This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
\N is no longer experimental
[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
d28d8023
KW
356characters C<[ \f\n\r\t]>, and starting in Perl v5.18, experimentally,
357the vertical tab; C<\w> means the 63 characters
808432af
KW
358C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
359C<[[:print:]]> match only the appropriate ASCII-range characters.
360
361This modifier is useful for people who only incidentally use Unicode,
362and who do not wish to be burdened with its complexities and security
363concerns.
364
365With C</a>, one can write C<\d> with confidence that it will only match
366ASCII characters, and should the need arise to match beyond ASCII, you
367can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>). There are
368similar C<\p{...}> constructs that can match beyond ASCII both white
369space (see L<perlrecharclass/Whitespace>), and Posix classes (see
370L<perlrecharclass/POSIX Character Classes>). Thus, this modifier
371doesn't mean you can't use Unicode, it means that to get Unicode
372matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
373signals Unicode.
374
375As you would expect, this modifier causes, for example, C<\D> to mean
376the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
377C<\D>, C<\S>, and C<\W>. C<\b> still means to match at the boundary
378between C<\w> and C<\W>, using the C</a> definitions of them (similarly
379for C<\B>).
380
381Otherwise, C</a> behaves like the C</u> modifier, in that
382case-insensitive matching uses Unicode semantics; for example, "k" will
383match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
384points in the Latin1 range, above ASCII will have Unicode rules when it
385comes to case-insensitive matching.
386
387To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
388specify the "a" twice, for example C</aai> or C</aia>. (The first
389occurrence of "a" restricts the C<\d>, etc., and the second occurrence
390adds the C</i> restrictions.) But, note that code points outside the
391ASCII range will use Unicode rules for C</i> matching, so the modifier
392doesn't really restrict things to just ASCII; it just forbids the
393intermixing of ASCII and non-ASCII.
394
395To summarize, this modifier provides protection for applications that
396don't wish to be exposed to all of Unicode. Specifying it twice
397gives added protection.
398
399This modifier may be specified to be the default by C<use re '/a'>
400or C<use re '/aa'>. If you do so, you may actually have occasion to use
401the C</u> modifier explictly if there are a few regular expressions
402where you do want full Unicode rules (but even here, it's best if
403everything were under feature C<"unicode_strings">, along with the
404C<use re '/aa'>). Also see L</Which character set modifier is in
405effect?>.
406X</a>
407X</aa>
408
ed7efc79
KW
409=head4 Which character set modifier is in effect?
410
411Which of these modifiers is in effect at any given point in a regular
808432af
KW
412expression depends on a fairly complex set of interactions. These have
413been designed so that in general you don't have to worry about it, but
414this section gives the gory details. As
ed7efc79
KW
415explained below in L</Extended Patterns> it is possible to explicitly
416specify modifiers that apply only to portions of a regular expression.
417The innermost always has priority over any outer ones, and one applying
6368643f
KW
418to the whole expression has priority over any of the default settings that are
419described in the remainder of this section.
ed7efc79 420
916cec3f 421The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
ed7efc79
KW
422default modifiers (including these) for regular expressions compiled
423within its scope. This pragma has precedence over the other pragmas
516074bb 424listed below that also change the defaults.
ed7efc79
KW
425
426Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
66cbab2c 427and C<L<use feature 'unicode_strings|feature>>, or
ed7efc79
KW
428C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
429C</u> when not in the same scope as either C<L<use locale|perllocale>>
66cbab2c
KW
430or C<L<use bytes|bytes>>.
431(C<L<use locale ':not_characters'|perllocale/Unicode and UTF-8>> also
432sets the default to C</u>, overriding any plain C<use locale>.)
433Unlike the mechanisms mentioned above, these
6368643f
KW
434affect operations besides regular expressions pattern matching, and so
435give more consistent results with other operators, including using
436C<\U>, C<\l>, etc. in substitution replacements.
ed7efc79
KW
437
438If none of the above apply, for backwards compatibility reasons, the
439C</d> modifier is the one in effect by default. As this can lead to
440unexpected results, it is best to specify which other rule set should be
441used.
442
443=head4 Character set modifier behavior prior to Perl 5.14
444
445Prior to 5.14, there were no explicit modifiers, but C</l> was implied
446for regexes compiled within the scope of C<use locale>, and C</d> was
447implied otherwise. However, interpolating a regex into a larger regex
448would ignore the original compilation in favor of whatever was in effect
449at the time of the second compilation. There were a number of
450inconsistencies (bugs) with the C</d> modifier, where Unicode rules
451would be used when inappropriate, and vice versa. C<\p{}> did not imply
452Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
b6fa137b 453
a0d0e21e
LW
454=head2 Regular Expressions
455
04838cea
RGS
456=head3 Metacharacters
457
384f06ae 458The patterns used in Perl pattern matching evolved from those supplied in
14218588 459the Version 8 regex routines. (The routines are derived
19799a22
GS
460(distantly) from Henry Spencer's freely redistributable reimplementation
461of the V8 routines.) See L<Version 8 Regular Expressions> for
462details.
a0d0e21e
LW
463
464In particular the following metacharacters have their standard I<egrep>-ish
465meanings:
d74e8afc
ITB
466X<metacharacter>
467X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
468
a0d0e21e 469
f793d64a
KW
470 \ Quote the next metacharacter
471 ^ Match the beginning of the line
472 . Match any character (except newline)
473 $ Match the end of the line (or before newline at the end)
474 | Alternation
475 () Grouping
476 [] Bracketed Character class
a0d0e21e 477
14218588
GS
478By default, the "^" character is guaranteed to match only the
479beginning of the string, the "$" character only the end (or before the
480newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
481assumption that the string contains only one line. Embedded newlines
482will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 483string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
484newline within the string (except if the newline is the last character in
485the string), and "$" will match before any newline. At the
a0d0e21e
LW
486cost of a little more overhead, you can do this by using the /m modifier
487on the pattern match operator. (Older programs did this by setting C<$*>,
db7cd43a 488but this option was removed in perl 5.10.)
d74e8afc 489X<^> X<$> X</m>
a0d0e21e 490
14218588 491To simplify multi-line substitutions, the "." character never matches a
55497cff 492newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 493the string is a single line--even if it isn't.
d74e8afc 494X<.> X</s>
a0d0e21e 495
04838cea
RGS
496=head3 Quantifiers
497
a0d0e21e 498The following standard quantifiers are recognized:
d74e8afc 499X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 500
f793d64a
KW
501 * Match 0 or more times
502 + Match 1 or more times
503 ? Match 1 or 0 times
504 {n} Match exactly n times
505 {n,} Match at least n times
506 {n,m} Match at least n but not more than m times
a0d0e21e 507
0b928c2f 508(If a curly bracket occurs in any other context and does not form part of
4d68ffa0
KW
509a backslashed sequence like C<\x{...}>, it is treated as a regular
510character. In particular, the lower quantifier bound is not optional,
511and a typo in a quantifier silently causes it to be treated as the
512literal characters. For example,
9af81bfe
KW
513
514 /o{4,3}/
515
516looks like a quantifier that matches 0 times, since 4 is greater than 3,
517but it really means to match the sequence of six characters
4d68ffa0
KW
518S<C<"o { 4 , 3 }">>. It is planned to eventually require literal uses
519of curly brackets to be escaped, say by preceding them with a backslash
520or enclosing them within square brackets, (C<"\{"> or C<"[{]">). This
521change will allow for future syntax extensions (like making the lower
522bound of a quantifier optional), and better error checking. In the
523meantime, you should get in the habit of escaping all instances where
524you mean a literal "{".)
9af81bfe
KW
525
526The "*" quantifier is equivalent to C<{0,}>, the "+"
527e91da 527quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 528to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
529This is usually 32766 on the most common platforms. The actual limit can
530be seen in the error message generated by code such as this:
531
820475bd 532 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 533
54310121
PP
534By default, a quantified subpattern is "greedy", that is, it will match as
535many times as possible (given a particular starting location) while still
536allowing the rest of the pattern to match. If you want it to match the
537minimum number of times possible, follow the quantifier with a "?". Note
538that the meanings don't change, just the "greediness":
0d017f4d 539X<metacharacter> X<greedy> X<greediness>
d74e8afc 540X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 541
f793d64a
KW
542 *? Match 0 or more times, not greedily
543 +? Match 1 or more times, not greedily
544 ?? Match 0 or 1 time, not greedily
0b928c2f 545 {n}? Match exactly n times, not greedily (redundant)
f793d64a
KW
546 {n,}? Match at least n times, not greedily
547 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 548
b9b4dddf
YO
549By default, when a quantified subpattern does not allow the rest of the
550overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 551sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
552as well.
553
f793d64a
KW
554 *+ Match 0 or more times and give nothing back
555 ++ Match 1 or more times and give nothing back
556 ?+ Match 0 or 1 time and give nothing back
557 {n}+ Match exactly n times and give nothing back (redundant)
558 {n,}+ Match at least n times and give nothing back
559 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
560
561For instance,
562
563 'aaaa' =~ /a++a/
564
565will never match, as the C<a++> will gobble up all the C<a>'s in the
566string and won't leave any for the remaining part of the pattern. This
567feature can be extremely useful to give perl hints about where it
568shouldn't backtrack. For instance, the typical "match a double-quoted
569string" problem can be most efficiently performed when written as:
570
571 /"(?:[^"\\]++|\\.)*+"/
572
0d017f4d 573as we know that if the final quote does not match, backtracking will not
0b928c2f
FC
574help. See the independent subexpression
575L</C<< (?>pattern) >>> for more details;
b9b4dddf
YO
576possessive quantifiers are just syntactic sugar for that construct. For
577instance the above example could also be written as follows:
578
579 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
580
04838cea
RGS
581=head3 Escape sequences
582
0b928c2f 583Because patterns are processed as double-quoted strings, the following
a0d0e21e
LW
584also work:
585
f793d64a
KW
586 \t tab (HT, TAB)
587 \n newline (LF, NL)
588 \r return (CR)
589 \f form feed (FF)
590 \a alarm (bell) (BEL)
591 \e escape (think troff) (ESC)
f793d64a 592 \cK control char (example: VT)
dc0d9c48 593 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 594 \N{name} named Unicode character or character sequence
f793d64a 595 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 596 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
597 \l lowercase next char (think vi)
598 \u uppercase next char (think vi)
599 \L lowercase till \E (think vi)
600 \U uppercase till \E (think vi)
601 \Q quote (disable) pattern metacharacters till \E
602 \E end either case modification or quoted section, think vi
a0d0e21e 603
9bb1f947 604Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 605
e1d1eefb 606=head3 Character Classes and other Special Escapes
04838cea 607
a0d0e21e 608In addition, Perl defines the following:
d0b16107 609X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 610
f793d64a
KW
611 Sequence Note Description
612 [...] [1] Match a character according to the rules of the
613 bracketed character class defined by the "...".
