This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
typo in perlre.pod
[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 "$" from matching
14218588 34the start or end of the string to matching the start or end of any
7f761169 35line anywhere within the string.
55497cff 36
54310121 37=item s
d74e8afc
ITB
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
55497cff
PP
40
41Treat string as single line. That is, change "." to match any character
19799a22 42whatsoever, even a newline, which normally it would not match.
55497cff 43
34d67d80 44Used together, as C</ms>, they let the "." match any character whatsoever,
fb55449c 45while still allowing "^" and "$" 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
1f59b283
KW
75Perl doesn't match multiple characters in an inverted bracketed
76character class, which otherwise could be highly confusing. See
77L<perlrecharclass/Negation>.
78
25275596
KW
79Another bug involves character classes that match both a sequence of
80multiple characters, and an initial sub-string of that sequence. For
81example,
82
83 /[s\xDF]/i
84
85should match both a single and a double "s", since C<\xDF> (on ASCII
86platforms) matches "ss". However, this bug
87(L<[perl #89774]|https://rt.perl.org/rt3/Ticket/Display.html?id=89774>)
88causes it to only match a single "s", even if the final larger match
89fails, and matching the double "ss" would have succeeded.
90
1f59b283 91Also, Perl matching doesn't fully conform to the current Unicode C</i>
ed7efc79
KW
92recommendations, which ask that the matching be made upon the NFD
93(Normalization Form Decomposed) of the text. However, Unicode is
94in the process of reconsidering and revising their recommendations.
87e95b7f 95
54310121 96=item x
d74e8afc 97X</x>
55497cff
PP
98
99Extend your pattern's legibility by permitting whitespace and comments.
ed7efc79 100Details in L</"/x">
55497cff 101
87e95b7f
YO
102=item p
103X</p> X<regex, preserve> X<regexp, preserve>
104
632a1772 105Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
87e95b7f
YO
106${^POSTMATCH} are available for use after matching.
107
e2e6bec7
DN
108=item g and c
109X</g> X</c>
110
111Global matching, and keep the Current position after failed matching.
112Unlike i, m, s and x, these two flags affect the way the regex is used
113rather than the regex itself. See
114L<perlretut/"Using regular expressions in Perl"> for further explanation
115of the g and c modifiers.
116
b6fa137b
FC
117=item a, d, l and u
118X</a> X</d> X</l> X</u>
119
516074bb
KW
120These modifiers, all new in 5.14, affect which character-set semantics
121(Unicode, etc.) are used, as described below in
ed7efc79 122L</Character set modifiers>.
b6fa137b 123
55497cff 124=back
a0d0e21e 125
516074bb
KW
126Regular expression modifiers are usually written in documentation
127as e.g., "the C</x> modifier", even though the delimiter
b6fa137b 128in question might not really be a slash. The modifiers C</imsxadlup>
ab7bb42d 129may also be embedded within the regular expression itself using
ed7efc79 130the C<(?...)> construct, see L</Extended Patterns> below.
b6fa137b 131
ed7efc79
KW
132=head3 /x
133
b6fa137b 134C</x> tells
7b059540 135the regular expression parser to ignore most whitespace that is neither
55497cff 136backslashed nor within a character class. You can use this to break up
4633a7c4 137your regular expression into (slightly) more readable parts. The C<#>
54310121 138character is also treated as a metacharacter introducing a comment,
55497cff 139just as in ordinary Perl code. This also means that if you want real
14218588 140whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a 141class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
142escape them (using backslashes or C<\Q...\E>) or encode them using octal,
143hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
144making Perl's regular expressions more readable. Note that you have to
145be careful not to include the pattern delimiter in the comment--perl has
146no way of knowing you did not intend to close the pattern early. See
147the C-comment deletion code in L<perlop>. Also note that anything inside
7651b971 148a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
0b928c2f 149space interpretation within a single multi-character construct. For
7651b971 150example in C<\x{...}>, regardless of the C</x> modifier, there can be no
9bb1f947 151spaces. Same for a L<quantifier|/Quantifiers> such as C<{3}> or
f9e949fd
KW
152C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<?> and C<:>,
153but can between the C<(> and C<?>. Within any delimiters for such a
154construct, allowed spaces are not affected by C</x>, and depend on the
155construct. For example, C<\x{...}> can't have spaces because hexadecimal
156numbers don't have spaces in them. But, Unicode properties can have spaces, so
0b928c2f 157in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
9bb1f947 158L<perluniprops/Properties accessible through \p{} and \P{}>.
d74e8afc 159X</x>
a0d0e21e 160
ed7efc79
KW
161=head3 Character set modifiers
162
163C</d>, C</u>, C</a>, and C</l>, available starting in 5.14, are called
164the character set modifiers; they affect the character set semantics
165used for the regular expression.
