other operators, including using
435C<\U>, C<\l>, etc. in substitution replacements.
ed7efc79
KW
436
437If none of the above apply, for backwards compatibility reasons, the
438C</d> modifier is the one in effect by default. As this can lead to
439unexpected results, it is best to specify which other rule set should be
440used.
441
442=head4 Character set modifier behavior prior to Perl 5.14
443
444Prior to 5.14, there were no explicit modifiers, but C</l> was implied
445for regexes compiled within the scope of C<use locale>, and C</d> was
446implied otherwise. However, interpolating a regex into a larger regex
447would ignore the original compilation in favor of whatever was in effect
448at the time of the second compilation. There were a number of
449inconsistencies (bugs) with the C</d> modifier, where Unicode rules
450would be used when inappropriate, and vice versa. C<\p{}> did not imply
451Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
b6fa137b 452
a0d0e21e
LW
453=head2 Regular Expressions
454
04838cea
RGS
455=head3 Metacharacters
456
384f06ae 457The patterns used in Perl pattern matching evolved from those supplied in
14218588 458the Version 8 regex routines. (The routines are derived
19799a22
GS
459(distantly) from Henry Spencer's freely redistributable reimplementation
460of the V8 routines.) See L<Version 8 Regular Expressions> for
461details.
a0d0e21e
LW
462
463In particular the following metacharacters have their standard I<egrep>-ish
464meanings:
d74e8afc
ITB
465X<metacharacter>
466X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
467
a0d0e21e 468
f793d64a
KW
469 \ Quote the next metacharacter
470 ^ Match the beginning of the line
471 . Match any character (except newline)
472 $ Match the end of the line (or before newline at the end)
473 | Alternation
474 () Grouping
475 [] Bracketed Character class
a0d0e21e 476
14218588
GS
477By default, the "^" character is guaranteed to match only the
478beginning of the string, the "$" character only the end (or before the
479newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
480assumption that the string contains only one line. Embedded newlines
481will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 482string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
483newline within the string (except if the newline is the last character in
484the string), and "$" will match before any newline. At the
a0d0e21e
LW
485cost of a little more overhead, you can do this by using the /m modifier
486on the pattern match operator. (Older programs did this by setting C<$*>,
db7cd43a 487but this option was removed in perl 5.10.)
d74e8afc 488X<^> X<$> X</m>
a0d0e21e 489
14218588 490To simplify multi-line substitutions, the "." character never matches a
55497cff 491newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 492the string is a single line--even if it isn't.
d74e8afc 493X<.> X</s>
a0d0e21e 494
04838cea
RGS
495=head3 Quantifiers
496
a0d0e21e 497The following standard quantifiers are recognized:
d74e8afc 498X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 499
f793d64a
KW
500 * Match 0 or more times
501 + Match 1 or more times
502 ? Match 1 or 0 times
503 {n} Match exactly n times
504 {n,} Match at least n times
505 {n,m} Match at least n but not more than m times
a0d0e21e 506
0b928c2f
FC
507(If a curly bracket occurs in any other context and does not form part of
508a backslashed sequence like C<\x{...}>, it is treated
9af81bfe
KW
509as a regular character. In particular, the lower quantifier bound
510is not optional. However, in Perl v5.18, it is planned to issue a
511deprecation warning for all such occurrences, and in Perl v5.20 to
512require literal uses of a curly bracket to be escaped, say by preceding
513them with a backslash or enclosing them within square brackets, (C<"\{">
514or C<"[{]">). This change will allow for future syntax extensions (like
515making the lower bound of a quantifier optional), and better error
516checking of quantifiers. Now, a typo in a quantifier silently causes
517it to be treated as the literal characters. For example,
518
519 /o{4,3}/
520
521looks like a quantifier that matches 0 times, since 4 is greater than 3,
522but it really means to match the sequence of six characters
523S<C<"o { 4 , 3 }">>.)
524
525The "*" quantifier is equivalent to C<{0,}>, the "+"
527e91da 526quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 527to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
528This is usually 32766 on the most common platforms. The actual limit can
529be seen in the error message generated by code such as this:
530
820475bd 531 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 532
54310121
PP
533By default, a quantified subpattern is "greedy", that is, it will match as
534many times as possible (given a particular starting location) while still
535allowing the rest of the pattern to match. If you want it to match the
536minimum number of times possible, follow the quantifier with a "?". Note
537that the meanings don't change, just the "greediness":
0d017f4d 538X<metacharacter> X<greedy> X<greediness>
d74e8afc 539X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 540
f793d64a
KW
541 *? Match 0 or more times, not greedily
542 +? Match 1 or more times, not greedily
543 ?? Match 0 or 1 time, not greedily
0b928c2f 544 {n}? Match exactly n times, not greedily (redundant)
f793d64a
KW
545 {n,}? Match at least n times, not greedily
546 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 547
b9b4dddf
YO
548By default, when a quantified subpattern does not allow the rest of the
549overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 550sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
551as well.
552
f793d64a
KW
553 *+ Match 0 or more times and give nothing back
554 ++ Match 1 or more times and give nothing back
555 ?+ Match 0 or 1 time and give nothing back
556 {n}+ Match exactly n times and give nothing back (redundant)
557 {n,}+ Match at least n times and give nothing back
558 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
559
560For instance,
561
562 'aaaa' =~ /a++a/
563
564will never match, as the C<a++> will gobble up all the C<a>'s in the
565string and won't leave any for the remaining part of the pattern. This
566feature can be extremely useful to give perl hints about where it
567shouldn't backtrack. For instance, the typical "match a double-quoted
568string" problem can be most efficiently performed when written as:
569
570 /"(?:[^"\\]++|\\.)*+"/
571
0d017f4d 572as we know that if the final quote does not match, backtracking will not
0b928c2f
FC
573help. See the independent subexpression
574L</C<< (?>pattern) >>> for more details;
b9b4dddf
YO
575possessive quantifiers are just syntactic sugar for that construct. For
576instance the above example could also be written as follows:
577
578 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
579
04838cea
RGS
580=head3 Escape sequences
581
0b928c2f 582Because patterns are processed as double-quoted strings, the following
a0d0e21e
LW
583also work:
584
f793d64a
KW
585 \t tab (HT, TAB)
586 \n newline (LF, NL)
587 \r return (CR)
588 \f form feed (FF)
589 \a alarm (bell) (BEL)
590 \e escape (think troff) (ESC)
f793d64a 591 \cK control char (example: VT)
dc0d9c48 592 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 593 \N{name} named Unicode character or character sequence
f793d64a 594 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 595 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
596 \l lowercase next char (think vi)
597 \u uppercase next char (think vi)
598 \L lowercase till \E (think vi)
599 \U uppercase till \E (think vi)
600 \Q quote (disable) pattern metacharacters till \E
601 \E end either case modification or quoted section, think vi
a0d0e21e 602
9bb1f947 603Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 604
e1d1eefb 605=head3 Character Classes and other Special Escapes
04838cea 606
a0d0e21e 607In addition, Perl defines the following:
d0b16107 608X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 609
f793d64a
KW
610 Sequence Note Description
611 [...] [1] Match a character according to the rules of the
612 bracketed character class defined by the "...".
613 Example: [a-z] matches "a" or "b" or "c" ... or "z"
614 [[:...:]] [2] Match a character according to the rules of the POSIX
615 character class "..." within the outer bracketed
616 character class. Example: [[:upper:]] matches any
617 uppercase character.
d35dd6c6
KW
618 \w [3] Match a "word" character (alphanumeric plus "_", plus
619 other connector punctuation chars plus Unicode
0b928c2f 620 marks)
f793d64a
KW
621 \W [3] Match a non-"word" character
622 \s [3] Match a whitespace character
623 \S [3] Match a non-whitespace character
624 \d [3] Match a decimal digit character
625 \D [3] Match a non-digit character
626 \pP [3] Match P, named property. Use \p{Prop} for longer names
627 \PP [3] Match non-P
628 \X [4] Match Unicode "eXtended grapheme cluster"
629 \C Match a single C-language char (octet) even if that is
630 part of a larger UTF-8 character. Thus it breaks up
631 characters into their UTF-8 bytes, so you may end up
632 with malformed pieces of UTF-8. Unsupported in
633 lookbehind.
c27a5cfe 634 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
635 '1' may actually be any positive integer.
636 \g1 [5] Backreference to a specific or previous group,
637 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 638 previous group and may optionally be wrapped in
f793d64a
KW
639 curly brackets for safer parsing.
640 \g{name} [5] Named backreference
641 \k<name> [5] Named backreference
642 \K [6] Keep the stuff left of the \K, don't include it in $&
643 \N [7] Any character but \n (experimental). Not affected by
644 /s modifier
645 \v [3] Vertical whitespace
646 \V [3] Not vertical whitespace
647 \h [3] Horizontal whitespace
648 \H [3] Not horizontal whitespace
649 \R [4] Linebreak
e1d1eefb 650
9bb1f947
KW
651=over 4
652
653=item [1]
654
655See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 656
9bb1f947 657=item [2]
b8c5462f 658
9bb1f947 659See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 660
9bb1f947 661=item [3]
5496314a 662
9bb1f947 663See L<perlrecharclass/Backslash sequences> for details.
5496314a 664
9bb1f947 665=item [4]
5496314a 666
9bb1f947 667See L<perlrebackslash/Misc> for details.
d0b16107 668
9bb1f947 669=item [5]
b8c5462f 670
c27a5cfe 671See L</Capture groups> below for details.
93733859 672
9bb1f947 673=item [6]
b8c5462f 674
9bb1f947
KW
675See L</Extended Patterns> below for details.
676
677=item [7]
678
679Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
680character or character sequence whose name is C<NAME>; and similarly
681when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
682code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947
KW
683
684=back
d0b16107 685
04838cea
RGS
686=head3 Assertions
687
a0d0e21e 688Perl defines the following zero-width assertions:
d74e8afc
ITB
689X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
690X<regexp, zero-width assertion>
691X<regular expression, zero-width assertion>
692X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 693
9bb1f947
KW
694 \b Match a word boundary
695 \B Match except at a word boundary
696 \A Match only at beginning of string
697 \Z Match only at end of string, or before newline at the end
698 \z Match only at end of string
699 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 700 of prior m//g)
a0d0e21e 701
14218588 702A word boundary (C<\b>) is a spot between two characters
19799a22
GS
703that has a C<\w> on one side of it and a C<\W> on the other side
704of it (in either order), counting the imaginary characters off the
705beginning and end of the string as matching a C<\W>. (Within
706character classes C<\b> represents backspace rather than a word
707boundary, just as it normally does in any double-quoted string.)