614 Example: [a-z] matches "a" or "b" or "c" ... or "z"
615 [[:...:]] [2] Match a character according to the rules of the POSIX
616 character class "..." within the outer bracketed
617 character class. Example: [[:upper:]] matches any
618 uppercase character.
572224ce 619 (?[...]) [8] Extended bracketed character class
d35dd6c6
KW
620 \w [3] Match a "word" character (alphanumeric plus "_", plus
621 other connector punctuation chars plus Unicode
0b928c2f 622 marks)
f793d64a
KW
623 \W [3] Match a non-"word" character
624 \s [3] Match a whitespace character
625 \S [3] Match a non-whitespace character
626 \d [3] Match a decimal digit character
627 \D [3] Match a non-digit character
628 \pP [3] Match P, named property. Use \p{Prop} for longer names
629 \PP [3] Match non-P
630 \X [4] Match Unicode "eXtended grapheme cluster"
631 \C Match a single C-language char (octet) even if that is
632 part of a larger UTF-8 character. Thus it breaks up
633 characters into their UTF-8 bytes, so you may end up
634 with malformed pieces of UTF-8. Unsupported in
635 lookbehind.
c27a5cfe 636 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
637 '1' may actually be any positive integer.
638 \g1 [5] Backreference to a specific or previous group,
639 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 640 previous group and may optionally be wrapped in
f793d64a
KW
641 curly brackets for safer parsing.
642 \g{name} [5] Named backreference
643 \k<name> [5] Named backreference
644 \K [6] Keep the stuff left of the \K, don't include it in $&
2171640d 645 \N [7] Any character but \n. Not affected by /s modifier
f793d64a
KW
646 \v [3] Vertical whitespace
647 \V [3] Not vertical whitespace
648 \h [3] Horizontal whitespace
649 \H [3] Not horizontal whitespace
650 \R [4] Linebreak
e1d1eefb 651
9bb1f947
KW
652=over 4
653
654=item [1]
655
656See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 657
9bb1f947 658=item [2]
b8c5462f 659
9bb1f947 660See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 661
9bb1f947 662=item [3]
5496314a 663
9bb1f947 664See L<perlrecharclass/Backslash sequences> for details.
5496314a 665
9bb1f947 666=item [4]
5496314a 667
9bb1f947 668See L<perlrebackslash/Misc> for details.
d0b16107 669
9bb1f947 670=item [5]
b8c5462f 671
c27a5cfe 672See L</Capture groups> below for details.
93733859 673
9bb1f947 674=item [6]
b8c5462f 675
9bb1f947
KW
676See L</Extended Patterns> below for details.
677
678=item [7]
679
680Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
681character or character sequence whose name is C<NAME>; and similarly
682when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
683code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947 684
572224ce
KW
685=item [8]
686
687See L<perlrecharclass/Extended Bracketed Character Classes> for details.
688
9bb1f947 689=back
d0b16107 690
04838cea
RGS
691=head3 Assertions
692
a0d0e21e 693Perl defines the following zero-width assertions:
d74e8afc
ITB
694X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
695X<regexp, zero-width assertion>
696X<regular expression, zero-width assertion>
697X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 698
9bb1f947
KW
699 \b Match a word boundary
700 \B Match except at a word boundary
701 \A Match only at beginning of string
702 \Z Match only at end of string, or before newline at the end
703 \z Match only at end of string
704 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 705 of prior m//g)
a0d0e21e 706
14218588 707A word boundary (C<\b>) is a spot between two characters
19799a22
GS
708that has a C<\w> on one side of it and a C<\W> on the other side
709of it (in either order), counting the imaginary characters off the
710beginning and end of the string as matching a C<\W>. (Within
711character classes C<\b> represents backspace rather than a word
712boundary, just as it normally does in any double-quoted string.)
713The C<\A> and C<\Z> are just like "^" and "$", except that they
714won't match multiple times when the C</m> modifier is used, while
715"^" and "$" will match at every internal line boundary. To match
716the actual end of the string and not ignore an optional trailing
717newline, use C<\z>.
d74e8afc 718X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
719
720The C<\G> assertion can be used to chain global matches (using
721C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
722It is also useful when writing C<lex>-like scanners, when you have
723several patterns that you want to match against consequent substrings
0b928c2f 724of your string; see the previous reference. The actual location
19799a22 725where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d 726an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
0b928c2f
FC
727matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
728is modified somewhat, in that contents to the left of C<\G> are
58e23c8d
YO
729not counted when determining the length of the match. Thus the following
730will not match forever:
d74e8afc 731X<\G>
c47ff5f1 732
e761bb84
CO
733 my $string = 'ABC';
734 pos($string) = 1;
735 while ($string =~ /(.\G)/g) {
736 print $1;
737 }
58e23c8d
YO
738
739It will print 'A' and then terminate, as it considers the match to
740be zero-width, and thus will not match at the same position twice in a
741row.
742
743It is worth noting that C<\G> improperly used can result in an infinite
744loop. Take care when using patterns that include C<\G> in an alternation.
745
c27a5cfe 746=head3 Capture groups
04838cea 747
c27a5cfe
KW
748The bracketing construct C<( ... )> creates capture groups (also referred to as
749capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
750the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
751for the second, and so on.
752This is called a I<backreference>.
d74e8afc 753X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 754X<regex, capture group> X<regexp, capture group>
d74e8afc 755X<regular expression, capture buffer> X<backreference>
c27a5cfe 756X<regular expression, capture group> X<backreference>
1f1031fe 757X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
758X<named capture buffer> X<regular expression, named capture buffer>
759X<named capture group> X<regular expression, named capture group>
760X<%+> X<$+{name}> X<< \k<name> >>
761There is no limit to the number of captured substrings that you may use.
762Groups are numbered with the leftmost open parenthesis being number 1, etc. If
763a group did not match, the associated backreference won't match either. (This
764can happen if the group is optional, or in a different branch of an
765alternation.)
766You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
767this form, described below.
768
769You can also refer to capture groups relatively, by using a negative number, so
770that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
771group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
772example:
5624f11d
YO
773
774 /
c27a5cfe
KW
775 (Y) # group 1
776 ( # group 2
777 (X) # group 3
778 \g{-1} # backref to group 3
779 \g{-3} # backref to group 1
5624f11d
YO
780 )
781 /x
782
d8b950dc
KW
783would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
784interpolate regexes into larger regexes and not have to worry about the
785capture groups being renumbered.
786
787You can dispense with numbers altogether and create named capture groups.
788The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
789reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
790also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
791I<name> must not begin with a number, nor contain hyphens.
792When different groups within the same pattern have the same name, any reference
793to that name assumes the leftmost defined group. Named groups count in
794absolute and relative numbering, and so can also be referred to by those
795numbers.
796(It's possible to do things with named capture groups that would otherwise
797require C<(??{})>.)
798
799Capture group contents are dynamically scoped and available to you outside the
800pattern until the end of the enclosing block or until the next successful
801match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
802You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
803etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
804
805Braces are required in referring to named capture groups, but are optional for
806absolute or relative numbered ones. Braces are safer when creating a regex by
807concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
808contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
809is probably not what you intended.
810
811The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
812there were no named nor relative numbered capture groups. Absolute numbered
0b928c2f
FC
813groups were referred to using C<\1>,
814C<\2>, etc., and this notation is still
d8b950dc
KW
815accepted (and likely always will be). But it leads to some ambiguities if
816there are more than 9 capture groups, as C<\10> could mean either the tenth
817capture group, or the character whose ordinal in octal is 010 (a backspace in
818ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
819only if at least 10 left parentheses have opened before it. Likewise C<\11> is
820a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
821And so on. C<\1> through C<\9> are always interpreted as backreferences.
822There are several examples below that illustrate these perils. You can avoid
823the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
824and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
825digits padded with leading zeros, since a leading zero implies an octal
826constant.
d8b950dc
KW
827
828The C<\I<digit>> notation also works in certain circumstances outside
ed7efc79 829the pattern. See L</Warning on \1 Instead of $1> below for details.
81714fb9 830
14218588 831Examples:
a0d0e21e
LW
832
833 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
834
d8b950dc 835 /(.)\g1/ # find first doubled char
81714fb9
YO
836 and print "'$1' is the first doubled character\n";
837
838 /(?<char>.)\k<char>/ # ... a different way
839 and print "'$+{char}' is the first doubled character\n";
840
d8b950dc 841 /(?'char'.)\g1/ # ... mix and match
81714fb9 842 and print "'$1' is the first doubled character\n";
c47ff5f1 843
14218588 844 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
845 $hours = $1;
846 $minutes = $2;
847 $seconds = $3;
a0d0e21e 848 }
c47ff5f1 849
9d860678
KW
850 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
851 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
852 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
853 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
854
855 $a = '(.)\1'; # Creates problems when concatenated.
856 $b = '(.)\g{1}'; # Avoids the problems.
857 "aa" =~ /${a}/; # True
858 "aa" =~ /${b}/; # True
859 "aa0" =~ /${a}0/; # False!
860 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
861 "aa\x08" =~ /${a}0/; # True!
862 "aa\x08" =~ /${b}0/; # False
9d860678 863
14218588
GS
864Several special variables also refer back to portions of the previous
865match. C<$+> returns whatever the last bracket match matched.