166
808432af
KW
167The C</d>, C</u>, and C</l> modifiers are not likely to be of much use
168to you, and so you need not worry about them very much. They exist for
169Perl's internal use, so that complex regular expression data structures
170can be automatically serialized and later exactly reconstituted,
171including all their nuances. But, since Perl can't keep a secret, and
172there may be rare instances where they are useful, they are documented
173here.
ed7efc79 174
808432af
KW
175The C</a> modifier, on the other hand, may be useful. Its purpose is to
176allow code that is to work mostly on ASCII data to not have to concern
177itself with Unicode.
ca9560b2 178
808432af
KW
179Briefly, C</l> sets the character set to that of whatever B<L>ocale is in
180effect at the time of the execution of the pattern match.
ca9560b2 181
808432af
KW
182C</u> sets the character set to B<U>nicode.
183
184C</a> also sets the character set to Unicode, BUT adds several
185restrictions for B<A>SCII-safe matching.
186
187C</d> is the old, problematic, pre-5.14 B<D>efault character set
188behavior. Its only use is to force that old behavior.
189
190At any given time, exactly one of these modifiers is in effect. Their
191existence allows Perl to keep the originally compiled behavior of a
192regular expression, regardless of what rules are in effect when it is
193actually executed. And if it is interpolated into a larger regex, the
194original's rules continue to apply to it, and only it.
195
196The C</l> and C</u> modifiers are automatically selected for
197regular expressions compiled within the scope of various pragmas,
198and we recommend that in general, you use those pragmas instead of
199specifying these modifiers explicitly. For one thing, the modifiers
200affect only pattern matching, and do not extend to even any replacement
201done, whereas using the pragmas give consistent results for all
202appropriate operations within their scopes. For example,
203
204 s/foo/\Ubar/il
205
206will match "foo" using the locale's rules for case-insensitive matching,
207but the C</l> does not affect how the C<\U> operates. Most likely you
208want both of them to use locale rules. To do this, instead compile the
209regular expression within the scope of C<use locale>. This both
210implicitly adds the C</l> and applies locale rules to the C<\U>. The
211lesson is to C<use locale> and not C</l> explicitly.
212
213Similarly, it would be better to use C<use feature 'unicode_strings'>
214instead of,
215
216 s/foo/\Lbar/iu
217
218to get Unicode rules, as the C<\L> in the former (but not necessarily
219the latter) would also use Unicode rules.
220
221More detail on each of the modifiers follows. Most likely you don't
222need to know this detail for C</l>, C</u>, and C</d>, and can skip ahead
223to L<E<sol>a|/E<sol>a (and E<sol>aa)>.
ca9560b2 224
ed7efc79
KW
225=head4 /l
226
227means to use the current locale's rules (see L<perllocale>) when pattern
228matching. For example, C<\w> will match the "word" characters of that
229locale, and C<"/i"> case-insensitive matching will match according to
230the locale's case folding rules. The locale used will be the one in
231effect at the time of execution of the pattern match. This may not be
232the same as the compilation-time locale, and can differ from one match
233to another if there is an intervening call of the
b6fa137b 234L<setlocale() function|perllocale/The setlocale function>.
ed7efc79
KW
235
236Perl only supports single-byte locales. This means that code points
237above 255 are treated as Unicode no matter what locale is in effect.
238Under Unicode rules, there are a few case-insensitive matches that cross
239the 255/256 boundary. These are disallowed under C</l>. For example,
516074bb
KW
2400xFF (on ASCII platforms) does not caselessly match the character at
2410x178, C<LATIN CAPITAL LETTER Y WITH DIAERESIS>, because 0xFF may not be
242C<LATIN SMALL LETTER Y WITH DIAERESIS> in the current locale, and Perl
243has no way of knowing if that character even exists in the locale, much
244less what code point it is.
ed7efc79
KW
245
246This modifier may be specified to be the default by C<use locale>, but
247see L</Which character set modifier is in effect?>.
b6fa137b
FC
248X</l>
249
ed7efc79
KW
250=head4 /u
251
252means to use Unicode rules when pattern matching. On ASCII platforms,
253this means that the code points between 128 and 255 take on their
808432af
KW
254Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's).