708The C<\A> and C<\Z> are just like "^" and "$", except that they
709won't match multiple times when the C</m> modifier is used, while
710"^" and "$" will match at every internal line boundary. To match
711the actual end of the string and not ignore an optional trailing
712newline, use C<\z>.
d74e8afc 713X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
714
715The C<\G> assertion can be used to chain global matches (using
716C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
717It is also useful when writing C<lex>-like scanners, when you have
718several patterns that you want to match against consequent substrings
0b928c2f 719of your string; see the previous reference. The actual location
19799a22 720where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d 721an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
0b928c2f
FC
722matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
723is modified somewhat, in that contents to the left of C<\G> are
58e23c8d
YO
724not counted when determining the length of the match. Thus the following
725will not match forever:
d74e8afc 726X<\G>
c47ff5f1 727
e761bb84
CO
728 my $string = 'ABC';
729 pos($string) = 1;
730 while ($string =~ /(.\G)/g) {
731 print $1;
732 }
58e23c8d
YO
733
734It will print 'A' and then terminate, as it considers the match to
735be zero-width, and thus will not match at the same position twice in a
736row.
737
738It is worth noting that C<\G> improperly used can result in an infinite
739loop. Take care when using patterns that include C<\G> in an alternation.
740
c27a5cfe 741=head3 Capture groups
04838cea 742
c27a5cfe
KW
743The bracketing construct C<( ... )> creates capture groups (also referred to as
744capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
745the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
746for the second, and so on.
747This is called a I<backreference>.
d74e8afc 748X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 749X<regex, capture group> X<regexp, capture group>
d74e8afc 750X<regular expression, capture buffer> X<backreference>
c27a5cfe 751X<regular expression, capture group> X<backreference>
1f1031fe 752X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
753X<named capture buffer> X<regular expression, named capture buffer>
754X<named capture group> X<regular expression, named capture group>
755X<%+> X<$+{name}> X<< \k<name> >>
756There is no limit to the number of captured substrings that you may use.
757Groups are numbered with the leftmost open parenthesis being number 1, etc. If
758a group did not match, the associated backreference won't match either. (This
759can happen if the group is optional, or in a different branch of an
760alternation.)
761You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
762this form, described below.
763
764You can also refer to capture groups relatively, by using a negative number, so
765that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
766group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
767example:
5624f11d
YO
768
769 /
c27a5cfe
KW
770 (Y) # group 1
771 ( # group 2
772 (X) # group 3
773 \g{-1} # backref to group 3
774 \g{-3} # backref to group 1
5624f11d
YO
775 )
776 /x
777
d8b950dc
KW
778would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
779interpolate regexes into larger regexes and not have to worry about the
780capture groups being renumbered.
781
782You can dispense with numbers altogether and create named capture groups.
783The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
784reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
785also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
786I<name> must not begin with a number, nor contain hyphens.
787When different groups within the same pattern have the same name, any reference
788to that name assumes the leftmost defined group. Named groups count in
789absolute and relative numbering, and so can also be referred to by those
790numbers.
791(It's possible to do things with named capture groups that would otherwise
792require C<(??{})>.)
793
794Capture group contents are dynamically scoped and available to you outside the
795pattern until the end of the enclosing block or until the next successful
796match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
797You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
798etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
799
800Braces are required in referring to named capture groups, but are optional for
801absolute or relative numbered ones. Braces are safer when creating a regex by
802concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
803contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
804is probably not what you intended.
805
806The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
807there were no named nor relative numbered capture groups. Absolute numbered
0b928c2f
FC
808groups were referred to using C<\1>,
809C<\2>, etc., and this notation is still
d8b950dc
KW
810accepted (and likely always will be). But it leads to some ambiguities if
811there are more than 9 capture groups, as C<\10> could mean either the tenth
812capture group, or the character whose ordinal in octal is 010 (a backspace in
813ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
814only if at least 10 left parentheses have opened before it. Likewise C<\11> is
815a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
816And so on. C<\1> through C<\9> are always interpreted as backreferences.
817There are several examples below that illustrate these perils. You can avoid
818the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
819and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
820digits padded with leading zeros, since a leading zero implies an octal
821constant.
d8b950dc
KW
822
823The C<\I<digit>> notation also works in certain circumstances outside
ed7efc79 824the pattern. See L</Warning on \1 Instead of $1> below for details.
81714fb9 825
14218588 826Examples:
a0d0e21e
LW
827
828 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
829
d8b950dc 830 /(.)\g1/ # find first doubled char
81714fb9
YO
831 and print "'$1' is the first doubled character\n";
832
833 /(?<char>.)\k<char>/ # ... a different way
834 and print "'$+{char}' is the first doubled character\n";
835
d8b950dc 836 /(?'char'.)\g1/ # ... mix and match
81714fb9 837 and print "'$1' is the first doubled character\n";
c47ff5f1 838
14218588 839 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
840 $hours = $1;
841 $minutes = $2;
842 $seconds = $3;
a0d0e21e 843 }
c47ff5f1 844
9d860678
KW
845 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
846 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
847 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
848 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
849
850 $a = '(.)\1'; # Creates problems when concatenated.
851 $b = '(.)\g{1}'; # Avoids the problems.
852 "aa" =~ /${a}/; # True
853 "aa" =~ /${b}/; # True
854 "aa0" =~ /${a}0/; # False!
855 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
856 "aa\x08" =~ /${a}0/; # True!
857 "aa\x08" =~ /${b}0/; # False
9d860678 858
14218588
GS
859Several special variables also refer back to portions of the previous
860match. C<$+> returns whatever the last bracket match matched.
861C<$&> returns the entire matched string. (At one point C<$0> did
862also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
863everything before the matched string. C<$'> returns everything
864after the matched string. And C<$^N> contains whatever was matched by
865the most-recently closed group (submatch). C<$^N> can be used in
866extended patterns (see below), for example to assign a submatch to a
81714fb9 867variable.
d74e8afc 868X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 869
d8b950dc
KW
870These special variables, like the C<%+> hash and the numbered match variables
871(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
872until the end of the enclosing block or until the next successful
873match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
874X<$+> X<$^N> X<$&> X<$`> X<$'>
875X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
876
0d017f4d 877B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 878which makes it easier to write code that tests for a series of more
665e98b9
JH
879specific cases and remembers the best match.
880
d78f32f6
FC
881B<WARNING>: If your code is to run on Perl 5.16 or earlier,
882beware that once Perl sees that you need one of C<$&>, C<$`>, or
14218588 883C<$'> anywhere in the program, it has to provide them for every
d78f32f6
FC
884pattern match. This may substantially slow your program. (In Perl 5.18 a
885more efficient mechanism is used, eliminating any slowdown.) Perl
d8b950dc 886uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
887price for each pattern that contains capturing parentheses. (To
888avoid this cost while retaining the grouping behaviour, use the
889extended regular expression C<(?: ... )> instead.) But if you never
890use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
891parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
892if you can, but if you can't (and some algorithms really appreciate
893them), once you've used them once, use them at will, because you've
d78f32f6 894already paid the price.
d74e8afc 895X<$&> X<$`> X<$'>
68dc0745 896
d78f32f6 897As a workaround for this problem, Perl 5.10.0 introduced C<${^PREMATCH}>,
cde0cee5
YO
898C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
899and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 900successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
901The use of these variables incurs no global performance penalty, unlike
902their punctuation char equivalents, however at the trade-off that you
d78f32f6
FC
903have to tell perl when you want to use them. As of Perl 5.18, these three
904variables are equivalent to C<$`>, C<$&> and C<$'>, and C</p> is ignored.
87e95b7f 905X</p> X<p modifier>
cde0cee5 906
9d727203
KW
907=head2 Quoting metacharacters
908
19799a22
GS
909Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
910C<\w>, C<\n>. Unlike some other regular expression languages, there
911are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 912that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
913interpreted as a literal character, not a metacharacter. This was
914once used in a common idiom to disable or quote the special meanings
915of regular expression metacharacters in a string that you want to
36bbe248 916use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
917
918 $pattern =~ s/(\W)/\\$1/g;
919
f1cbbd6e 920(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
921Today it is more common to use the quotemeta() function or the C<\Q>
922metaquoting escape sequence to disable all metacharacters' special
923meanings like this:
a0d0e21e
LW
924
925 /$unquoted\Q$quoted\E$unquoted/
926
9da458fc
IZ
927Beware that if you put literal backslashes (those not inside
928interpolated variables) between C<\Q> and C<\E>, double-quotish
929backslash interpolation may lead to confusing results. If you
930I<need> to use literal backslashes within C<\Q...\E>,
931consult L<perlop/"Gory details of parsing quoted constructs">.
932
736fe711
KW
933C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
934
19799a22
GS
935=head2 Extended Patterns
936
14218588 937Perl also defines a consistent extension syntax for features not
0b928c2f
FC
938found in standard tools like B<awk> and
939B<lex>. The syntax for most of these is a
14218588
GS
940pair of parentheses with a question mark as the first thing within
941the parentheses. The character after the question mark indicates
942the extension.
19799a22 943
14218588
GS
944The stability of these extensions varies widely. Some have been
945part of the core language for many years. Others are experimental
946and may change without warning or be completely removed. Check
947the documentation on an individual feature to verify its current
948status.
19799a22 949
14218588
GS
950A question mark was chosen for this and for the minimal-matching
951construct because 1) question marks are rare in older regular
952expressions, and 2) whenever you see one, you should stop and
0b928c2f 953"question" exactly what is going on. That's psychology....
a0d0e21e 954
70ca8714 955=over 4
a0d0e21e 956
cc6b7395 957=item C<(?#text)>
d74e8afc 958X<(?#)>
a0d0e21e 959
14218588 960A comment. The text is ignored. If the C</x> modifier enables
19799a22 961whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
962the comment as soon as it sees a C<)>, so there is no way to put a literal
963C<)> in the comment.
a0d0e21e 964
cfaf538b 965=item C<(?adlupimsx-imsx)>
fb85c044 966
cfaf538b 967=item C<(?^alupimsx)>
fb85c044 968X<(?)> X<(?^)>
19799a22 969
0b6d1084
JH
970One or more embedded pattern-match modifiers, to be turned on (or
971turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
972the remainder of the enclosing pattern group (if any).