866C<$&> returns the entire matched string. (At one point C<$0> did
867also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
868everything before the matched string. C<$'> returns everything
869after the matched string. And C<$^N> contains whatever was matched by
870the most-recently closed group (submatch). C<$^N> can be used in
871extended patterns (see below), for example to assign a submatch to a
81714fb9 872variable.
d74e8afc 873X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 874
d8b950dc
KW
875These special variables, like the C<%+> hash and the numbered match variables
876(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
877until the end of the enclosing block or until the next successful
878match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
879X<$+> X<$^N> X<$&> X<$`> X<$'>
880X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
881
0d017f4d 882B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 883which makes it easier to write code that tests for a series of more
665e98b9
JH
884specific cases and remembers the best match.
885
d78f32f6
FC
886B<WARNING>: If your code is to run on Perl 5.16 or earlier,
887beware that once Perl sees that you need one of C<$&>, C<$`>, or
14218588 888C<$'> anywhere in the program, it has to provide them for every
d78f32f6
FC
889pattern match. This may substantially slow your program. (In Perl 5.18 a
890more efficient mechanism is used, eliminating any slowdown.) Perl
d8b950dc 891uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
892price for each pattern that contains capturing parentheses. (To
893avoid this cost while retaining the grouping behaviour, use the
894extended regular expression C<(?: ... )> instead.) But if you never
895use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
896parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
897if you can, but if you can't (and some algorithms really appreciate
898them), once you've used them once, use them at will, because you've
d78f32f6 899already paid the price.
d74e8afc 900X<$&> X<$`> X<$'>
68dc0745 901
d78f32f6 902As a workaround for this problem, Perl 5.10.0 introduced C<${^PREMATCH}>,
cde0cee5
YO
903C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
904and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 905successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
906The use of these variables incurs no global performance penalty, unlike
907their punctuation char equivalents, however at the trade-off that you
d78f32f6
FC
908have to tell perl when you want to use them. As of Perl 5.18, these three
909variables are equivalent to C<$`>, C<$&> and C<$'>, and C</p> is ignored.
87e95b7f 910X</p> X<p modifier>
cde0cee5 911
9d727203
KW
912=head2 Quoting metacharacters
913
19799a22
GS
914Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
915C<\w>, C<\n>. Unlike some other regular expression languages, there
916are no backslashed symbols that aren't alphanumeric. So anything
0f264506 917that looks like \\, \(, \), \[, \], \{, or \} is always
19799a22
GS
918interpreted as a literal character, not a metacharacter. This was
919once used in a common idiom to disable or quote the special meanings
920of regular expression metacharacters in a string that you want to
36bbe248 921use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
922
923 $pattern =~ s/(\W)/\\$1/g;
924
f1cbbd6e 925(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
926Today it is more common to use the quotemeta() function or the C<\Q>
927metaquoting escape sequence to disable all metacharacters' special
928meanings like this:
a0d0e21e
LW
929
930 /$unquoted\Q$quoted\E$unquoted/
931
9da458fc
IZ
932Beware that if you put literal backslashes (those not inside
933interpolated variables) between C<\Q> and C<\E>, double-quotish
934backslash interpolation may lead to confusing results. If you
935I<need> to use literal backslashes within C<\Q...\E>,
936consult L<perlop/"Gory details of parsing quoted constructs">.
937
736fe711
KW
938C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
939
19799a22
GS
940=head2 Extended Patterns
941
14218588 942Perl also defines a consistent extension syntax for features not
0b928c2f
FC
943found in standard tools like B<awk> and
944B<lex>. The syntax for most of these is a
14218588
GS
945pair of parentheses with a question mark as the first thing within
946the parentheses. The character after the question mark indicates
947the extension.
19799a22 948
14218588
GS
949The stability of these extensions varies widely. Some have been
950part of the core language for many years. Others are experimental
951and may change without warning or be completely removed. Check
952the documentation on an individual feature to verify its current
953status.
19799a22 954
14218588
GS
955A question mark was chosen for this and for the minimal-matching
956construct because 1) question marks are rare in older regular
957expressions, and 2) whenever you see one, you should stop and
0b928c2f 958"question" exactly what is going on. That's psychology....
a0d0e21e 959
70ca8714 960=over 4
a0d0e21e 961
cc6b7395 962=item C<(?#text)>
d74e8afc 963X<(?#)>
a0d0e21e 964
14218588 965A comment. The text is ignored. If the C</x> modifier enables
19799a22 966whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
967the comment as soon as it sees a C<)>, so there is no way to put a literal
968C<)> in the comment.
a0d0e21e 969
cfaf538b 970=item C<(?adlupimsx-imsx)>
fb85c044 971
cfaf538b 972=item C<(?^alupimsx)>
fb85c044 973X<(?)> X<(?^)>
19799a22 974
0b6d1084
JH
975One or more embedded pattern-match modifiers, to be turned on (or
976turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
977the remainder of the enclosing pattern group (if any).
978
fb85c044 979This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 980configuration file, taken from an argument, or specified in a table
0b928c2f
FC
981somewhere. Consider the case where some patterns want to be
982case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 983include C<(?i)> at the front of the pattern. For example:
19799a22
GS
984
985 $pattern = "foobar";
5d458dd8 986 if ( /$pattern/i ) { }
19799a22
GS
987
988 # more flexible:
989
990 $pattern = "(?i)foobar";
5d458dd8 991 if ( /$pattern/ ) { }
19799a22 992
0b6d1084 993These modifiers are restored at the end of the enclosing group. For example,
19799a22 994
d8b950dc 995 ( (?i) blah ) \s+ \g1
19799a22 996
0d017f4d
WL
997will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
998repetition of the previous word, assuming the C</x> modifier, and no C</i>
999modifier outside this group.
19799a22 1000
8eb5594e 1001These modifiers do not carry over into named subpatterns called in the
dd72e27b 1002enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
8eb5594e
DR
1003change the case-sensitivity of the "NAME" pattern.
1004
dc925305
KW
1005Any of these modifiers can be set to apply globally to all regular
1006expressions compiled within the scope of a C<use re>. See
a0bbd6ff 1007L<re/"'/flags' mode">.
dc925305 1008
9de15fec
KW
1009Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1010after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
1011C<"d">) may follow the caret to override it.
1012But a minus sign is not legal with it.
1013
dc925305 1014Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 1015that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 1016C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
1017others, and a maximum of one (or two C<a>'s) may appear in the
1018construct. Thus, for
0b928c2f 1019example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 1020C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1021
1022Note also that the C<p> modifier is special in that its presence
1023anywhere in a pattern has a global effect.
cde0cee5 1024
5a964f20 1025=item C<(?:pattern)>
d74e8afc 1026X<(?:)>
a0d0e21e 1027
cfaf538b 1028=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1029
cfaf538b 1030=item C<(?^aluimsx:pattern)>
fb85c044
KW
1031X<(?^:)>
1032
5a964f20
TC
1033This is for clustering, not capturing; it groups subexpressions like
1034"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1035
5a964f20 1036 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1037
1038is like
1039
5a964f20 1040 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1041
19799a22
GS
1042but doesn't spit out extra fields. It's also cheaper not to capture
1043characters if you don't need to.
a0d0e21e 1044
19799a22 1045Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1046C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1047
1048 /(?s-i:more.*than).*million/i
1049
14218588 1050is equivalent to the more verbose
ca9dfc88
IZ
1051
1052 /(?:(?s-i)more.*than).*million/i
1053
fb85c044 1054Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1055after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1056flags (except C<"d">) may follow the caret, so
fb85c044
KW
1057
1058 (?^x:foo)
1059
1060is equivalent to
1061
1062 (?x-ims:foo)
1063
1064The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1065surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1066modified by any flags specified.
1067
1068The caret allows for simpler stringification of compiled regular
1069expressions. These look like
1070
1071 (?^:pattern)
1072
1073with any non-default flags appearing between the caret and the colon.
1074A test that looks at such stringification thus doesn't need to have the
1075system default flags hard-coded in it, just the caret. If new flags are
1076added to Perl, the meaning of the caret's expansion will change to include
1077the default for those flags, so the test will still work, unchanged.
1078
1079Specifying a negative flag after the caret is an error, as the flag is
1080redundant.
1081
1082Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1083to match at the beginning.
1084
594d7033
YO
1085=item C<(?|pattern)>
1086X<(?|)> X<Branch reset>
1087
1088This is the "branch reset" pattern, which has the special property
c27a5cfe 1089that the capture groups are numbered from the same starting point
99d59c4d 1090in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1091
c27a5cfe 1092Capture groups are numbered from left to right, but inside this
693596a8 1093construct the numbering is restarted for each branch.
4deaaa80 1094
c27a5cfe 1095The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1096following this construct will be numbered as though the construct
1097contained only one branch, that being the one with the most capture
c27a5cfe 1098groups in it.
4deaaa80 1099
0b928c2f 1100This construct is useful when you want to capture one of a
4deaaa80
PJ
1101number of alternative matches.
1102
1103Consider the following pattern. The numbers underneath show in
c27a5cfe 1104which group the captured content will be stored.
594d7033
YO
1105
1106
1107 # before ---------------branch-reset----------- after
1108 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1109 # 1 2 2 3 2 3 4
1110
ab106183
A
1111Be careful when using the branch reset pattern in combination with
1112named captures. Named captures are implemented as being aliases to
c27a5cfe 1113numbered groups holding the captures, and that interferes with the
ab106183
A
1114implementation of the branch reset pattern. If you are using named
1115captures in a branch reset pattern, it's best to use the same names,
1116in the same order, in each of the alternations:
1117
1118 /(?| (?<a> x ) (?<b> y )
1119 | (?<a> z ) (?<b> w )) /x
1120
1121Not doing so may lead to surprises:
1122
1123 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1124 say $+ {a}; # Prints '12'
1125 say $+ {b}; # *Also* prints '12'.
1126
c27a5cfe
KW
1127The problem here is that both the group named C<< a >> and the group
1128named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1129
ee9b8eae
YO
1130=item Look-Around Assertions
1131X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1132
0b928c2f 1133Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1134pattern without including it in C<$&>. Positive assertions match when
1135their subpattern matches, negative assertions match when their subpattern
1136fails. Look-behind matches text up to the current match position,
1137look-ahead matches text following the current match position.
1138
1139=over 4
1140
5a964f20 1141=item C<(?=pattern)>
d74e8afc 1142X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1143
19799a22 1144A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1145matches a word followed by a tab, without including the tab in C<$&>.