255(Otherwise Perl considers their meanings to be undefined.) Thus,
256under this modifier, the ASCII platform effectively becomes a Unicode
257platform; and hence, for example, C<\w> will match any of the more than
258100_000 word characters in Unicode.
ed7efc79
KW
259
260Unlike most locales, which are specific to a language and country pair,
516074bb
KW
261Unicode classifies all the characters that are letters I<somewhere> in
262the world as
ed7efc79
KW
263C<\w>. For example, your locale might not think that C<LATIN SMALL
264LETTER ETH> is a letter (unless you happen to speak Icelandic), but
265Unicode does. Similarly, all the characters that are decimal digits
266somewhere in the world will match C<\d>; this is hundreds, not 10,
267possible matches. And some of those digits look like some of the 10
268ASCII digits, but mean a different number, so a human could easily think
269a number is a different quantity than it really is. For example,
270C<BENGALI DIGIT FOUR> (U+09EA) looks very much like an
271C<ASCII DIGIT EIGHT> (U+0038). And, C<\d+>, may match strings of digits
272that are a mixture from different writing systems, creating a security
67592e11 273issue. L<Unicode::UCD/num()> can be used to sort
516074bb
KW
274this out. Or the C</a> modifier can be used to force C<\d> to match
275just the ASCII 0 through 9.
ed7efc79 276
516074bb
KW
277Also, under this modifier, case-insensitive matching works on the full
278set of Unicode
ed7efc79
KW
279characters. The C<KELVIN SIGN>, for example matches the letters "k" and
280"K"; and C<LATIN SMALL LIGATURE FF> matches the sequence "ff", which,
281if you're not prepared, might make it look like a hexadecimal constant,
282presenting another potential security issue. See
283L<http://unicode.org/reports/tr36> for a detailed discussion of Unicode
284security issues.
285
6368643f
KW
286On the EBCDIC platforms that Perl handles, the native character set is
287equivalent to Latin-1. Thus this modifier changes behavior only when
ed7efc79
KW
288the C<"/i"> modifier is also specified, and it turns out it affects only
289two characters, giving them full Unicode semantics: the C<MICRO SIGN>
6368643f 290will match the Greek capital and small letters C<MU>, otherwise not; and
ed7efc79
KW
291the C<LATIN CAPITAL LETTER SHARP S> will match any of C<SS>, C<Ss>,
292C<sS>, and C<ss>, otherwise not.
293
294This modifier may be specified to be the default by C<use feature
808432af
KW
295'unicode_strings> or C<L<use 5.012|perlfunc/use VERSION>> (or higher),
296but see L</Which character set modifier is in effect?>.
b6fa137b
FC
297X</u>
298
ed7efc79
KW
299=head4 /d
300
301This modifier means to use the "Default" native rules of the platform
302except when there is cause to use Unicode rules instead, as follows:
303
304=over 4
305
306=item 1
307
308the target string is encoded in UTF-8; or
309
310=item 2
311
312the pattern is encoded in UTF-8; or
313
314=item 3
315
316the pattern explicitly mentions a code point that is above 255 (say by
317C<\x{100}>); or
318
319=item 4
b6fa137b 320
ed7efc79
KW
321the pattern uses a Unicode name (C<\N{...}>); or
322
323=item 5
324
325the pattern uses a Unicode property (C<\p{...}>)
326
327=back
328
329Another mnemonic for this modifier is "Depends", as the rules actually
330used depend on various things, and as a result you can get unexpected
808432af
KW
331results. See L<perlunicode/The "Unicode Bug">. The Unicode Bug has
332become rather infamous, leading to yet another (printable) name for this
333modifier, "Dodgy".
ed7efc79
KW
334
335On ASCII platforms, the native rules are ASCII, and on EBCDIC platforms
336(at least the ones that Perl handles), they are Latin-1.
337
338Here are some examples of how that works on an ASCII platform:
339
340 $str = "\xDF"; # $str is not in UTF-8 format.
341 $str =~ /^\w/; # No match, as $str isn't in UTF-8 format.
342 $str .= "\x{0e0b}"; # Now $str is in UTF-8 format.
343 $str =~ /^\w/; # Match! $str is now in UTF-8 format.
344 chop $str;
345 $str =~ /^\w/; # Still a match! $str remains in UTF-8 format.
346
808432af
KW
347This modifier is automatically selected by default when none of the
348others are, so yet another name for it is "Default".
349
350Because of the unexpected behaviors associated with this modifier, you
351probably should only use it to maintain weird backward compatibilities.