973
fb85c044 974This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 975configuration file, taken from an argument, or specified in a table
0b928c2f
FC
976somewhere. Consider the case where some patterns want to be
977case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 978include C<(?i)> at the front of the pattern. For example:
19799a22
GS
979
980 $pattern = "foobar";
5d458dd8 981 if ( /$pattern/i ) { }
19799a22
GS
982
983 # more flexible:
984
985 $pattern = "(?i)foobar";
5d458dd8 986 if ( /$pattern/ ) { }
19799a22 987
0b6d1084 988These modifiers are restored at the end of the enclosing group. For example,
19799a22 989
d8b950dc 990 ( (?i) blah ) \s+ \g1
19799a22 991
0d017f4d
WL
992will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
993repetition of the previous word, assuming the C</x> modifier, and no C</i>
994modifier outside this group.
19799a22 995
8eb5594e 996These modifiers do not carry over into named subpatterns called in the
dd72e27b 997enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
8eb5594e
DR
998change the case-sensitivity of the "NAME" pattern.
999
dc925305
KW
1000Any of these modifiers can be set to apply globally to all regular
1001expressions compiled within the scope of a C<use re>. See
a0bbd6ff 1002L<re/"'/flags' mode">.
dc925305 1003
9de15fec
KW
1004Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1005after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
1006C<"d">) may follow the caret to override it.
1007But a minus sign is not legal with it.
1008
dc925305 1009Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 1010that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 1011C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
1012others, and a maximum of one (or two C<a>'s) may appear in the
1013construct. Thus, for
0b928c2f 1014example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 1015C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1016
1017Note also that the C<p> modifier is special in that its presence
1018anywhere in a pattern has a global effect.
cde0cee5 1019
5a964f20 1020=item C<(?:pattern)>
d74e8afc 1021X<(?:)>
a0d0e21e 1022
cfaf538b 1023=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1024
cfaf538b 1025=item C<(?^aluimsx:pattern)>
fb85c044
KW
1026X<(?^:)>
1027
5a964f20
TC
1028This is for clustering, not capturing; it groups subexpressions like
1029"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1030
5a964f20 1031 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1032
1033is like
1034
5a964f20 1035 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1036
19799a22
GS
1037but doesn't spit out extra fields. It's also cheaper not to capture
1038characters if you don't need to.
a0d0e21e 1039
19799a22 1040Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1041C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1042
1043 /(?s-i:more.*than).*million/i
1044
14218588 1045is equivalent to the more verbose
ca9dfc88
IZ
1046
1047 /(?:(?s-i)more.*than).*million/i
1048
fb85c044 1049Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1050after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1051flags (except C<"d">) may follow the caret, so
fb85c044
KW
1052
1053 (?^x:foo)
1054
1055is equivalent to
1056
1057 (?x-ims:foo)
1058
1059The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1060surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1061modified by any flags specified.
1062
1063The caret allows for simpler stringification of compiled regular
1064expressions. These look like
1065
1066 (?^:pattern)
1067
1068with any non-default flags appearing between the caret and the colon.
1069A test that looks at such stringification thus doesn't need to have the
1070system default flags hard-coded in it, just the caret. If new flags are
1071added to Perl, the meaning of the caret's expansion will change to include
1072the default for those flags, so the test will still work, unchanged.
1073
1074Specifying a negative flag after the caret is an error, as the flag is
1075redundant.
1076
1077Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1078to match at the beginning.
1079
594d7033
YO
1080=item C<(?|pattern)>
1081X<(?|)> X<Branch reset>
1082
1083This is the "branch reset" pattern, which has the special property
c27a5cfe 1084that the capture groups are numbered from the same starting point
99d59c4d 1085in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1086
c27a5cfe 1087Capture groups are numbered from left to right, but inside this
693596a8 1088construct the numbering is restarted for each branch.
4deaaa80 1089
c27a5cfe 1090The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1091following this construct will be numbered as though the construct
1092contained only one branch, that being the one with the most capture
c27a5cfe 1093groups in it.
4deaaa80 1094
0b928c2f 1095This construct is useful when you want to capture one of a
4deaaa80
PJ
1096number of alternative matches.
1097
1098Consider the following pattern. The numbers underneath show in
c27a5cfe 1099which group the captured content will be stored.
594d7033
YO
1100
1101
1102 # before ---------------branch-reset----------- after
1103 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1104 # 1 2 2 3 2 3 4
1105
ab106183
A
1106Be careful when using the branch reset pattern in combination with
1107named captures. Named captures are implemented as being aliases to
c27a5cfe 1108numbered groups holding the captures, and that interferes with the
ab106183
A
1109implementation of the branch reset pattern. If you are using named
1110captures in a branch reset pattern, it's best to use the same names,
1111in the same order, in each of the alternations:
1112
1113 /(?| (?<a> x ) (?<b> y )
1114 | (?<a> z ) (?<b> w )) /x
1115
1116Not doing so may lead to surprises:
1117
1118 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1119 say $+ {a}; # Prints '12'
1120 say $+ {b}; # *Also* prints '12'.
1121
c27a5cfe
KW
1122The problem here is that both the group named C<< a >> and the group
1123named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1124
ee9b8eae
YO
1125=item Look-Around Assertions
1126X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1127
0b928c2f 1128Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1129pattern without including it in C<$&>. Positive assertions match when
1130their subpattern matches, negative assertions match when their subpattern
1131fails. Look-behind matches text up to the current match position,
1132look-ahead matches text following the current match position.
1133
1134=over 4
1135
5a964f20 1136=item C<(?=pattern)>
d74e8afc 1137X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1138
19799a22 1139A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1140matches a word followed by a tab, without including the tab in C<$&>.
1141
5a964f20 1142=item C<(?!pattern)>
d74e8afc 1143X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1144
19799a22 1145A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1146matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1147however that look-ahead and look-behind are NOT the same thing. You cannot
1148use this for look-behind.
7b8d334a 1149
5a964f20 1150If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1151will not do what you want. That's because the C<(?!foo)> is just saying that
1152the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1153match. Use look-behind instead (see below).
c277df42 1154
ee9b8eae
YO
1155=item C<(?<=pattern)> C<\K>
1156X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1157
c47ff5f1 1158A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1159matches a word that follows a tab, without including the tab in C<$&>.
1160Works only for fixed-width look-behind.
c277df42 1161
ee9b8eae
YO
1162There is a special form of this construct, called C<\K>, which causes the
1163regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1164not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1165look-behind. The use of C<\K> inside of another look-around assertion
1166is allowed, but the behaviour is currently not well defined.
1167
c62285ac 1168For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1169equivalent C<< (?<=...) >> construct, and it is especially useful in
1170situations where you want to efficiently remove something following
1171something else in a string. For instance
1172
1173 s/(foo)bar/$1/g;
1174
1175can be rewritten as the much more efficient
1176
1177 s/foo\Kbar//g;
1178
5a964f20 1179=item C<(?<!pattern)>
d74e8afc 1180X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1181
19799a22
GS
1182A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1183matches any occurrence of "foo" that does not follow "bar". Works
1184only for fixed-width look-behind.
c277df42 1185
ee9b8eae
YO
1186=back
1187
81714fb9
YO
1188=item C<(?'NAME'pattern)>
1189
1190=item C<< (?<NAME>pattern) >>
1191X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1192
c27a5cfe 1193A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1194parentheses C<()> but for the additional fact that the group
1195can be referred to by name in various regular expression
1196constructs (like C<\g{NAME}>) and can be accessed by name
1197after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1198for more details on the C<%+> and C<%-> hashes.
81714fb9 1199
c27a5cfe
KW
1200If multiple distinct capture groups have the same name then the
1201$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1202
0d017f4d 1203The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1204
1205B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1206function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1207numbered sequentially regardless of being named or not. Thus in the
1208pattern
1209
1210 /(x)(?<foo>y)(z)/
1211
1212$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1213the opposite which is what a .NET regex hacker might expect.
1214
1f1031fe
YO
1215Currently NAME is restricted to simple identifiers only.
1216In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1217its Unicode extension (see L<utf8>),
1218though it isn't extended by the locale (see L<perllocale>).
81714fb9 1219
1f1031fe 1220B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1221with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1222may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1223support the use of single quotes as a delimiter for the name.
81714fb9 1224
1f1031fe
YO
1225=item C<< \k<NAME> >>
1226
1227=item C<< \k'NAME' >>
81714fb9
YO
1228
1229Named backreference. Similar to numeric backreferences, except that
1230the group is designated by name and not number. If multiple groups
1231have the same name then it refers to the leftmost defined group in
1232the current match.
1233
0d017f4d 1234It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1235earlier in the pattern.
1236
1237Both forms are equivalent.
1238
1f1031fe 1239B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1240with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1241may be used instead of C<< \k<NAME> >>.
1f1031fe 1242
cc6b7395 1243=item C<(?{ code })>
d74e8afc 1244X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1245
19799a22 1246B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1247experimental, and may be changed without notice. Code executed that
1248has side effects may not perform identically from version to version
e128ab2c
DM
1249due to the effect of future optimisations in the regex engine. The
1250implementation of this feature was radically overhauled for the 5.18.0
1251release, and its behaviour in earlier versions of perl was much buggier,
1252especially in relation to parsing, lexical vars, scoping, recursion and
1253reentrancy.
c277df42 1254
e128ab2c
DM
1255This zero-width assertion executes any embedded Perl code. It always
1256succeeds, and its return value is set as C<$^R>.