1146
5a964f20 1147=item C<(?!pattern)>
d74e8afc 1148X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1149
19799a22 1150A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1151matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1152however that look-ahead and look-behind are NOT the same thing. You cannot
1153use this for look-behind.
7b8d334a 1154
5a964f20 1155If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1156will not do what you want. That's because the C<(?!foo)> is just saying that
1157the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1158match. Use look-behind instead (see below).
c277df42 1159
ee9b8eae
YO
1160=item C<(?<=pattern)> C<\K>
1161X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1162
c47ff5f1 1163A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1164matches a word that follows a tab, without including the tab in C<$&>.
1165Works only for fixed-width look-behind.
c277df42 1166
ee9b8eae
YO
1167There is a special form of this construct, called C<\K>, which causes the
1168regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1169not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1170look-behind. The use of C<\K> inside of another look-around assertion
1171is allowed, but the behaviour is currently not well defined.
1172
c62285ac 1173For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1174equivalent C<< (?<=...) >> construct, and it is especially useful in
1175situations where you want to efficiently remove something following
1176something else in a string. For instance
1177
1178 s/(foo)bar/$1/g;
1179
1180can be rewritten as the much more efficient
1181
1182 s/foo\Kbar//g;
1183
5a964f20 1184=item C<(?<!pattern)>
d74e8afc 1185X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1186
19799a22
GS
1187A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1188matches any occurrence of "foo" that does not follow "bar". Works
1189only for fixed-width look-behind.
c277df42 1190
ee9b8eae
YO
1191=back
1192
81714fb9
YO
1193=item C<(?'NAME'pattern)>
1194
1195=item C<< (?<NAME>pattern) >>
1196X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1197
c27a5cfe 1198A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1199parentheses C<()> but for the additional fact that the group
1200can be referred to by name in various regular expression
1201constructs (like C<\g{NAME}>) and can be accessed by name
1202after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1203for more details on the C<%+> and C<%-> hashes.
81714fb9 1204
c27a5cfe
KW
1205If multiple distinct capture groups have the same name then the
1206$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1207
0d017f4d 1208The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1209
1210B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1211function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1212numbered sequentially regardless of being named or not. Thus in the
1213pattern
1214
1215 /(x)(?<foo>y)(z)/
1216
1217$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1218the opposite which is what a .NET regex hacker might expect.
1219
1f1031fe
YO
1220Currently NAME is restricted to simple identifiers only.
1221In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1222its Unicode extension (see L<utf8>),
1223though it isn't extended by the locale (see L<perllocale>).
81714fb9 1224
1f1031fe 1225B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1226with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1227may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1228support the use of single quotes as a delimiter for the name.
81714fb9 1229
1f1031fe
YO
1230=item C<< \k<NAME> >>
1231
1232=item C<< \k'NAME' >>
81714fb9
YO
1233
1234Named backreference. Similar to numeric backreferences, except that
1235the group is designated by name and not number. If multiple groups
1236have the same name then it refers to the leftmost defined group in
1237the current match.
1238
0d017f4d 1239It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1240earlier in the pattern.
1241
1242Both forms are equivalent.
1243
1f1031fe 1244B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1245with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1246may be used instead of C<< \k<NAME> >>.
1f1031fe 1247
cc6b7395 1248=item C<(?{ code })>
d74e8afc 1249X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1250
19799a22 1251B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1252experimental, and may be changed without notice. Code executed that
1253has side effects may not perform identically from version to version
e128ab2c
DM
1254due to the effect of future optimisations in the regex engine. The
1255implementation of this feature was radically overhauled for the 5.18.0
1256release, and its behaviour in earlier versions of perl was much buggier,
1257especially in relation to parsing, lexical vars, scoping, recursion and
1258reentrancy.
c277df42 1259
e128ab2c
DM
1260This zero-width assertion executes any embedded Perl code. It always
1261succeeds, and its return value is set as C<$^R>.
19799a22 1262
e128ab2c
DM
1263In literal patterns, the code is parsed at the same time as the
1264surrounding code. While within the pattern, control is passed temporarily
1265back to the perl parser, until the logically-balancing closing brace is
1266encountered. This is similar to the way that an array index expression in
1267a literal string is handled, for example
77ea4f6d 1268
e128ab2c
DM
1269 "abc$array[ 1 + f('[') + g()]def"
1270
1271In particular, braces do not need to be balanced:
1272
576fa024 1273 s/abc(?{ f('{'); })/def/
e128ab2c
DM
1274
1275Even in a pattern that is interpolated and compiled at run-time, literal
1276code blocks will be compiled once, at perl compile time; the following
1277prints "ABCD":
1278
1279 print "D";
1280 my $qr = qr/(?{ BEGIN { print "A" } })/;
1281 my $foo = "foo";
1282 /$foo$qr(?{ BEGIN { print "B" } })/;
1283 BEGIN { print "C" }
1284
1285In patterns where the text of the code is derived from run-time
1286information rather than appearing literally in a source code /pattern/,
1287the code is compiled at the same time that the pattern is compiled, and
5771dda0 1288for reasons of security, C<use re 'eval'> must be in scope. This is to
e128ab2c
DM
1289stop user-supplied patterns containing code snippets from being
1290executable.
1291
5771dda0 1292In situations where you need to enable this with C<use re 'eval'>, you should
e128ab2c
DM
1293also have taint checking enabled. Better yet, use the carefully
1294constrained evaluation within a Safe compartment. See L<perlsec> for
1295details about both these mechanisms.
1296
1297From the viewpoint of parsing, lexical variable scope and closures,
1298
1299 /AAA(?{ BBB })CCC/
1300
1301behaves approximately like
1302
1303 /AAA/ && do { BBB } && /CCC/
1304
1305Similarly,
1306
1307 qr/AAA(?{ BBB })CCC/
1308
1309behaves approximately like
77ea4f6d 1310
e128ab2c
DM
1311 sub { /AAA/ && do { BBB } && /CCC/ }
1312
1313In particular:
1314
1315 { my $i = 1; $r = qr/(?{ print $i })/ }
1316 my $i = 2;
1317 /$r/; # prints "1"
1318
1319Inside a C<(?{...})> block, C<$_> refers to the string the regular
754091cb 1320expression is matching against. You can also use C<pos()> to know what is
fa11829f 1321the current position of matching within this string.
754091cb 1322
e128ab2c
DM
1323The code block introduces a new scope from the perspective of lexical
1324variable declarations, but B<not> from the perspective of C<local> and
1325similar localizing behaviours. So later code blocks within the same
1326pattern will still see the values which were localized in earlier blocks.
1327These accumulated localizations are undone either at the end of a
1328successful match, or if the assertion is backtracked (compare
1329L<"Backtracking">). For example,
b9ac3b5b
GS
1330
1331 $_ = 'a' x 8;
5d458dd8 1332 m<
d1fbf752 1333 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1334 (
5d458dd8 1335 a
b9ac3b5b 1336 (?{
d1fbf752
KW
1337 local $cnt = $cnt + 1; # Update $cnt,
1338 # backtracking-safe.
b9ac3b5b 1339 })
5d458dd8 1340 )*
b9ac3b5b 1341 aaaa
d1fbf752
KW
1342 (?{ $res = $cnt }) # On success copy to
1343 # non-localized location.
b9ac3b5b
GS
1344 >x;
1345
e128ab2c
DM
1346will initially increment C<$cnt> up to 8; then during backtracking, its
1347value will be unwound back to 4, which is the value assigned to C<$res>.
1348At the end of the regex execution, $cnt will be wound back to its initial
1349value of 0.
1350
1351This assertion may be used as the condition in a
b9ac3b5b 1352
e128ab2c
DM
1353 (?(condition)yes-pattern|no-pattern)
1354
1355switch. If I<not> used in this way, the result of evaluation of C<code>
1356is put into the special variable C<$^R>. This happens immediately, so
1357C<$^R> can be used from other C<(?{ code })> assertions inside the same
1358regular expression.
b9ac3b5b 1359
19799a22
GS
1360The assignment to C<$^R> above is properly localized, so the old
1361value of C<$^R> is restored if the assertion is backtracked; compare
1362L<"Backtracking">.
b9ac3b5b 1363
e128ab2c
DM
1364Note that the special variable C<$^N> is particularly useful with code
1365blocks to capture the results of submatches in variables without having to
1366keep track of the number of nested parentheses. For example:
1367
1368 $_ = "The brown fox jumps over the lazy dog";
1369 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1370 print "color = $color, animal = $animal\n";
1371
8988a1bb 1372
14455d6c 1373=item C<(??{ code })>
d74e8afc
ITB
1374X<(??{})>
1375X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1376
19799a22 1377B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1378experimental, and may be changed without notice. Code executed that
1379has side effects may not perform identically from version to version
1380due to the effect of future optimisations in the regex engine.
0f5d15d6 1381
e128ab2c
DM
1382This is a "postponed" regular subexpression. It behaves in I<exactly> the
1383same way as a C<(?{ code })> code block as described above, except that
1384its return value, rather than being assigned to C<$^R>, is treated as a
1385pattern, compiled if it's a string (or used as-is if its a qr// object),
1386then matched as if it were inserted instead of this construct.
6bda09f9 1387
e128ab2c
DM
1388During the matching of this sub-pattern, it has its own set of
1389captures which are valid during the sub-match, but are discarded once
1390control returns to the main pattern. For example, the following matches,
1391with the inner pattern capturing "B" and matching "BB", while the outer
1392pattern captures "A";
1393
1394 my $inner = '(.)\1';
1395 "ABBA" =~ /^(.)(??{ $inner })\1/;
1396 print $1; # prints "A";
6bda09f9 1397
e128ab2c
DM
1398Note that this means that there is no way for the inner pattern to refer
1399to a capture group defined outside. (The code block itself can use C<$1>,
1400etc., to refer to the enclosing pattern's capture groups.) Thus, although
0f5d15d6 1401
e128ab2c
DM
1402 ('a' x 100)=~/(??{'(.)' x 100})/
1403
1404I<will> match, it will I<not> set $1 on exit.