352
353=head4 /a (and /aa)
354
355This modifier stands for ASCII-restrict (or ASCII-safe). This modifier,
356unlike the others, may be doubled-up to increase its effect.
357
358When it appears singly, it causes the sequences C<\d>, C<\s>, C<\w>, and
359the Posix character classes to match only in the ASCII range. They thus
360revert to their pre-5.6, pre-Unicode meanings. Under C</a>, C<\d>
361always means precisely the digits C<"0"> to C<"9">; C<\s> means the five
362characters C<[ \f\n\r\t]>; C<\w> means the 63 characters
363C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
364C<[[:print:]]> match only the appropriate ASCII-range characters.
365
366This modifier is useful for people who only incidentally use Unicode,
367and who do not wish to be burdened with its complexities and security
368concerns.
369
370With C</a>, one can write C<\d> with confidence that it will only match
371ASCII characters, and should the need arise to match beyond ASCII, you
372can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>). There are
373similar C<\p{...}> constructs that can match beyond ASCII both white
374space (see L<perlrecharclass/Whitespace>), and Posix classes (see
375L<perlrecharclass/POSIX Character Classes>). Thus, this modifier
376doesn't mean you can't use Unicode, it means that to get Unicode
377matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
378signals Unicode.
379
380As you would expect, this modifier causes, for example, C<\D> to mean
381the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
382C<\D>, C<\S>, and C<\W>. C<\b> still means to match at the boundary
383between C<\w> and C<\W>, using the C</a> definitions of them (similarly
384for C<\B>).
385
386Otherwise, C</a> behaves like the C</u> modifier, in that
387case-insensitive matching uses Unicode semantics; for example, "k" will
388match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
389points in the Latin1 range, above ASCII will have Unicode rules when it
390comes to case-insensitive matching.
391
392To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
393specify the "a" twice, for example C</aai> or C</aia>. (The first
394occurrence of "a" restricts the C<\d>, etc., and the second occurrence
395adds the C</i> restrictions.) But, note that code points outside the
396ASCII range will use Unicode rules for C</i> matching, so the modifier
397doesn't really restrict things to just ASCII; it just forbids the
398intermixing of ASCII and non-ASCII.
399
400To summarize, this modifier provides protection for applications that
401don't wish to be exposed to all of Unicode. Specifying it twice
402gives added protection.
403
404This modifier may be specified to be the default by C<use re '/a'>
405or C<use re '/aa'>. If you do so, you may actually have occasion to use
406the C</u> modifier explictly if there are a few regular expressions
407where you do want full Unicode rules (but even here, it's best if
408everything were under feature C<"unicode_strings">, along with the
409C<use re '/aa'>). Also see L</Which character set modifier is in
410effect?>.
411X</a>
412X</aa>
413
ed7efc79
KW
414=head4 Which character set modifier is in effect?
415
416Which of these modifiers is in effect at any given point in a regular
808432af
KW
417expression depends on a fairly complex set of interactions. These have
418been designed so that in general you don't have to worry about it, but
419this section gives the gory details. As
ed7efc79
KW
420explained below in L</Extended Patterns> it is possible to explicitly
421specify modifiers that apply only to portions of a regular expression.
422The innermost always has priority over any outer ones, and one applying
6368643f
KW
423to the whole expression has priority over any of the default settings that are
424described in the remainder of this section.
ed7efc79 425
916cec3f 426The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
ed7efc79
KW
427default modifiers (including these) for regular expressions compiled
428within its scope. This pragma has precedence over the other pragmas
516074bb 429listed below that also change the defaults.
ed7efc79
KW
430
431Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
432and C<L<use feature 'unicode_strings|feature>> or
433C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
434C</u> when not in the same scope as either C<L<use locale|perllocale>>
6368643f
KW
435or C<L<use bytes|bytes>>. Unlike the mechanisms mentioned above, these
436affect operations besides regular expressions pattern matching, and so
437give more consistent results with other operators, including using
438C<\U>, C<\l>, etc. in substitution replacements.
ed7efc79
KW
439
440If none of the above apply, for backwards compatibility reasons, the
441C</d> modifier is the one in effect by default. As this can lead to
442unexpected results, it is best to specify which other rule set should be
443used.