19799a22 1257
e128ab2c
DM
1258In literal patterns, the code is parsed at the same time as the
1259surrounding code. While within the pattern, control is passed temporarily
1260back to the perl parser, until the logically-balancing closing brace is
1261encountered. This is similar to the way that an array index expression in
1262a literal string is handled, for example
77ea4f6d 1263
e128ab2c
DM
1264 "abc$array[ 1 + f('[') + g()]def"
1265
1266In particular, braces do not need to be balanced:
1267
576fa024 1268 s/abc(?{ f('{'); })/def/
e128ab2c
DM
1269
1270Even in a pattern that is interpolated and compiled at run-time, literal
1271code blocks will be compiled once, at perl compile time; the following
1272prints "ABCD":
1273
1274 print "D";
1275 my $qr = qr/(?{ BEGIN { print "A" } })/;
1276 my $foo = "foo";
1277 /$foo$qr(?{ BEGIN { print "B" } })/;
1278 BEGIN { print "C" }
1279
1280In patterns where the text of the code is derived from run-time
1281information rather than appearing literally in a source code /pattern/,
1282the code is compiled at the same time that the pattern is compiled, and
5771dda0 1283for reasons of security, C<use re 'eval'> must be in scope. This is to
e128ab2c
DM
1284stop user-supplied patterns containing code snippets from being
1285executable.
1286
5771dda0 1287In situations where you need to enable this with C<use re 'eval'>, you should
e128ab2c
DM
1288also have taint checking enabled. Better yet, use the carefully
1289constrained evaluation within a Safe compartment. See L<perlsec> for
1290details about both these mechanisms.
1291
1292From the viewpoint of parsing, lexical variable scope and closures,
1293
1294 /AAA(?{ BBB })CCC/
1295
1296behaves approximately like
1297
1298 /AAA/ && do { BBB } && /CCC/
1299
1300Similarly,
1301
1302 qr/AAA(?{ BBB })CCC/
1303
1304behaves approximately like
77ea4f6d 1305
e128ab2c
DM
1306 sub { /AAA/ && do { BBB } && /CCC/ }
1307
1308In particular:
1309
1310 { my $i = 1; $r = qr/(?{ print $i })/ }
1311 my $i = 2;
1312 /$r/; # prints "1"
1313
1314Inside a C<(?{...})> block, C<$_> refers to the string the regular
754091cb 1315expression is matching against. You can also use C<pos()> to know what is
fa11829f 1316the current position of matching within this string.
754091cb 1317
e128ab2c
DM
1318The code block introduces a new scope from the perspective of lexical
1319variable declarations, but B<not> from the perspective of C<local> and
1320similar localizing behaviours. So later code blocks within the same
1321pattern will still see the values which were localized in earlier blocks.
1322These accumulated localizations are undone either at the end of a
1323successful match, or if the assertion is backtracked (compare
1324L<"Backtracking">). For example,
b9ac3b5b
GS
1325
1326 $_ = 'a' x 8;
5d458dd8 1327 m<
d1fbf752 1328 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1329 (
5d458dd8 1330 a
b9ac3b5b 1331 (?{
d1fbf752
KW
1332 local $cnt = $cnt + 1; # Update $cnt,
1333 # backtracking-safe.
b9ac3b5b 1334 })
5d458dd8 1335 )*
b9ac3b5b 1336 aaaa
d1fbf752
KW
1337 (?{ $res = $cnt }) # On success copy to
1338 # non-localized location.
b9ac3b5b
GS
1339 >x;
1340
e128ab2c
DM
1341will initially increment C<$cnt> up to 8; then during backtracking, its
1342value will be unwound back to 4, which is the value assigned to C<$res>.
1343At the end of the regex execution, $cnt will be wound back to its initial
1344value of 0.
1345
1346This assertion may be used as the condition in a
b9ac3b5b 1347
e128ab2c
DM
1348 (?(condition)yes-pattern|no-pattern)
1349
1350switch. If I<not> used in this way, the result of evaluation of C<code>
1351is put into the special variable C<$^R>. This happens immediately, so
1352C<$^R> can be used from other C<(?{ code })> assertions inside the same
1353regular expression.
b9ac3b5b 1354
19799a22
GS
1355The assignment to C<$^R> above is properly localized, so the old
1356value of C<$^R> is restored if the assertion is backtracked; compare
1357L<"Backtracking">.
b9ac3b5b 1358
e128ab2c
DM
1359Note that the special variable C<$^N> is particularly useful with code
1360blocks to capture the results of submatches in variables without having to
1361keep track of the number of nested parentheses. For example:
1362
1363 $_ = "The brown fox jumps over the lazy dog";
1364 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1365 print "color = $color, animal = $animal\n";
1366
8988a1bb 1367
14455d6c 1368=item C<(??{ code })>
d74e8afc
ITB
1369X<(??{})>
1370X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1371
19799a22 1372B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1373experimental, and may be changed without notice. Code executed that
1374has side effects may not perform identically from version to version
1375due to the effect of future optimisations in the regex engine.
0f5d15d6 1376
e128ab2c
DM
1377This is a "postponed" regular subexpression. It behaves in I<exactly> the
1378same way as a C<(?{ code })> code block as described above, except that
1379its return value, rather than being assigned to C<$^R>, is treated as a
1380pattern, compiled if it's a string (or used as-is if its a qr// object),
1381then matched as if it were inserted instead of this construct.
6bda09f9 1382
e128ab2c
DM
1383During the matching of this sub-pattern, it has its own set of
1384captures which are valid during the sub-match, but are discarded once
1385control returns to the main pattern. For example, the following matches,
1386with the inner pattern capturing "B" and matching "BB", while the outer
1387pattern captures "A";
1388
1389 my $inner = '(.)\1';
1390 "ABBA" =~ /^(.)(??{ $inner })\1/;
1391 print $1; # prints "A";
6bda09f9 1392
e128ab2c
DM
1393Note that this means that there is no way for the inner pattern to refer
1394to a capture group defined outside. (The code block itself can use C<$1>,
1395etc., to refer to the enclosing pattern's capture groups.) Thus, although
0f5d15d6 1396
e128ab2c
DM
1397 ('a' x 100)=~/(??{'(.)' x 100})/
1398
1399I<will> match, it will I<not> set $1 on exit.
19799a22
GS
1400
1401The following pattern matches a parenthesized group:
0f5d15d6 1402
d1fbf752
KW
1403 $re = qr{
1404 \(
1405 (?:
1406 (?> [^()]+ ) # Non-parens without backtracking
1407 |
1408 (??{ $re }) # Group with matching parens
1409 )*
1410 \)
1411 }x;
0f5d15d6 1412
6bda09f9
YO
1413See also C<(?PARNO)> for a different, more efficient way to accomplish
1414the same task.
1415
e128ab2c
DM
1416Executing a postponed regular expression 50 times without consuming any
1417input string will result in a fatal error. The maximum depth is compiled
1418into perl, so changing it requires a custom build.
6bda09f9 1419
542fa716
YO
1420=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1421X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1422X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1423X<regex, relative recursion>
6bda09f9 1424
e128ab2c
DM
1425Similar to C<(??{ code })> except that it does not involve executing any
1426code or potentially compiling a returned pattern string; instead it treats
1427the part of the current pattern contained within a specified capture group
1428as an independent pattern that must match at the current position.
1429Capture groups contained by the pattern will have the value as determined
1430by the outermost recursion.
6bda09f9 1431
894be9b7 1432PARNO is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1433the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1434the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716 1435C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
c27a5cfe 1436to be relative, with negative numbers indicating preceding capture groups
542fa716 1437and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1438declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1439Note that the counting for relative recursion differs from that of
c27a5cfe 1440relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1441included.
6bda09f9 1442
81714fb9 1443The following pattern matches a function foo() which may contain
f145b7e9 1444balanced parentheses as the argument.
6bda09f9 1445
d1fbf752 1446 $re = qr{ ( # paren group 1 (full function)
81714fb9 1447 foo
d1fbf752 1448 ( # paren group 2 (parens)
6bda09f9 1449 \(
d1fbf752 1450 ( # paren group 3 (contents of parens)
6bda09f9 1451 (?:
d1fbf752 1452 (?> [^()]+ ) # Non-parens without backtracking
6bda09f9 1453 |
d1fbf752 1454 (?2) # Recurse to start of paren group 2
6bda09f9
YO
1455 )*
1456 )
1457 \)
1458 )
1459 )
1460 }x;
1461
1462If the pattern was used as follows
1463
1464 'foo(bar(baz)+baz(bop))'=~/$re/
1465 and print "\$1 = $1\n",
1466 "\$2 = $2\n",
1467 "\$3 = $3\n";
1468
1469the output produced should be the following:
1470
1471 $1 = foo(bar(baz)+baz(bop))
1472 $2 = (bar(baz)+baz(bop))
81714fb9 1473 $3 = bar(baz)+baz(bop)
6bda09f9 1474
c27a5cfe 1475If there is no corresponding capture group defined, then it is a
61528107 1476fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1477string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1478into perl, so changing it requires a custom build.
1479
542fa716
YO
1480The following shows how using negative indexing can make it
1481easier to embed recursive patterns inside of a C<qr//> construct
1482for later use:
1483
1484 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
c77257ed 1485 if (/foo $parens \s+ \+ \s+ bar $parens/x) {
542fa716
YO
1486 # do something here...
1487 }
1488
81714fb9 1489B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1490PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1491a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1492as atomic. Also, modifiers are resolved at compile time, so constructs
1493like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1494be processed.
6bda09f9 1495
894be9b7
YO
1496=item C<(?&NAME)>
1497X<(?&NAME)>
1498
0d017f4d
WL
1499Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1500parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1501the same name, then it recurses to the leftmost.
1502
1503It is an error to refer to a name that is not declared somewhere in the
1504pattern.
1505
1f1031fe
YO
1506B<NOTE:> In order to make things easier for programmers with experience
1507with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1508may be used instead of C<< (?&NAME) >>.
1f1031fe 1509
e2e6a0f1
YO
1510=item C<(?(condition)yes-pattern|no-pattern)>
1511X<(?()>
286f584a 1512
e2e6a0f1 1513=item C<(?(condition)yes-pattern)>
286f584a 1514
41ef34de
ML
1515Conditional expression. Matches C<yes-pattern> if C<condition> yields
1516a true value, matches C<no-pattern> otherwise. A missing pattern always
1517matches.
1518
25e26d77 1519C<(condition)> should be one of: 1) an integer in
e2e6a0f1 1520parentheses (which is valid if the corresponding pair of parentheses
25e26d77 1521matched); 2) a look-ahead/look-behind/evaluate zero-width assertion; 3) a
c27a5cfe 1522name in angle brackets or single quotes (which is valid if a group
25e26d77 1523with the given name matched); or 4) the special symbol (R) (true when
e2e6a0f1
YO
1524evaluated inside of recursion or eval). Additionally the R may be
1525followed by a number, (which will be true when evaluated when recursing
1526inside of the appropriate group), or by C<&NAME>, in which case it will
1527be true only when evaluated during recursion in the named group.