19799a22
GS
1405
1406The following pattern matches a parenthesized group:
0f5d15d6 1407
d1fbf752
KW
1408 $re = qr{
1409 \(
1410 (?:
1411 (?> [^()]+ ) # Non-parens without backtracking
1412 |
1413 (??{ $re }) # Group with matching parens
1414 )*
1415 \)
1416 }x;
0f5d15d6 1417
6bda09f9
YO
1418See also C<(?PARNO)> for a different, more efficient way to accomplish
1419the same task.
1420
e128ab2c
DM
1421Executing a postponed regular expression 50 times without consuming any
1422input string will result in a fatal error. The maximum depth is compiled
1423into perl, so changing it requires a custom build.
6bda09f9 1424
542fa716
YO
1425=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1426X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1427X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1428X<regex, relative recursion>
6bda09f9 1429
e128ab2c
DM
1430Similar to C<(??{ code })> except that it does not involve executing any
1431code or potentially compiling a returned pattern string; instead it treats
1432the part of the current pattern contained within a specified capture group
1433as an independent pattern that must match at the current position.
1434Capture groups contained by the pattern will have the value as determined
1435by the outermost recursion.
6bda09f9 1436
894be9b7 1437PARNO is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1438the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1439the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716 1440C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
c27a5cfe 1441to be relative, with negative numbers indicating preceding capture groups
542fa716 1442and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1443declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1444Note that the counting for relative recursion differs from that of
c27a5cfe 1445relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1446included.
6bda09f9 1447
81714fb9 1448The following pattern matches a function foo() which may contain
f145b7e9 1449balanced parentheses as the argument.
6bda09f9 1450
d1fbf752 1451 $re = qr{ ( # paren group 1 (full function)
81714fb9 1452 foo
d1fbf752 1453 ( # paren group 2 (parens)
6bda09f9 1454 \(
d1fbf752 1455 ( # paren group 3 (contents of parens)
6bda09f9 1456 (?:
d1fbf752 1457 (?> [^()]+ ) # Non-parens without backtracking
6bda09f9 1458 |
d1fbf752 1459 (?2) # Recurse to start of paren group 2
6bda09f9
YO
1460 )*
1461 )
1462 \)
1463 )
1464 )
1465 }x;
1466
1467If the pattern was used as follows
1468
1469 'foo(bar(baz)+baz(bop))'=~/$re/
1470 and print "\$1 = $1\n",
1471 "\$2 = $2\n",
1472 "\$3 = $3\n";
1473
1474the output produced should be the following:
1475
1476 $1 = foo(bar(baz)+baz(bop))
1477 $2 = (bar(baz)+baz(bop))
81714fb9 1478 $3 = bar(baz)+baz(bop)
6bda09f9 1479
c27a5cfe 1480If there is no corresponding capture group defined, then it is a
61528107 1481fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1482string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1483into perl, so changing it requires a custom build.
1484
542fa716
YO
1485The following shows how using negative indexing can make it
1486easier to embed recursive patterns inside of a C<qr//> construct
1487for later use:
1488
1489 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
c77257ed 1490 if (/foo $parens \s+ \+ \s+ bar $parens/x) {
542fa716
YO
1491 # do something here...
1492 }
1493
81714fb9 1494B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1495PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1496a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1497as atomic. Also, modifiers are resolved at compile time, so constructs
1498like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1499be processed.
6bda09f9 1500
894be9b7
YO
1501=item C<(?&NAME)>
1502X<(?&NAME)>
1503
0d017f4d
WL
1504Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1505parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1506the same name, then it recurses to the leftmost.
1507
1508It is an error to refer to a name that is not declared somewhere in the
1509pattern.
1510
1f1031fe
YO
1511B<NOTE:> In order to make things easier for programmers with experience
1512with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1513may be used instead of C<< (?&NAME) >>.
1f1031fe 1514
e2e6a0f1
YO
1515=item C<(?(condition)yes-pattern|no-pattern)>
1516X<(?()>
286f584a 1517
e2e6a0f1 1518=item C<(?(condition)yes-pattern)>
286f584a 1519
41ef34de
ML
1520Conditional expression. Matches C<yes-pattern> if C<condition> yields
1521a true value, matches C<no-pattern> otherwise. A missing pattern always
1522matches.
1523
25e26d77 1524C<(condition)> should be one of: 1) an integer in
e2e6a0f1 1525parentheses (which is valid if the corresponding pair of parentheses
25e26d77 1526matched); 2) a look-ahead/look-behind/evaluate zero-width assertion; 3) a
c27a5cfe 1527name in angle brackets or single quotes (which is valid if a group
25e26d77 1528with the given name matched); or 4) the special symbol (R) (true when
e2e6a0f1
YO
1529evaluated inside of recursion or eval). Additionally the R may be
1530followed by a number, (which will be true when evaluated when recursing
1531inside of the appropriate group), or by C<&NAME>, in which case it will
1532be true only when evaluated during recursion in the named group.
1533
1534Here's a summary of the possible predicates:
1535
1536=over 4
1537
1538=item (1) (2) ...
1539
c27a5cfe 1540Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1541
1542=item (<NAME>) ('NAME')
1543
c27a5cfe 1544Checks if a group with the given name has matched something.
e2e6a0f1 1545
f01cd190
FC
1546=item (?=...) (?!...) (?<=...) (?<!...)
1547
1548Checks whether the pattern matches (or does not match, for the '!'
1549variants).
1550
e2e6a0f1
YO
1551=item (?{ CODE })
1552
f01cd190 1553Treats the return value of the code block as the condition.
e2e6a0f1
YO
1554
1555=item (R)
1556
1557Checks if the expression has been evaluated inside of recursion.
1558
1559=item (R1) (R2) ...
1560
1561Checks if the expression has been evaluated while executing directly
1562inside of the n-th capture group. This check is the regex equivalent of
1563
1564 if ((caller(0))[3] eq 'subname') { ... }
1565
1566In other words, it does not check the full recursion stack.
1567
1568=item (R&NAME)
1569
1570Similar to C<(R1)>, this predicate checks to see if we're executing
1571directly inside of the leftmost group with a given name (this is the same
1572logic used by C<(?&NAME)> to disambiguate). It does not check the full
1573stack, but only the name of the innermost active recursion.
1574
1575=item (DEFINE)
1576
1577In this case, the yes-pattern is never directly executed, and no
1578no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1579See below for details.
1580
1581=back
1582
1583For example:
1584
1585 m{ ( \( )?
1586 [^()]+
1587 (?(1) \) )
1588 }x
1589
1590matches a chunk of non-parentheses, possibly included in parentheses
1591themselves.
1592
0b928c2f
FC
1593A special form is the C<(DEFINE)> predicate, which never executes its
1594yes-pattern directly, and does not allow a no-pattern. This allows one to
1595define subpatterns which will be executed only by the recursion mechanism.
e2e6a0f1
YO
1596This way, you can define a set of regular expression rules that can be
1597bundled into any pattern you choose.
1598
1599It is recommended that for this usage you put the DEFINE block at the
1600end of the pattern, and that you name any subpatterns defined within it.
1601
1602Also, it's worth noting that patterns defined this way probably will
1603not be as efficient, as the optimiser is not very clever about
1604handling them.
1605
1606An example of how this might be used is as follows:
1607
2bf803e2 1608 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1609 (?(DEFINE)
2bf803e2
YO
1610 (?<NAME_PAT>....)
1611 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1612 )/x
1613
c27a5cfe
KW
1614Note that capture groups matched inside of recursion are not accessible
1615after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1616necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1617C<$+{NAME}> would be.
286f584a 1618
51a1303c
BF
1619Finally, keep in mind that subpatterns created inside a DEFINE block
1620count towards the absolute and relative number of captures, so this:
1621
1622 my @captures = "a" =~ /(.) # First capture
1623 (?(DEFINE)
1624 (?<EXAMPLE> 1 ) # Second capture
1625 )/x;
1626 say scalar @captures;
1627
1628Will output 2, not 1. This is particularly important if you intend to
1629compile the definitions with the C<qr//> operator, and later
1630interpolate them in another pattern.
1631
c47ff5f1 1632=item C<< (?>pattern) >>
6bda09f9 1633X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1634
19799a22
GS
1635An "independent" subexpression, one which matches the substring
1636that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1637position, and it matches I<nothing other than this substring>. This
19799a22
GS
1638construct is useful for optimizations of what would otherwise be
1639"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1640It may also be useful in places where the "grab all you can, and do not
1641give anything back" semantic is desirable.
19799a22 1642
c47ff5f1 1643For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1644(anchored at the beginning of string, as above) will match I<all>
1645characters C<a> at the beginning of string, leaving no C<a> for
1646C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1647since the match of the subgroup C<a*> is influenced by the following
1648group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1649C<a*ab> will match fewer characters than a standalone C<a*>, since
1650this makes the tail match.
1651
0b928c2f
FC
1652C<< (?>pattern) >> does not disable backtracking altogether once it has
1653matched. It is still possible to backtrack past the construct, but not
1654into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1655
c47ff5f1 1656An effect similar to C<< (?>pattern) >> may be achieved by writing
0b928c2f
FC
1657C<(?=(pattern))\g{-1}>. This matches the same substring as a standalone
1658C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
c47ff5f1 1659makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1660(The difference between these two constructs is that the second one
1661uses a capturing group, thus shifting ordinals of backreferences
1662in the rest of a regular expression.)
1663
1664Consider this pattern:
c277df42 1665
871b0233 1666 m{ \(
e2e6a0f1 1667 (
f793d64a 1668 [^()]+ # x+
e2e6a0f1 1669 |
871b0233
IZ
1670 \( [^()]* \)
1671 )+
e2e6a0f1 1672 \)
871b0233 1673 }x
5a964f20 1674
19799a22
GS
1675That will efficiently match a nonempty group with matching parentheses
1676two levels deep or less. However, if there is no such group, it
1677will take virtually forever on a long string. That's because there
1678are so many different ways to split a long string into several
1679substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1680to a subpattern of the above pattern. Consider how the pattern
1681above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1682seconds, but that each extra letter doubles this time. This
1683exponential performance will make it appear that your program has
14218588 1684hung. However, a tiny change to this pattern
5a964f20 1685
e2e6a0f1
YO
1686 m{ \(
1687 (
f793d64a 1688 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1689 |
871b0233
IZ
1690 \( [^()]* \)
1691 )+
e2e6a0f1 1692 \)
871b0233 1693 }x
c277df42 1694
c47ff5f1 1695which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1696this yourself would be a productive exercise), but finishes in a fourth
1697the time when used on a similar string with 1000000 C<a>s. Be aware,
0b928c2f
FC
1698however, that, when this construct is followed by a
1699quantifier, it currently triggers a warning message under
9f1b1f2d 1700the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1701C<"matches null string many times in regex">.
c277df42 1702
c47ff5f1 1703On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1704effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1705This was only 4 times slower on a string with 1000000 C<a>s.
1706
9da458fc
IZ
1707The "grab all you can, and do not give anything back" semantic is desirable
1708in many situations where on the first sight a simple C<()*> looks like
1709the correct solution. Suppose we parse text with comments being delimited
1710by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1711its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1712the comment delimiter, because it may "give up" some whitespace if
1713the remainder of the pattern can be made to match that way. The correct
1714answer is either one of these:
1715
1716 (?>#[ \t]*)
1717 #[ \t]*(?![ \t])
1718
1719For example, to grab non-empty comments into $1, one should use either
1720one of these:
1721
1722 / (?> \# [ \t]* ) ( .+ ) /x;
1723 / \# [ \t]* ( [^ \t] .* ) /x;
1724
1725Which one you pick depends on which of these expressions better reflects
1726the above specification of comments.
1727
6bda09f9
YO
1728In some literature this construct is called "atomic matching" or
1729"possessive matching".
1730
b9b4dddf
YO
1731Possessive quantifiers are equivalent to putting the item they are applied
1732to inside of one of these constructs. The following equivalences apply:
1733
1734 Quantifier Form Bracketing Form
1735 --------------- ---------------
1736 PAT*+ (?>PAT*)
1737 PAT++ (?>PAT+)
1738 PAT?+ (?>PAT?)
1739 PAT{min,max}+ (?>PAT{min,max})
1740
9d1a5160 1741=item C<(?[ ])>
f4f5fe57 1742
572224ce 1743See L<perlrecharclass/Extended Bracketed Character Classes>.
9d1a5160 1744
e2e6a0f1
YO
1745=back
1746
1747=head2 Special Backtracking Control Verbs
1748
1749B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1750removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1751be noted to avoid problems during upgrades.
1752
1753These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1754otherwise stated the ARG argument is optional; in some cases, it is
1755forbidden.
1756
1757Any pattern containing a special backtracking verb that allows an argument
e1020413 1758has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1759C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1760rules apply:
e2e6a0f1 1761
5d458dd8
YO
1762On failure, the C<$REGERROR> variable will be set to the ARG value of the
1763verb pattern, if the verb was involved in the failure of the match. If the
1764ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1765name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1766none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1767
5d458dd8
YO
1768On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1769the C<$REGMARK> variable will be set to the name of the last
1770C<(*MARK:NAME)> pattern executed. See the explanation for the
1771C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1772
5d458dd8 1773B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1774and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1775readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1776Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1777
1778If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1779argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1780
70ca8714 1781=over 3
e2e6a0f1
YO
1782
1783=item Verbs that take an argument
1784
1785=over 4
1786
5d458dd8 1787=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1788X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1789
5d458dd8
YO
1790This zero-width pattern prunes the backtracking tree at the current point
1791when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1792where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1793A may backtrack as necessary to match. Once it is reached, matching
1794continues in B, which may also backtrack as necessary; however, should B
1795not match, then no further backtracking will take place, and the pattern
1796will fail outright at the current starting position.
54612592
YO
1797
1798The following example counts all the possible matching strings in a
1799pattern (without actually matching any of them).
1800
e2e6a0f1 1801 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1802 print "Count=$count\n";
1803
1804which produces:
1805
1806 aaab
1807 aaa
1808 aa
1809 a
1810 aab
1811 aa
1812 a
1813 ab
1814 a
1815 Count=9
1816
5d458dd8 1817If we add a C<(*PRUNE)> before the count like the following
54612592 1818
5d458dd8 1819 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1820 print "Count=$count\n";
1821
0b928c2f 1822we prevent backtracking and find the count of the longest matching string
353c6505 1823at each matching starting point like so:
54612592
YO
1824
1825 aaab
1826 aab
1827 ab
1828 Count=3
1829
5d458dd8 1830Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1831
5d458dd8
YO
1832See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1833control backtracking. In some cases, the use of C<(*PRUNE)> can be
1834replaced with a C<< (?>pattern) >> with no functional difference; however,
1835C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1836C<< (?>pattern) >> alone.
54612592 1837
5d458dd8
YO
1838=item C<(*SKIP)> C<(*SKIP:NAME)>
1839X<(*SKIP)>
e2e6a0f1 1840
5d458dd8 1841This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1842failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1843to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1844of this pattern. This effectively means that the regex engine "skips" forward
1845to this position on failure and tries to match again, (assuming that
1846there is sufficient room to match).
1847
1848The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1849C<(*MARK:NAME)> was encountered while matching, then it is that position
1850which is used as the "skip point". If no C<(*MARK)> of that name was
1851encountered, then the C<(*SKIP)> operator has no effect. When used
1852without a name the "skip point" is where the match point was when
1853executing the (*SKIP) pattern.
1854
0b928c2f 1855Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1856is twice as long:
1857
d1fbf752
KW
1858 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1859 print "Count=$count\n";
24b23f37
YO
1860
1861outputs
1862
1863 aaab
1864 aaab
1865 Count=2
1866
5d458dd8 1867Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1868executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1869C<(*SKIP)> was executed.
1870
5d458dd8 1871=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1872X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1873
1874This zero-width pattern can be used to mark the point reached in a string
1875when a certain part of the pattern has been successfully matched. This
1876mark may be given a name. A later C<(*SKIP)> pattern will then skip
1877forward to that point if backtracked into on failure. Any number of
b4222fa9 1878C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1879
1880In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1881can be used to "label" a pattern branch, so that after matching, the
1882program can determine which branches of the pattern were involved in the
1883match.
1884
1885When a match is successful, the C<$REGMARK> variable will be set to the
1886name of the most recently executed C<(*MARK:NAME)> that was involved
1887in the match.
1888
1889This can be used to determine which branch of a pattern was matched
c27a5cfe 1890without using a separate capture group for each branch, which in turn
5d458dd8
YO
1891can result in a performance improvement, as perl cannot optimize
1892C</(?:(x)|(y)|(z))/> as efficiently as something like
1893C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1894
1895When a match has failed, and unless another verb has been involved in
1896failing the match and has provided its own name to use, the C<$REGERROR>
1897variable will be set to the name of the most recently executed
1898C<(*MARK:NAME)>.
1899
42ac7c82 1900See L</(*SKIP)> for more details.
5d458dd8 1901
b62d2d15
YO
1902As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1903
5d458dd8
YO
1904=item C<(*THEN)> C<(*THEN:NAME)>
1905
ac9d8485 1906This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1907C<(*PRUNE)>, this verb always matches, and when backtracked into on
1908failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
1909innermost enclosing group (capturing or otherwise) that has alternations.
1910The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1911count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
1912
1913Its name comes from the observation that this operation combined with the
1914alternation operator (C<|>) can be used to create what is essentially a
1915pattern-based if/then/else block:
1916
1917 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1918
1919Note that if this operator is used and NOT inside of an alternation then
1920it acts exactly like the C<(*PRUNE)> operator.
1921
1922 / A (*PRUNE) B /
1923
1924is the same as
1925
1926 / A (*THEN) B /
1927
1928but
1929
25e26d77 1930 / ( A (*THEN) B | C ) /
5d458dd8
YO
1931
1932is not the same as
1933
25e26d77 1934 / ( A (*PRUNE) B | C ) /
5d458dd8
YO
1935
1936as after matching the A but failing on the B the C<(*THEN)> verb will
1937backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1938
cbeadc21
JV
1939=back
1940
1941=item Verbs without an argument
1942
1943=over 4
1944
e2e6a0f1
YO
1945=item C<(*COMMIT)>
1946X<(*COMMIT)>
24b23f37 1947
241e7389 1948This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1949zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1950into on failure it causes the match to fail outright. No further attempts
1951to find a valid match by advancing the start pointer will occur again.
1952For example,
24b23f37 1953
d1fbf752
KW
1954 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
1955 print "Count=$count\n";
24b23f37
YO
1956
1957outputs
1958
1959 aaab
1960 Count=1
1961
e2e6a0f1
YO
1962In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1963does not match, the regex engine will not try any further matching on the
1964rest of the string.
c277df42 1965
e2e6a0f1
YO
1966=item C<(*FAIL)> C<(*F)>
1967X<(*FAIL)> X<(*F)>
9af228c6 1968
e2e6a0f1
YO
1969This pattern matches nothing and always fails. It can be used to force the
1970engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1971fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1972
e2e6a0f1 1973It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1974
e2e6a0f1
YO
1975=item C<(*ACCEPT)>
1976X<(*ACCEPT)>
9af228c6 1977
e2e6a0f1
YO
1978B<WARNING:> This feature is highly experimental. It is not recommended
1979for production code.
9af228c6 1980
e2e6a0f1
YO
1981This pattern matches nothing and causes the end of successful matching at
1982the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1983whether there is actually more to match in the string. When inside of a
0d017f4d 1984nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1985via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1986
c27a5cfe 1987If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1988marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1989For instance:
9af228c6 1990
e2e6a0f1 1991 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1992
e2e6a0f1 1993will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 1994be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 1995string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1996
1997=back
c277df42 1998
a0d0e21e
LW
1999=back
2000
c07a80fd 2001=head2 Backtracking
d74e8afc 2002X<backtrack> X<backtracking>
c07a80fd 2003
35a734be
IZ
2004NOTE: This section presents an abstract approximation of regular
2005expression behavior. For a more rigorous (and complicated) view of
2006the rules involved in selecting a match among possible alternatives,
0d017f4d 2007see L<Combining RE Pieces>.