1528
1529Here's a summary of the possible predicates:
1530
1531=over 4
1532
1533=item (1) (2) ...
1534
c27a5cfe 1535Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1536
1537=item (<NAME>) ('NAME')
1538
c27a5cfe 1539Checks if a group with the given name has matched something.
e2e6a0f1 1540
f01cd190
FC
1541=item (?=...) (?!...) (?<=...) (?<!...)
1542
1543Checks whether the pattern matches (or does not match, for the '!'
1544variants).
1545
e2e6a0f1
YO
1546=item (?{ CODE })
1547
f01cd190 1548Treats the return value of the code block as the condition.
e2e6a0f1
YO
1549
1550=item (R)
1551
1552Checks if the expression has been evaluated inside of recursion.
1553
1554=item (R1) (R2) ...
1555
1556Checks if the expression has been evaluated while executing directly
1557inside of the n-th capture group. This check is the regex equivalent of
1558
1559 if ((caller(0))[3] eq 'subname') { ... }
1560
1561In other words, it does not check the full recursion stack.
1562
1563=item (R&NAME)
1564
1565Similar to C<(R1)>, this predicate checks to see if we're executing
1566directly inside of the leftmost group with a given name (this is the same
1567logic used by C<(?&NAME)> to disambiguate). It does not check the full
1568stack, but only the name of the innermost active recursion.
1569
1570=item (DEFINE)
1571
1572In this case, the yes-pattern is never directly executed, and no
1573no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1574See below for details.
1575
1576=back
1577
1578For example:
1579
1580 m{ ( \( )?
1581 [^()]+
1582 (?(1) \) )
1583 }x
1584
1585matches a chunk of non-parentheses, possibly included in parentheses
1586themselves.
1587
0b928c2f
FC
1588A special form is the C<(DEFINE)> predicate, which never executes its
1589yes-pattern directly, and does not allow a no-pattern. This allows one to
1590define subpatterns which will be executed only by the recursion mechanism.
e2e6a0f1
YO
1591This way, you can define a set of regular expression rules that can be
1592bundled into any pattern you choose.
1593
1594It is recommended that for this usage you put the DEFINE block at the
1595end of the pattern, and that you name any subpatterns defined within it.
1596
1597Also, it's worth noting that patterns defined this way probably will
1598not be as efficient, as the optimiser is not very clever about
1599handling them.
1600
1601An example of how this might be used is as follows:
1602
2bf803e2 1603 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1604 (?(DEFINE)
2bf803e2
YO
1605 (?<NAME_PAT>....)
1606 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1607 )/x
1608
c27a5cfe
KW
1609Note that capture groups matched inside of recursion are not accessible
1610after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1611necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1612C<$+{NAME}> would be.
286f584a 1613
51a1303c
BF
1614Finally, keep in mind that subpatterns created inside a DEFINE block
1615count towards the absolute and relative number of captures, so this:
1616
1617 my @captures = "a" =~ /(.) # First capture
1618 (?(DEFINE)
1619 (?<EXAMPLE> 1 ) # Second capture
1620 )/x;
1621 say scalar @captures;
1622
1623Will output 2, not 1. This is particularly important if you intend to
1624compile the definitions with the C<qr//> operator, and later
1625interpolate them in another pattern.
1626
c47ff5f1 1627=item C<< (?>pattern) >>
6bda09f9 1628X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1629
19799a22
GS
1630An "independent" subexpression, one which matches the substring
1631that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1632position, and it matches I<nothing other than this substring>. This
19799a22
GS
1633construct is useful for optimizations of what would otherwise be
1634"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1635It may also be useful in places where the "grab all you can, and do not
1636give anything back" semantic is desirable.
19799a22 1637
c47ff5f1 1638For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1639(anchored at the beginning of string, as above) will match I<all>
1640characters C<a> at the beginning of string, leaving no C<a> for
1641C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1642since the match of the subgroup C<a*> is influenced by the following
1643group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1644C<a*ab> will match fewer characters than a standalone C<a*>, since
1645this makes the tail match.
1646
0b928c2f
FC
1647C<< (?>pattern) >> does not disable backtracking altogether once it has
1648matched. It is still possible to backtrack past the construct, but not
1649into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1650
c47ff5f1 1651An effect similar to C<< (?>pattern) >> may be achieved by writing
0b928c2f
FC
1652C<(?=(pattern))\g{-1}>. This matches the same substring as a standalone
1653C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
c47ff5f1 1654makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1655(The difference between these two constructs is that the second one
1656uses a capturing group, thus shifting ordinals of backreferences
1657in the rest of a regular expression.)
1658
1659Consider this pattern:
c277df42 1660
871b0233 1661 m{ \(
e2e6a0f1 1662 (
f793d64a 1663 [^()]+ # x+
e2e6a0f1 1664 |
871b0233
IZ
1665 \( [^()]* \)
1666 )+
e2e6a0f1 1667 \)
871b0233 1668 }x
5a964f20 1669
19799a22
GS
1670That will efficiently match a nonempty group with matching parentheses
1671two levels deep or less. However, if there is no such group, it
1672will take virtually forever on a long string. That's because there
1673are so many different ways to split a long string into several
1674substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1675to a subpattern of the above pattern. Consider how the pattern
1676above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1677seconds, but that each extra letter doubles this time. This
1678exponential performance will make it appear that your program has
14218588 1679hung. However, a tiny change to this pattern
5a964f20 1680
e2e6a0f1
YO
1681 m{ \(
1682 (
f793d64a 1683 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1684 |
871b0233
IZ
1685 \( [^()]* \)
1686 )+
e2e6a0f1 1687 \)
871b0233 1688 }x
c277df42 1689
c47ff5f1 1690which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1691this yourself would be a productive exercise), but finishes in a fourth
1692the time when used on a similar string with 1000000 C<a>s. Be aware,
0b928c2f
FC
1693however, that, when this construct is followed by a
1694quantifier, it currently triggers a warning message under
9f1b1f2d 1695the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1696C<"matches null string many times in regex">.
c277df42 1697
c47ff5f1 1698On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1699effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1700This was only 4 times slower on a string with 1000000 C<a>s.
1701
9da458fc
IZ
1702The "grab all you can, and do not give anything back" semantic is desirable
1703in many situations where on the first sight a simple C<()*> looks like
1704the correct solution. Suppose we parse text with comments being delimited
1705by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1706its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1707the comment delimiter, because it may "give up" some whitespace if
1708the remainder of the pattern can be made to match that way. The correct
1709answer is either one of these:
1710
1711 (?>#[ \t]*)
1712 #[ \t]*(?![ \t])
1713
1714For example, to grab non-empty comments into $1, one should use either
1715one of these:
1716
1717 / (?> \# [ \t]* ) ( .+ ) /x;
1718 / \# [ \t]* ( [^ \t] .* ) /x;
1719
1720Which one you pick depends on which of these expressions better reflects
1721the above specification of comments.
1722
6bda09f9
YO
1723In some literature this construct is called "atomic matching" or
1724"possessive matching".
1725
b9b4dddf
YO
1726Possessive quantifiers are equivalent to putting the item they are applied
1727to inside of one of these constructs. The following equivalences apply:
1728
1729 Quantifier Form Bracketing Form
1730 --------------- ---------------
1731 PAT*+ (?>PAT*)
1732 PAT++ (?>PAT+)
1733 PAT?+ (?>PAT?)
1734 PAT{min,max}+ (?>PAT{min,max})
1735
9d1a5160
KW
1736=item C<(?[ ])>
1737X<set operations>
1738
1739This is an experimental feature present starting in 5.18, but is subject
1740to change as we gain field experience with it. Any attempt to use it
1741will raise a warning, unless disabled via
1742
1743 no warnings "experimental::regex_sets";
1744
1745Comments on this feature are welcome; send email to
1746C<perl5-porters@perl.org>.
1747
1748This is a fancy bracketed character class that can be used for more
1749readable and less error-prone classes, and to perform set operations,
1750such as intersection. An example is
1751
1752 /(?[ \p{Thai} & \p{Digit} ])/
1753
1754This will match all the digit characters that are in the Thai script.
1755We can extend this by
1756
1757 /(?[ ( \p{Thai} + \p{Lao} ) & \p{Digit} ])/
1758
0b34b372 1759This matches digits that are in either the Thai or Laotian scripts.
9d1a5160
KW
1760
1761Notice the white space in these examples. This construct always has
1762L</C<E<sol>x>> turned on.
1763
1764The available binary operators are:
1765
1766 & intersection
1767 + union
1768 | another name for '+', hence means union
1769 - subtraction (the result matches the set consisting of those
1770 code points matched by the first operand, excluding any that
1771 are also matched by the second operand)
1772 ^ symmetric difference (the union minus the intersection). This
1773 is like an exclusive or, in that the result is the set of code
1774 points that are matched by either, but not both, of the
1775 operands.
1776
1777There is one unary operator:
1778
1779 ! complement
1780
1781All the binary operators left associate, and are of equal precedence.
1782The unary operator right associates, and has higher precedence. Use
1783parentheses to override the default associations.
1784
1785The main restriction is that everything is a metacharacter. Thus,
1786you cannot refer to single characters by doing something like this:
1787
1788 /(?[ a + b ])/ # Syntax error!
1789
1790The easiest way to specify an individual typable character is to enclose
1791it in brackets:
1792
1793 /(?[ [a] + [b] ])/
1794
1795(This is the same thing as C<[ab]>.) You could also have said the
1796equivalent
1797
1798 /(?[[ a b ]])/
1799
1800(You can, of course, specify single characters by using, C<\x{ }>,
1801C<\N{ }>, etc.)
1802
1803This last example shows the use of this construct to specify an ordinary
1804bracketed character class without set operations. Note the white space
1805within it. To specify a matchable white space character, you can escape
1806it with a backslash, like:
1807
1808 /(?[ [ a e i o u \ ] ])/
1809
1810This matches the English vowels plus the SPACE character.
1811All the other escapes accepted by normal bracketed character classes are
1812accepted here as well; but unlike the normal ones, unrecognized escapes are
1813fatal errors here.
1814
1815All warnings from these class elements are fatal, as well as some
1816practices that don't currently warn. For example you cannot say
1817
1818 /(?[ [ \xF ] ])/ # Syntax error!
1819
1820You have to have two hex digits after a braceless C<\x> (use a leading
1821zero to make two). These restrictions are to lower the incidence of
1822typos causing the class to not match what you thought it would.
1823
1824The final difference between regular bracketed character classes and
1825these, is that it is not possible to get the latter to match a
1826multi-character fold. Thus,
1827
1828 /(?[ [\xDF] ])/iu
1829
1830does not match the string C<ss>.
1831
1832You don't have to enclose Posix class names inside double brackets. The
1833following works
1834
1835 /(?[ [:word:] - [:lower:] ])/
1836
1837C<< (?[ ]) >> is a compile-time construct. Any attempt to use something
1838which isn't knowable until run-time is a fatal error. Thus, this
1839construct cannot be used within the scope of C<use locale> (or the
1840L</C<E<sol>l>> regex modifier). Any L<user-defined
1841property|perlunicode/"User-Defined Character Properties"> used must be
1842already defined by the time the regular expression is compiled; but note
1843that this construct can be used to avoid defining such properties.
1844
1845A regular expression using this construct that otherwise would compile
1846using L</C<E<sol>d>> rules will instead use L</C<E<sol>u>>.
1847
1848The L</C<E<sol>x>> processing within this class is an extended form.
1849Besides the characters that are considered white space in normal C</x>
1850processing, there are 5 others, recommended by the Unicode standard:
1851
1852 U+0085 NEXT LINE
1853 U+200E LEFT-TO-RIGHT MARK
1854 U+200F RIGHT-TO-LEFT MARK
1855 U+2028 LINE SEPARATOR
1856 U+2029 PARAGRAPH SEPARATOR
1857
1858Note that skipping white space applies only to the interior of this
1859construct. There must not be any space between any of the characters
1860that form the initial C<(?[>. Nor may there be space between the
1861closing C<])> characters.
1862
1863Due to the way that Perl parses things, your parentheses and brackets
1864may need to be balanced, even including comments.
1865
1866Since this experimental, we may change this so that other legal uses of
1867normal bracketed character classes might become illegal. One proposal,
1868for example, is to forbid adjacent uses of the same character, as in
1869C<[aa]>. This is likely a typo, as the second "a" adds nothing.
1870
e2e6a0f1
YO
1871=back
1872
1873=head2 Special Backtracking Control Verbs
1874
1875B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1876removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1877be noted to avoid problems during upgrades.
1878
1879These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1880otherwise stated the ARG argument is optional; in some cases, it is
1881forbidden.
1882
1883Any pattern containing a special backtracking verb that allows an argument
e1020413 1884has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1885C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1886rules apply:
e2e6a0f1 1887
5d458dd8
YO
1888On failure, the C<$REGERROR> variable will be set to the ARG value of the
1889verb pattern, if the verb was involved in the failure of the match. If the
1890ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1891name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1892none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1893
5d458dd8
YO
1894On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1895the C<$REGMARK> variable will be set to the name of the last
1896C<(*MARK:NAME)> pattern executed. See the explanation for the
1897C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1898
5d458dd8 1899B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1900and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1901readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1902Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1903
1904If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1905argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1906
70ca8714 1907=over 3
e2e6a0f1
YO
1908
1909=item Verbs that take an argument
1910
1911=over 4
1912
5d458dd8 1913=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1914X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1915
5d458dd8
YO
1916This zero-width pattern prunes the backtracking tree at the current point
1917when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1918where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1919A may backtrack as necessary to match. Once it is reached, matching
1920continues in B, which may also backtrack as necessary; however, should B
1921not match, then no further backtracking will take place, and the pattern
1922will fail outright at the current starting position.
54612592
YO
1923
1924The following example counts all the possible matching strings in a
1925pattern (without actually matching any of them).
1926
e2e6a0f1 1927 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1928 print "Count=$count\n";
1929
1930which produces:
1931
1932 aaab
1933 aaa
1934 aa
1935 a
1936 aab
1937 aa
1938 a
1939 ab
1940 a
1941 Count=9
1942
5d458dd8 1943If we add a C<(*PRUNE)> before the count like the following
54612592 1944
5d458dd8 1945 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1946 print "Count=$count\n";
1947
0b928c2f 1948we prevent backtracking and find the count of the longest matching string
353c6505 1949at each matching starting point like so:
54612592
YO
1950
1951 aaab
1952 aab
1953 ab
1954 Count=3
1955
5d458dd8 1956Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1957
5d458dd8
YO
1958See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1959control backtracking. In some cases, the use of C<(*PRUNE)> can be
1960replaced with a C<< (?>pattern) >> with no functional difference; however,
1961C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1962C<< (?>pattern) >> alone.
54612592 1963
5d458dd8
YO
1964=item C<(*SKIP)> C<(*SKIP:NAME)>
1965X<(*SKIP)>
e2e6a0f1 1966
5d458dd8 1967This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1968failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1969to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1970of this pattern. This effectively means that the regex engine "skips" forward
1971to this position on failure and tries to match again, (assuming that
1972there is sufficient room to match).
1973
1974The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1975C<(*MARK:NAME)> was encountered while matching, then it is that position
1976which is used as the "skip point". If no C<(*MARK)> of that name was
1977encountered, then the C<(*SKIP)> operator has no effect. When used
1978without a name the "skip point" is where the match point was when
1979executing the (*SKIP) pattern.
1980
0b928c2f 1981Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1982is twice as long:
1983
d1fbf752
KW
1984 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1985 print "Count=$count\n";
24b23f37
YO
1986
1987outputs
1988
1989 aaab
1990 aaab
1991 Count=2
1992
5d458dd8 1993Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1994executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1995C<(*SKIP)> was executed.
1996
5d458dd8 1997=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1998X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1999
2000This zero-width pattern can be used to mark the point reached in a string
2001when a certain part of the pattern has been successfully matched. This
2002mark may be given a name. A later C<(*SKIP)> pattern will then skip
2003forward to that point if backtracked into on failure. Any number of
b4222fa9 2004C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
2005
2006In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
2007can be used to "label" a pattern branch, so that after matching, the
2008program can determine which branches of the pattern were involved in the
2009match.
2010
2011When a match is successful, the C<$REGMARK> variable will be set to the
2012name of the most recently executed C<(*MARK:NAME)> that was involved
2013in the match.
2014
2015This can be used to determine which branch of a pattern was matched
c27a5cfe 2016without using a separate capture group for each branch, which in turn
5d458dd8
YO
2017can result in a performance improvement, as perl cannot optimize
2018C</(?:(x)|(y)|(z))/> as efficiently as something like
2019C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
2020
2021When a match has failed, and unless another verb has been involved in
2022failing the match and has provided its own name to use, the C<$REGERROR>
2023variable will be set to the name of the most recently executed
2024C<(*MARK:NAME)>.
2025
42ac7c82 2026See L</(*SKIP)> for more details.
5d458dd8 2027
b62d2d15
YO
2028As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
2029
5d458dd8
YO
2030=item C<(*THEN)> C<(*THEN:NAME)>
2031
ac9d8485 2032This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
2033C<(*PRUNE)>, this verb always matches, and when backtracked into on
2034failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
2035innermost enclosing group (capturing or otherwise) that has alternations.
2036The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
2037count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
2038
2039Its name comes from the observation that this operation combined with the
2040alternation operator (C<|>) can be used to create what is essentially a
2041pattern-based if/then/else block:
2042
2043 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
2044
2045Note that if this operator is used and NOT inside of an alternation then
2046it acts exactly like the C<(*PRUNE)> operator.
2047
2048 / A (*PRUNE) B /
2049
2050is the same as
2051
2052 / A (*THEN) B /
2053
2054but
2055
25e26d77 2056 / ( A (*THEN) B | C ) /
5d458dd8
YO
2057
2058is not the same as
2059
25e26d77 2060 / ( A (*PRUNE) B | C ) /
5d458dd8
YO
2061
2062as after matching the A but failing on the B the C<(*THEN)> verb will
2063backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 2064
cbeadc21
JV
2065=back
2066
2067=item Verbs without an argument
2068
2069=over 4
2070
e2e6a0f1
YO
2071=item C<(*COMMIT)>
2072X<(*COMMIT)>
24b23f37 2073
241e7389 2074This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
2075zero-width pattern similar to C<(*SKIP)>, except that when backtracked
2076into on failure it causes the match to fail outright. No further attempts
2077to find a valid match by advancing the start pointer will occur again.
2078For example,
24b23f37 2079
d1fbf752
KW
2080 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
2081 print "Count=$count\n";
24b23f37
YO
2082
2083outputs
2084
2085 aaab
2086 Count=1
2087
e2e6a0f1
YO
2088In other words, once the C<(*COMMIT)> has been entered, and if the pattern
2089does not match, the regex engine will not try any further matching on the
2090rest of the string.
c277df42 2091
e2e6a0f1
YO
2092=item C<(*FAIL)> C<(*F)>
2093X<(*FAIL)> X<(*F)>
9af228c6 2094
e2e6a0f1
YO
2095This pattern matches nothing and always fails. It can be used to force the
2096engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
2097fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 2098
e2e6a0f1 2099It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 2100
e2e6a0f1
YO
2101=item C<(*ACCEPT)>
2102X<(*ACCEPT)>
9af228c6 2103
e2e6a0f1
YO
2104B<WARNING:> This feature is highly experimental. It is not recommended
2105for production code.
9af228c6 2106
e2e6a0f1
YO
2107This pattern matches nothing and causes the end of successful matching at
2108the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
2109whether there is actually more to match in the string. When inside of a
0d017f4d 2110nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 2111via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 2112
c27a5cfe 2113If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
2114marked as ended at the point at which the C<(*ACCEPT)> was encountered.
2115For instance:
9af228c6 2116
e2e6a0f1 2117 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 2118
e2e6a0f1 2119will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 2120be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 2121string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
2122
2123=back
c277df42 2124
a0d0e21e
LW
2125=back
2126
c07a80fd 2127=head2 Backtracking
d74e8afc 2128X<backtrack> X<backtracking>
c07a80fd 2129
35a734be
IZ
2130NOTE: This section presents an abstract approximation of regular
2131expression behavior. For a more rigorous (and complicated) view of
2132the rules involved in selecting a match among possible alternatives,
0d017f4d 2133see L<Combining RE Pieces>.