35a734be 2008
c277df42 2009A fundamental feature of regular expression matching involves the
5a964f20 2010notion called I<backtracking>, which is currently used (when needed)
0d017f4d 2011by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
2012C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
2013internally, but the general principle outlined here is valid.
c07a80fd
PP
2014
2015For a regular expression to match, the I<entire> regular expression must
2016match, not just part of it. So if the beginning of a pattern containing a
2017quantifier succeeds in a way that causes later parts in the pattern to
2018fail, the matching engine backs up and recalculates the beginning
2019part--that's why it's called backtracking.
2020
2021Here is an example of backtracking: Let's say you want to find the
2022word following "foo" in the string "Food is on the foo table.":
2023
2024 $_ = "Food is on the foo table.";
2025 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 2026 print "$2 follows $1.\n";
c07a80fd
PP
2027 }
2028
2029When the match runs, the first part of the regular expression (C<\b(foo)>)
2030finds a possible match right at the beginning of the string, and loads up
2031$1 with "Foo". However, as soon as the matching engine sees that there's
2032no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 2033mistake and starts over again one character after where it had the
c07a80fd
PP
2034tentative match. This time it goes all the way until the next occurrence
2035of "foo". The complete regular expression matches this time, and you get
2036the expected output of "table follows foo."
2037
2038Sometimes minimal matching can help a lot. Imagine you'd like to match
2039everything between "foo" and "bar". Initially, you write something
2040like this:
2041
2042 $_ = "The food is under the bar in the barn.";
2043 if ( /foo(.*)bar/ ) {
f793d64a 2044 print "got <$1>\n";
c07a80fd
PP
2045 }
2046
2047Which perhaps unexpectedly yields:
2048
2049 got <d is under the bar in the >
2050
2051That's because C<.*> was greedy, so you get everything between the
14218588 2052I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
2053to use minimal matching to make sure you get the text between a "foo"
2054and the first "bar" thereafter.
2055
2056 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2057 got <d is under the >
2058
0d017f4d 2059Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2060of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2061So you write this:
2062
2063 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2064 if ( /(.*)(\d*)/ ) { # Wrong!
2065 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2066 }
2067
2068That won't work at all, because C<.*> was greedy and gobbled up the
2069whole string. As C<\d*> can match on an empty string the complete
2070regular expression matched successfully.
2071
8e1088bc 2072 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2073
2074Here are some variants, most of which don't work:
2075
2076 $_ = "I have 2 numbers: 53147";
2077 @pats = qw{
f793d64a
KW
2078 (.*)(\d*)
2079 (.*)(\d+)
2080 (.*?)(\d*)
2081 (.*?)(\d+)
2082 (.*)(\d+)$
2083 (.*?)(\d+)$
2084 (.*)\b(\d+)$
2085 (.*\D)(\d+)$
c07a80fd
PP
2086 };
2087
2088 for $pat (@pats) {
f793d64a
KW
2089 printf "%-12s ", $pat;
2090 if ( /$pat/ ) {
2091 print "<$1> <$2>\n";
2092 } else {
2093 print "FAIL\n";
2094 }
c07a80fd
PP
2095 }
2096
2097That will print out:
2098
2099 (.*)(\d*) <I have 2 numbers: 53147> <>
2100 (.*)(\d+) <I have 2 numbers: 5314> <7>
2101 (.*?)(\d*) <> <>
2102 (.*?)(\d+) <I have > <2>
2103 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2104 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2105 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2106 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2107
2108As you see, this can be a bit tricky. It's important to realize that a
2109regular expression is merely a set of assertions that gives a definition
2110of success. There may be 0, 1, or several different ways that the
2111definition might succeed against a particular string. And if there are
5a964f20
TC
2112multiple ways it might succeed, you need to understand backtracking to
2113know which variety of success you will achieve.
c07a80fd 2114
19799a22 2115When using look-ahead assertions and negations, this can all get even
8b19b778 2116trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2117followed by "123". You might try to write that as
2118
871b0233 2119 $_ = "ABC123";
f793d64a
KW
2120 if ( /^\D*(?!123)/ ) { # Wrong!
2121 print "Yup, no 123 in $_\n";
871b0233 2122 }
c07a80fd
PP
2123
2124But that isn't going to match; at least, not the way you're hoping. It
2125claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2126why that pattern matches, contrary to popular expectations:
c07a80fd 2127
4358a253
SS
2128 $x = 'ABC123';
2129 $y = 'ABC445';
c07a80fd 2130
4358a253
SS
2131 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2132 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2133
4358a253
SS
2134 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2135 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2136
2137This prints
2138
2139 2: got ABC
2140 3: got AB
2141 4: got ABC
2142
5f05dabc 2143You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2144general purpose version of test 1. The important difference between
2145them is that test 3 contains a quantifier (C<\D*>) and so can use
2146backtracking, whereas test 1 will not. What's happening is
2147that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2148non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2149let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2150fail.
14218588 2151
c07a80fd 2152The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2153try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2154a quantifier (C<\D*>) has been used in the regular expression, the
2155search engine can backtrack and retry the match differently
54310121 2156in the hope of matching the complete regular expression.
c07a80fd 2157
5a964f20
TC
2158The pattern really, I<really> wants to succeed, so it uses the
2159standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2160time. Now there's indeed something following "AB" that is not
14218588 2161"123". It's "C123", which suffices.
c07a80fd 2162
14218588
GS
2163We can deal with this by using both an assertion and a negation.
2164We'll say that the first part in $1 must be followed both by a digit
2165and by something that's not "123". Remember that the look-aheads
2166are zero-width expressions--they only look, but don't consume any
2167of the string in their match. So rewriting this way produces what
c07a80fd
PP
2168you'd expect; that is, case 5 will fail, but case 6 succeeds:
2169
4358a253
SS
2170 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2171 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2172
2173 6: got ABC
2174
5a964f20 2175In other words, the two zero-width assertions next to each other work as though
19799a22 2176they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2177matches only if you're at the beginning of the line AND the end of the
2178line simultaneously. The deeper underlying truth is that juxtaposition in
2179regular expressions always means AND, except when you write an explicit OR
2180using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2181although the attempted matches are made at different positions because "a"
2182is not a zero-width assertion, but a one-width assertion.
2183
0d017f4d 2184B<WARNING>: Particularly complicated regular expressions can take
14218588 2185exponential time to solve because of the immense number of possible
0d017f4d 2186ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2187internal optimizations done by the regular expression engine, this will
2188take a painfully long time to run:
c07a80fd 2189
e1901655
IZ
2190 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2191
2192And if you used C<*>'s in the internal groups instead of limiting them
2193to 0 through 5 matches, then it would take forever--or until you ran
2194out of stack space. Moreover, these internal optimizations are not
2195always applicable. For example, if you put C<{0,5}> instead of C<*>
2196on the external group, no current optimization is applicable, and the
2197match takes a long time to finish.
c07a80fd 2198
9da458fc
IZ
2199A powerful tool for optimizing such beasts is what is known as an
2200"independent group",
96090e4f 2201which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2202zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2203the tail match, since they are in "logical" context: only
14218588 2204whether they match is considered relevant. For an example
9da458fc 2205where side-effects of look-ahead I<might> have influenced the
96090e4f 2206following match, see L</C<< (?>pattern) >>>.
c277df42 2207
a0d0e21e 2208=head2 Version 8 Regular Expressions
d74e8afc 2209X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2210
5a964f20 2211In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2212routines, here are the pattern-matching rules not described above.
2213
54310121 2214Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2215with a special meaning described here or above. You can cause
5a964f20 2216characters that normally function as metacharacters to be interpreted
5f05dabc 2217literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2218character; "\\" matches a "\"). This escape mechanism is also required
2219for the character used as the pattern delimiter.
2220
2221A series of characters matches that series of characters in the target
0b928c2f 2222string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2223string.
a0d0e21e
LW
2224
2225You can specify a character class, by enclosing a list of characters
5d458dd8 2226in C<[]>, which will match any character from the list. If the
a0d0e21e 2227first character after the "[" is "^", the class matches any character not
14218588 2228in the list. Within a list, the "-" character specifies a
5a964f20 2229range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2230inclusive. If you want either "-" or "]" itself to be a member of a
2231class, put it at the start of the list (possibly after a "^"), or
2232escape it with a backslash. "-" is also taken literally when it is
2233at the end of the list, just before the closing "]". (The
84850974
DD
2234following all specify the same class of three characters: C<[-az]>,
2235C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2236specifies a class containing twenty-six characters, even on EBCDIC-based
2237character sets.) Also, if you try to use the character
2238classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2239a range, the "-" is understood literally.
a0d0e21e 2240
8ada0baa
JH
2241Note also that the whole range idea is rather unportable between
2242character sets--and even within character sets they may cause results
2243you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2244that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2245[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2246spell out the character sets in full.
2247
54310121 2248Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2249used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2250"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2251of three octal digits, matches the character whose coded character set value
5d458dd8 2252is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2253matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2254matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2255matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2256
2257You can specify a series of alternatives for a pattern using "|" to
2258separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2259or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2260first alternative includes everything from the last pattern delimiter
0b928c2f 2261("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2262the last alternative contains everything from the last "|" to the next
0b928c2f 2263closing pattern delimiter. That's why it's common practice to include
14218588 2264alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2265start and end.
2266
5a964f20 2267Alternatives are tried from left to right, so the first
a3cb178b
GS
2268alternative found for which the entire expression matches, is the one that
2269is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2270example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2271part will match, as that is the first alternative tried, and it successfully
2272matches the target string. (This might not seem important, but it is
2273important when you are capturing matched text using parentheses.)
2274
5a964f20 2275Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2276so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2277
14218588
GS
2278Within a pattern, you may designate subpatterns for later reference
2279by enclosing them in parentheses, and you may refer back to the
2280I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2281\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2282of their opening parenthesis. A backreference matches whatever
2283actually matched the subpattern in the string being examined, not
d8b950dc 2284the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2285match "0x1234 0x4321", but not "0x1234 01234", because subpattern
22861 matched "0x", even though the rule C<0|0x> could potentially match
2287the leading 0 in the second number.
cb1a09d0 2288
0d017f4d 2289=head2 Warning on \1 Instead of $1
cb1a09d0 2290
5a964f20 2291Some people get too used to writing things like:
cb1a09d0
AD
2292
2293 $pattern =~ s/(\W)/\\\1/g;
2294
3ff1c45a
KW
2295This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2296shocking the
cb1a09d0 2297B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2298PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2299the usual double-quoted string means a control-A. The customary Unix
2300meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2301of doing that, you get yourself into trouble if you then add an C</e>
2302modifier.