35a734be 2134
c277df42 2135A fundamental feature of regular expression matching involves the
5a964f20 2136notion called I<backtracking>, which is currently used (when needed)
0d017f4d 2137by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
2138C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
2139internally, but the general principle outlined here is valid.
c07a80fd
PP
2140
2141For a regular expression to match, the I<entire> regular expression must
2142match, not just part of it. So if the beginning of a pattern containing a
2143quantifier succeeds in a way that causes later parts in the pattern to
2144fail, the matching engine backs up and recalculates the beginning
2145part--that's why it's called backtracking.
2146
2147Here is an example of backtracking: Let's say you want to find the
2148word following "foo" in the string "Food is on the foo table.":
2149
2150 $_ = "Food is on the foo table.";
2151 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 2152 print "$2 follows $1.\n";
c07a80fd
PP
2153 }
2154
2155When the match runs, the first part of the regular expression (C<\b(foo)>)
2156finds a possible match right at the beginning of the string, and loads up
2157$1 with "Foo". However, as soon as the matching engine sees that there's
2158no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 2159mistake and starts over again one character after where it had the
c07a80fd
PP
2160tentative match. This time it goes all the way until the next occurrence
2161of "foo". The complete regular expression matches this time, and you get
2162the expected output of "table follows foo."
2163
2164Sometimes minimal matching can help a lot. Imagine you'd like to match
2165everything between "foo" and "bar". Initially, you write something
2166like this:
2167
2168 $_ = "The food is under the bar in the barn.";
2169 if ( /foo(.*)bar/ ) {
f793d64a 2170 print "got <$1>\n";
c07a80fd
PP
2171 }
2172
2173Which perhaps unexpectedly yields:
2174
2175 got <d is under the bar in the >
2176
2177That's because C<.*> was greedy, so you get everything between the
14218588 2178I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
2179to use minimal matching to make sure you get the text between a "foo"
2180and the first "bar" thereafter.
2181
2182 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2183 got <d is under the >
2184
0d017f4d 2185Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2186of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2187So you write this:
2188
2189 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2190 if ( /(.*)(\d*)/ ) { # Wrong!
2191 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2192 }
2193
2194That won't work at all, because C<.*> was greedy and gobbled up the
2195whole string. As C<\d*> can match on an empty string the complete
2196regular expression matched successfully.
2197
8e1088bc 2198 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2199
2200Here are some variants, most of which don't work:
2201
2202 $_ = "I have 2 numbers: 53147";
2203 @pats = qw{
f793d64a
KW
2204 (.*)(\d*)
2205 (.*)(\d+)
2206 (.*?)(\d*)
2207 (.*?)(\d+)
2208 (.*)(\d+)$
2209 (.*?)(\d+)$
2210 (.*)\b(\d+)$
2211 (.*\D)(\d+)$
c07a80fd
PP
2212 };
2213
2214 for $pat (@pats) {
f793d64a
KW
2215 printf "%-12s ", $pat;
2216 if ( /$pat/ ) {
2217 print "<$1> <$2>\n";
2218 } else {
2219 print "FAIL\n";
2220 }
c07a80fd
PP
2221 }
2222
2223That will print out:
2224
2225 (.*)(\d*) <I have 2 numbers: 53147> <>
2226 (.*)(\d+) <I have 2 numbers: 5314> <7>
2227 (.*?)(\d*) <> <>
2228 (.*?)(\d+) <I have > <2>
2229 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2230 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2231 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2232 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2233
2234As you see, this can be a bit tricky. It's important to realize that a
2235regular expression is merely a set of assertions that gives a definition
2236of success. There may be 0, 1, or several different ways that the
2237definition might succeed against a particular string. And if there are
5a964f20
TC
2238multiple ways it might succeed, you need to understand backtracking to
2239know which variety of success you will achieve.
c07a80fd 2240
19799a22 2241When using look-ahead assertions and negations, this can all get even
8b19b778 2242trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2243followed by "123". You might try to write that as
2244
871b0233 2245 $_ = "ABC123";
f793d64a
KW
2246 if ( /^\D*(?!123)/ ) { # Wrong!
2247 print "Yup, no 123 in $_\n";
871b0233 2248 }
c07a80fd
PP
2249
2250But that isn't going to match; at least, not the way you're hoping. It
2251claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2252why that pattern matches, contrary to popular expectations:
c07a80fd 2253
4358a253
SS
2254 $x = 'ABC123';
2255 $y = 'ABC445';
c07a80fd 2256
4358a253
SS
2257 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2258 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2259
4358a253
SS
2260 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2261 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2262
2263This prints
2264
2265 2: got ABC
2266 3: got AB
2267 4: got ABC
2268
5f05dabc 2269You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2270general purpose version of test 1. The important difference between
2271them is that test 3 contains a quantifier (C<\D*>) and so can use
2272backtracking, whereas test 1 will not. What's happening is
2273that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2274non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2275let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2276fail.
14218588 2277
c07a80fd 2278The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2279try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2280a quantifier (C<\D*>) has been used in the regular expression, the
2281search engine can backtrack and retry the match differently
54310121 2282in the hope of matching the complete regular expression.
c07a80fd 2283
5a964f20
TC
2284The pattern really, I<really> wants to succeed, so it uses the
2285standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2286time. Now there's indeed something following "AB" that is not
14218588 2287"123". It's "C123", which suffices.
c07a80fd 2288
14218588
GS
2289We can deal with this by using both an assertion and a negation.
2290We'll say that the first part in $1 must be followed both by a digit
2291and by something that's not "123". Remember that the look-aheads
2292are zero-width expressions--they only look, but don't consume any
2293of the string in their match. So rewriting this way produces what
c07a80fd
PP
2294you'd expect; that is, case 5 will fail, but case 6 succeeds:
2295
4358a253
SS
2296 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2297 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2298
2299 6: got ABC
2300
5a964f20 2301In other words, the two zero-width assertions next to each other work as though
19799a22 2302they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2303matches only if you're at the beginning of the line AND the end of the
2304line simultaneously. The deeper underlying truth is that juxtaposition in
2305regular expressions always means AND, except when you write an explicit OR
2306using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2307although the attempted matches are made at different positions because "a"
2308is not a zero-width assertion, but a one-width assertion.
2309
0d017f4d 2310B<WARNING>: Particularly complicated regular expressions can take
14218588 2311exponential time to solve because of the immense number of possible
0d017f4d 2312ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2313internal optimizations done by the regular expression engine, this will
2314take a painfully long time to run:
c07a80fd 2315
e1901655
IZ
2316 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2317
2318And if you used C<*>'s in the internal groups instead of limiting them
2319to 0 through 5 matches, then it would take forever--or until you ran
2320out of stack space. Moreover, these internal optimizations are not
2321always applicable. For example, if you put C<{0,5}> instead of C<*>
2322on the external group, no current optimization is applicable, and the
2323match takes a long time to finish.
c07a80fd 2324
9da458fc
IZ
2325A powerful tool for optimizing such beasts is what is known as an
2326"independent group",
96090e4f 2327which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2328zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2329the tail match, since they are in "logical" context: only
14218588 2330whether they match is considered relevant. For an example
9da458fc 2331where side-effects of look-ahead I<might> have influenced the
96090e4f 2332following match, see L</C<< (?>pattern) >>>.
c277df42 2333
a0d0e21e 2334=head2 Version 8 Regular Expressions
d74e8afc 2335X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2336
5a964f20 2337In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2338routines, here are the pattern-matching rules not described above.
2339
54310121 2340Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2341with a special meaning described here or above. You can cause
5a964f20 2342characters that normally function as metacharacters to be interpreted
5f05dabc 2343literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2344character; "\\" matches a "\"). This escape mechanism is also required
2345for the character used as the pattern delimiter.
2346
2347A series of characters matches that series of characters in the target
0b928c2f 2348string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2349string.
a0d0e21e
LW
2350
2351You can specify a character class, by enclosing a list of characters
5d458dd8 2352in C<[]>, which will match any character from the list. If the
a0d0e21e 2353first character after the "[" is "^", the class matches any character not
14218588 2354in the list. Within a list, the "-" character specifies a
5a964f20 2355range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2356inclusive. If you want either "-" or "]" itself to be a member of a
2357class, put it at the start of the list (possibly after a "^"), or
2358escape it with a backslash. "-" is also taken literally when it is
2359at the end of the list, just before the closing "]". (The
84850974
DD
2360following all specify the same class of three characters: C<[-az]>,
2361C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2362specifies a class containing twenty-six characters, even on EBCDIC-based
2363character sets.) Also, if you try to use the character
2364classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2365a range, the "-" is understood literally.
a0d0e21e 2366
8ada0baa
JH
2367Note also that the whole range idea is rather unportable between
2368character sets--and even within character sets they may cause results
2369you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2370that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2371[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2372spell out the character sets in full.
2373
54310121 2374Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2375used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2376"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2377of three octal digits, matches the character whose coded character set value
5d458dd8 2378is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2379matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2380matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2381matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2382
2383You can specify a series of alternatives for a pattern using "|" to
2384separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2385or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2386first alternative includes everything from the last pattern delimiter
0b928c2f 2387("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2388the last alternative contains everything from the last "|" to the next
0b928c2f 2389closing pattern delimiter. That's why it's common practice to include
14218588 2390alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2391start and end.
2392
5a964f20 2393Alternatives are tried from left to right, so the first
a3cb178b
GS
2394alternative found for which the entire expression matches, is the one that
2395is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2396example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2397part will match, as that is the first alternative tried, and it successfully
2398matches the target string. (This might not seem important, but it is
2399important when you are capturing matched text using parentheses.)