2303
f793d64a 2304 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2305
2306Or if you try to do
2307
2308 s/(\d+)/\1000/;
2309
2310You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2311C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2312with the operation of matching a backreference. Certainly they mean two
2313different things on the I<left> side of the C<s///>.
9fa51da4 2314
0d017f4d 2315=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2316
19799a22 2317B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2318
2319Regular expressions provide a terse and powerful programming language. As
2320with most other power tools, power comes together with the ability
2321to wreak havoc.
2322
2323A common abuse of this power stems from the ability to make infinite
628afcb5 2324loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2325
2326 'foo' =~ m{ ( o? )* }x;
2327
0d017f4d 2328The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2329in the string is not moved by the match, C<o?> would match again and again
527e91da 2330because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2331is with the looping modifier C<//g>:
2332
2333 @matches = ( 'foo' =~ m{ o? }xg );
2334
2335or
2336
2337 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2338
2339or the loop implied by split().
2340
2341However, long experience has shown that many programming tasks may
14218588
GS
2342be significantly simplified by using repeated subexpressions that
2343may match zero-length substrings. Here's a simple example being:
c84d73f1 2344
d1fbf752 2345 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2346 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2347
9da458fc 2348Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2349the infinite loop>. The rules for this are different for lower-level
527e91da 2350loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2351ones like the C</g> modifier or split() operator.
2352
19799a22
GS
2353The lower-level loops are I<interrupted> (that is, the loop is
2354broken) when Perl detects that a repeated expression matched a
2355zero-length substring. Thus
c84d73f1
IZ
2356
2357 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2358
5d458dd8 2359is made equivalent to
c84d73f1 2360
0b928c2f
FC
2361 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2362
2363For example, this program
2364
2365 #!perl -l
2366 "aaaaab" =~ /
2367 (?:
2368 a # non-zero
2369 | # or
2370 (?{print "hello"}) # print hello whenever this
2371 # branch is tried
2372 (?=(b)) # zero-width assertion
2373 )* # any number of times
2374 /x;
2375 print $&;
2376 print $1;
c84d73f1 2377
0b928c2f
FC
2378prints
2379
2380 hello
2381 aaaaa
2382 b
2383
2384Notice that "hello" is only printed once, as when Perl sees that the sixth
2385iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2386the C<*>.
2387
2388The higher-level loops preserve an additional state between iterations:
5d458dd8 2389whether the last match was zero-length. To break the loop, the following
c84d73f1 2390match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2391This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2392and so the I<second best> match is chosen if the I<best> match is of
2393zero length.
2394
19799a22 2395For example:
c84d73f1
IZ
2396
2397 $_ = 'bar';
2398 s/\w??/<$&>/g;
2399
20fb949f 2400results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2401match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2402best> match is what is matched by C<\w>. Thus zero-length matches
2403alternate with one-character-long matches.
2404
5d458dd8 2405Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2406position one notch further in the string.
2407
19799a22 2408The additional state of being I<matched with zero-length> is associated with
c84d73f1 2409the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2410Zero-length matches at the end of the previous match are ignored
2411during C<split>.
c84d73f1 2412
0d017f4d 2413=head2 Combining RE Pieces
35a734be
IZ
2414
2415Each of the elementary pieces of regular expressions which were described
2416before (such as C<ab> or C<\Z>) could match at most one substring
2417at the given position of the input string. However, in a typical regular
2418expression these elementary pieces are combined into more complicated
0b928c2f 2419patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2420(in these examples C<S> and C<T> are regular subexpressions).
2421
2422Such combinations can include alternatives, leading to a problem of choice:
2423if we match a regular expression C<a|ab> against C<"abc">, will it match
2424substring C<"a"> or C<"ab">? One way to describe which substring is
2425actually matched is the concept of backtracking (see L<"Backtracking">).
2426However, this description is too low-level and makes you think
2427in terms of a particular implementation.
2428
2429Another description starts with notions of "better"/"worse". All the
2430substrings which may be matched by the given regular expression can be
2431sorted from the "best" match to the "worst" match, and it is the "best"
2432match which is chosen. This substitutes the question of "what is chosen?"
2433by the question of "which matches are better, and which are worse?".
2434
2435Again, for elementary pieces there is no such question, since at most
2436one match at a given position is possible. This section describes the
2437notion of better/worse for combining operators. In the description
2438below C<S> and C<T> are regular subexpressions.
2439
13a2d996 2440=over 4
35a734be
IZ
2441
2442=item C<ST>
2443
2444Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2445substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2446which can be matched by C<T>.
35a734be 2447
0b928c2f 2448If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2449match than C<A'B'>.
2450
2451If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2452C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2453
2454=item C<S|T>
2455
2456When C<S> can match, it is a better match than when only C<T> can match.
2457
2458Ordering of two matches for C<S> is the same as for C<S>. Similar for
2459two matches for C<T>.
2460
2461=item C<S{REPEAT_COUNT}>
2462
2463Matches as C<SSS...S> (repeated as many times as necessary).
2464
2465=item C<S{min,max}>
2466
2467Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2468
2469=item C<S{min,max}?>
2470
2471Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2472
2473=item C<S?>, C<S*>, C<S+>
2474
2475Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2476
2477=item C<S??>, C<S*?>, C<S+?>
2478
2479Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2480
c47ff5f1 2481=item C<< (?>S) >>
35a734be
IZ
2482
2483Matches the best match for C<S> and only that.
2484
2485=item C<(?=S)>, C<(?<=S)>
2486
2487Only the best match for C<S> is considered. (This is important only if
2488C<S> has capturing parentheses, and backreferences are used somewhere
2489else in the whole regular expression.)
2490
2491=item C<(?!S)>, C<(?<!S)>
2492
2493For this grouping operator there is no need to describe the ordering, since
2494only whether or not C<S> can match is important.
2495
6bda09f9 2496=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2497
2498The ordering is the same as for the regular expression which is
c27a5cfe 2499the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
2500
2501=item C<(?(condition)yes-pattern|no-pattern)>
2502
2503Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2504already determined. The ordering of the matches is the same as for the
2505chosen subexpression.
2506
2507=back
2508
2509The above recipes describe the ordering of matches I<at a given position>.
2510One more rule is needed to understand how a match is determined for the
2511whole regular expression: a match at an earlier position is always better
2512than a match at a later position.
2513
0d017f4d 2514=head2 Creating Custom RE Engines
c84d73f1 2515
0b928c2f
FC
2516As of Perl 5.10.0, one can create custom regular expression engines. This
2517is not for the faint of heart, as they have to plug in at the C level. See
2518L<perlreapi> for more details.
2519
2520As an alternative, overloaded constants (see L<overload>) provide a simple
2521way to extend the functionality of the RE engine, by substituting one
2522pattern for another.
c84d73f1
IZ
2523
2524Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2525matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2526characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2527at these positions, so we want to have each C<\Y|> in the place of the
2528more complicated version. We can create a module C<customre> to do
2529this:
2530
2531 package customre;
2532 use overload;
2533
2534 sub import {
2535 shift;
2536 die "No argument to customre::import allowed" if @_;
2537 overload::constant 'qr' => \&convert;
2538 }
2539
2540 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2541
580a9fe1
RGS
2542 # We must also take care of not escaping the legitimate \\Y|
2543 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2544 my %rules = ( '\\' => '\\\\',
f793d64a 2545 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2546 sub convert {
2547 my $re = shift;
5d458dd8 2548 $re =~ s{
c84d73f1
IZ
2549 \\ ( \\ | Y . )
2550 }
5d458dd8 2551 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2552 return $re;
2553 }
2554
2555Now C<use customre> enables the new escape in constant regular
2556expressions, i.e., those without any runtime variable interpolations.
2557As documented in L<overload>, this conversion will work only over
2558literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2559part of this regular expression needs to be converted explicitly
2560(but only if the special meaning of C<\Y|> should be enabled inside $re):
2561
2562 use customre;
2563 $re = <>;
2564 chomp $re;
2565 $re = customre::convert $re;
2566 /\Y|$re\Y|/;
2567
0b928c2f 2568=head2 PCRE/Python Support
1f1031fe 2569
0b928c2f 2570As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2571to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2572Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2573
2574=over 4
2575
ae5648b3 2576=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2577
c27a5cfe 2578Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2579
2580=item C<< (?P=NAME) >>
2581
c27a5cfe 2582Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2583
2584=item C<< (?P>NAME) >>
2585
c27a5cfe 2586Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2587
ee9b8eae 2588=back
1f1031fe 2589
19799a22
GS
2590=head1 BUGS
2591
88c9975e
KW
2592Many regular expression constructs don't work on EBCDIC platforms.
2593
ed7efc79
KW
2594There are a number of issues with regard to case-insensitive matching
2595in Unicode rules. See C<i> under L</Modifiers> above.
2596
9da458fc
IZ
2597This document varies from difficult to understand to completely
2598and utterly opaque. The wandering prose riddled with jargon is
2599hard to fathom in several places.
2600
2601This document needs a rewrite that separates the tutorial content
2602from the reference content.
19799a22
GS
2603
2604=head1 SEE ALSO
9fa51da4 2605
91e0c79e
MJD
2606L<perlrequick>.
2607
2608L<perlretut>.
2609
9b599b2a
GS
2610L<perlop/"Regexp Quote-Like Operators">.
2611
1e66bd83
PP
2612L<perlop/"Gory details of parsing quoted constructs">.
2613
14218588
GS
2614L<perlfaq6>.
2615
9b599b2a
GS
2616L<perlfunc/pos>.
2617
2618L<perllocale>.
2619
fb55449c
JH
2620L<perlebcdic>.
2621
14218588
GS
2622I<Mastering Regular Expressions> by Jeffrey Friedl, published
2623by O'Reilly and Associates.