2400
5a964f20 2401Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2402so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2403
14218588
GS
2404Within a pattern, you may designate subpatterns for later reference
2405by enclosing them in parentheses, and you may refer back to the
2406I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2407\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2408of their opening parenthesis. A backreference matches whatever
2409actually matched the subpattern in the string being examined, not
d8b950dc 2410the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2411match "0x1234 0x4321", but not "0x1234 01234", because subpattern
24121 matched "0x", even though the rule C<0|0x> could potentially match
2413the leading 0 in the second number.
cb1a09d0 2414
0d017f4d 2415=head2 Warning on \1 Instead of $1
cb1a09d0 2416
5a964f20 2417Some people get too used to writing things like:
cb1a09d0
AD
2418
2419 $pattern =~ s/(\W)/\\\1/g;
2420
3ff1c45a
KW
2421This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2422shocking the
cb1a09d0 2423B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2424PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2425the usual double-quoted string means a control-A. The customary Unix
2426meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2427of doing that, you get yourself into trouble if you then add an C</e>
2428modifier.
2429
f793d64a 2430 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2431
2432Or if you try to do
2433
2434 s/(\d+)/\1000/;
2435
2436You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2437C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2438with the operation of matching a backreference. Certainly they mean two
2439different things on the I<left> side of the C<s///>.
9fa51da4 2440
0d017f4d 2441=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2442
19799a22 2443B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2444
2445Regular expressions provide a terse and powerful programming language. As
2446with most other power tools, power comes together with the ability
2447to wreak havoc.
2448
2449A common abuse of this power stems from the ability to make infinite
628afcb5 2450loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2451
2452 'foo' =~ m{ ( o? )* }x;
2453
0d017f4d 2454The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2455in the string is not moved by the match, C<o?> would match again and again
527e91da 2456because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2457is with the looping modifier C<//g>:
2458
2459 @matches = ( 'foo' =~ m{ o? }xg );
2460
2461or
2462
2463 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2464
2465or the loop implied by split().
2466
2467However, long experience has shown that many programming tasks may
14218588
GS
2468be significantly simplified by using repeated subexpressions that
2469may match zero-length substrings. Here's a simple example being:
c84d73f1 2470
d1fbf752 2471 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2472 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2473
9da458fc 2474Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2475the infinite loop>. The rules for this are different for lower-level
527e91da 2476loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2477ones like the C</g> modifier or split() operator.
2478
19799a22
GS
2479The lower-level loops are I<interrupted> (that is, the loop is
2480broken) when Perl detects that a repeated expression matched a
2481zero-length substring. Thus
c84d73f1
IZ
2482
2483 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2484
5d458dd8 2485is made equivalent to
c84d73f1 2486
0b928c2f
FC
2487 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2488
2489For example, this program
2490
2491 #!perl -l
2492 "aaaaab" =~ /
2493 (?:
2494 a # non-zero
2495 | # or
2496 (?{print "hello"}) # print hello whenever this
2497 # branch is tried
2498 (?=(b)) # zero-width assertion
2499 )* # any number of times
2500 /x;
2501 print $&;
2502 print $1;
c84d73f1 2503
0b928c2f
FC
2504prints
2505
2506 hello
2507 aaaaa
2508 b
2509
2510Notice that "hello" is only printed once, as when Perl sees that the sixth
2511iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2512the C<*>.
2513
2514The higher-level loops preserve an additional state between iterations:
5d458dd8 2515whether the last match was zero-length. To break the loop, the following
c84d73f1 2516match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2517This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2518and so the I<second best> match is chosen if the I<best> match is of
2519zero length.
2520
19799a22 2521For example:
c84d73f1
IZ
2522
2523 $_ = 'bar';
2524 s/\w??/<$&>/g;
2525
20fb949f 2526results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2527match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2528best> match is what is matched by C<\w>. Thus zero-length matches
2529alternate with one-character-long matches.
2530
5d458dd8 2531Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2532position one notch further in the string.
2533
19799a22 2534The additional state of being I<matched with zero-length> is associated with
c84d73f1 2535the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2536Zero-length matches at the end of the previous match are ignored
2537during C<split>.
c84d73f1 2538
0d017f4d 2539=head2 Combining RE Pieces
35a734be
IZ
2540
2541Each of the elementary pieces of regular expressions which were described
2542before (such as C<ab> or C<\Z>) could match at most one substring
2543at the given position of the input string. However, in a typical regular
2544expression these elementary pieces are combined into more complicated
0b928c2f 2545patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2546(in these examples C<S> and C<T> are regular subexpressions).
2547
2548Such combinations can include alternatives, leading to a problem of choice:
2549if we match a regular expression C<a|ab> against C<"abc">, will it match
2550substring C<"a"> or C<"ab">? One way to describe which substring is
2551actually matched is the concept of backtracking (see L<"Backtracking">).
2552However, this description is too low-level and makes you think
2553in terms of a particular implementation.
2554
2555Another description starts with notions of "better"/"worse". All the
2556substrings which may be matched by the given regular expression can be
2557sorted from the "best" match to the "worst" match, and it is the "best"
2558match which is chosen. This substitutes the question of "what is chosen?"
2559by the question of "which matches are better, and which are worse?".
2560
2561Again, for elementary pieces there is no such question, since at most
2562one match at a given position is possible. This section describes the
2563notion of better/worse for combining operators. In the description
2564below C<S> and C<T> are regular subexpressions.
2565
13a2d996 2566=over 4
35a734be
IZ
2567
2568=item C<ST>
2569
2570Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2571substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2572which can be matched by C<T>.
35a734be 2573
0b928c2f 2574If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2575match than C<A'B'>.
2576
2577If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2578C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2579
2580=item C<S|T>
2581
2582When C<S> can match, it is a better match than when only C<T> can match.
2583
2584Ordering of two matches for C<S> is the same as for C<S>. Similar for
2585two matches for C<T>.
2586
2587=item C<S{REPEAT_COUNT}>
2588
2589Matches as C<SSS...S> (repeated as many times as necessary).
2590
2591=item C<S{min,max}>
2592
2593Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2594
2595=item C<S{min,max}?>
2596
2597Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2598
2599=item C<S?>, C<S*>, C<S+>
2600
2601Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2602
2603=item C<S??>, C<S*?>, C<S+?>
2604
2605Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2606
c47ff5f1 2607=item C<< (?>S) >>
35a734be
IZ
2608
2609Matches the best match for C<S> and only that.
2610
2611=item C<(?=S)>, C<(?<=S)>
2612
2613Only the best match for C<S> is considered. (This is important only if
2614C<S> has capturing parentheses, and backreferences are used somewhere
2615else in the whole regular expression.)
2616
2617=item C<(?!S)>, C<(?<!S)>
2618
2619For this grouping operator there is no need to describe the ordering, since
2620only whether or not C<S> can match is important.
2621
6bda09f9 2622=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2623
2624The ordering is the same as for the regular expression which is
c27a5cfe 2625the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
2626
2627=item C<(?(condition)yes-pattern|no-pattern)>
2628
2629Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2630already determined. The ordering of the matches is the same as for the
2631chosen subexpression.
2632
2633=back
2634
2635The above recipes describe the ordering of matches I<at a given position>.
2636One more rule is needed to understand how a match is determined for the
2637whole regular expression: a match at an earlier position is always better
2638than a match at a later position.
2639
0d017f4d 2640=head2 Creating Custom RE Engines
c84d73f1 2641
0b928c2f
FC
2642As of Perl 5.10.0, one can create custom regular expression engines. This
2643is not for the faint of heart, as they have to plug in at the C level. See
2644L<perlreapi> for more details.
2645
2646As an alternative, overloaded constants (see L<overload>) provide a simple
2647way to extend the functionality of the RE engine, by substituting one
2648pattern for another.
c84d73f1
IZ
2649
2650Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2651matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2652characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2653at these positions, so we want to have each C<\Y|> in the place of the
2654more complicated version. We can create a module C<customre> to do
2655this:
2656
2657 package customre;
2658 use overload;
2659
2660 sub import {
2661 shift;
2662 die "No argument to customre::import allowed" if @_;
2663 overload::constant 'qr' => \&convert;
2664 }
2665
2666 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2667
580a9fe1
RGS
2668 # We must also take care of not escaping the legitimate \\Y|
2669 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2670 my %rules = ( '\\' => '\\\\',
f793d64a 2671 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2672 sub convert {
2673 my $re = shift;
5d458dd8 2674 $re =~ s{
c84d73f1
IZ
2675 \\ ( \\ | Y . )
2676 }
5d458dd8 2677 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2678 return $re;
2679 }
2680
2681Now C<use customre> enables the new escape in constant regular
2682expressions, i.e., those without any runtime variable interpolations.
2683As documented in L<overload>, this conversion will work only over
2684literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2685part of this regular expression needs to be converted explicitly
2686(but only if the special meaning of C<\Y|> should be enabled inside $re):
2687
2688 use customre;
2689 $re = <>;
2690 chomp $re;
2691 $re = customre::convert $re;
2692 /\Y|$re\Y|/;
2693
0b928c2f 2694=head2 PCRE/Python Support
1f1031fe 2695
0b928c2f 2696As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2697to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2698Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2699
2700=over 4
2701
ae5648b3 2702=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2703
c27a5cfe 2704Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2705
2706=item C<< (?P=NAME) >>
2707
c27a5cfe 2708Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2709
2710=item C<< (?P>NAME) >>
2711
c27a5cfe 2712Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2713
ee9b8eae 2714=back
1f1031fe 2715
19799a22
GS
2716=head1 BUGS
2717
88c9975e
KW
2718Many regular expression constructs don't work on EBCDIC platforms.
2719
ed7efc79
KW
2720There are a number of issues with regard to case-insensitive matching
2721in Unicode rules. See C<i> under L</Modifiers> above.
2722
9da458fc
IZ
2723This document varies from difficult to understand to completely
2724and utterly opaque. The wandering prose riddled with jargon is
2725hard to fathom in several places.
2726
2727This document needs a rewrite that separates the tutorial content
2728from the reference content.
19799a22
GS
2729
2730=head1 SEE ALSO
9fa51da4 2731
91e0c79e
MJD
2732L<perlrequick>.
2733
2734L<perlretut>.
2735
9b599b2a
GS
2736L<perlop/"Regexp Quote-Like Operators">.
2737
1e66bd83
PP
2738L<perlop/"Gory details of parsing quoted constructs">.
2739
14218588
GS
2740L<perlfaq6>.
2741
9b599b2a
GS
2742L<perlfunc/pos>.
2743
2744L<perllocale>.
2745
fb55449c
JH
2746L<perlebcdic>.
2747
14218588
GS
2748I<Mastering Regular Expressions> by Jeffrey Friedl, published
2749by O'Reilly and Associates.