444
445=head4 Character set modifier behavior prior to Perl 5.14
446
447Prior to 5.14, there were no explicit modifiers, but C</l> was implied
448for regexes compiled within the scope of C<use locale>, and C</d> was
449implied otherwise. However, interpolating a regex into a larger regex
450would ignore the original compilation in favor of whatever was in effect
451at the time of the second compilation. There were a number of
452inconsistencies (bugs) with the C</d> modifier, where Unicode rules
453would be used when inappropriate, and vice versa. C<\p{}> did not imply
454Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
b6fa137b 455
a0d0e21e
LW
456=head2 Regular Expressions
457
04838cea
RGS
458=head3 Metacharacters
459
384f06ae 460The patterns used in Perl pattern matching evolved from those supplied in
14218588 461the Version 8 regex routines. (The routines are derived
19799a22
GS
462(distantly) from Henry Spencer's freely redistributable reimplementation
463of the V8 routines.) See L<Version 8 Regular Expressions> for
464details.
a0d0e21e
LW
465
466In particular the following metacharacters have their standard I<egrep>-ish
467meanings:
d74e8afc
ITB
468X<metacharacter>
469X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
470
a0d0e21e 471
f793d64a
KW
472 \ Quote the next metacharacter
473 ^ Match the beginning of the line
474 . Match any character (except newline)
475 $ Match the end of the line (or before newline at the end)
476 | Alternation
477 () Grouping
478 [] Bracketed Character class
a0d0e21e 479
14218588
GS
480By default, the "^" character is guaranteed to match only the
481beginning of the string, the "$" character only the end (or before the
482newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
483assumption that the string contains only one line. Embedded newlines
484will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 485string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
486newline within the string (except if the newline is the last character in
487the string), and "$" will match before any newline. At the
a0d0e21e
LW
488cost of a little more overhead, you can do this by using the /m modifier
489on the pattern match operator. (Older programs did this by setting C<$*>,
0b928c2f 490but this option was removed in perl 5.9.)
d74e8afc 491X<^> X<$> X</m>
a0d0e21e 492
14218588 493To simplify multi-line substitutions, the "." character never matches a
55497cff 494newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 495the string is a single line--even if it isn't.
d74e8afc 496X<.> X</s>
a0d0e21e 497
04838cea
RGS
498=head3 Quantifiers
499
a0d0e21e 500The following standard quantifiers are recognized:
d74e8afc 501X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 502
f793d64a
KW
503 * Match 0 or more times
504 + Match 1 or more times
505 ? Match 1 or 0 times
506 {n} Match exactly n times
507 {n,} Match at least n times
508 {n,m} Match at least n but not more than m times
a0d0e21e 509
0b928c2f
FC
510(If a curly bracket occurs in any other context and does not form part of
511a backslashed sequence like C<\x{...}>, it is treated
b975c076 512as a regular character. In particular, the lower bound
527e91da
BB
513is not optional.) The "*" quantifier is equivalent to C<{0,}>, the "+"
514quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 515to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
516This is usually 32766 on the most common platforms. The actual limit can
517be seen in the error message generated by code such as this:
518
820475bd 519 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 520
54310121
PP
521By default, a quantified subpattern is "greedy", that is, it will match as
522many times as possible (given a particular starting location) while still
523allowing the rest of the pattern to match. If you want it to match the
524minimum number of times possible, follow the quantifier with a "?". Note
525that the meanings don't change, just the "greediness":
0d017f4d 526X<metacharacter> X<greedy> X<greediness>
d74e8afc 527X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 528
f793d64a
KW
529 *? Match 0 or more times, not greedily
530 +? Match 1 or more times, not greedily
531 ?? Match 0 or 1 time, not greedily
0b928c2f 532 {n}? Match exactly n times, not greedily (redundant)
f793d64a
KW
533 {n,}? Match at least n times, not greedily
534 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 535
b9b4dddf
YO
536By default, when a quantified subpattern does not allow the rest of the
537overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 538sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
539as well.
540
f793d64a
KW
541 *+ Match 0 or more times and give nothing back
542 ++ Match 1 or more times and give nothing back
543 ?+ Match 0 or 1 time and give nothing back
544 {n}+ Match exactly n times and give nothing back (redundant)
545 {n,}+ Match at least n times and give nothing back
546 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
547
548For instance,
549
550 'aaaa' =~ /a++a/
551
552will never match, as the C<a++> will gobble up all the C<a>'s in the
553string and won't leave any for the remaining part of the pattern. This
554feature can be extremely useful to give perl hints about where it
555shouldn't backtrack. For instance, the typical "match a double-quoted
556string" problem can be most efficiently performed when written as:
557
558 /"(?:[^"\\]++|\\.)*+"/
559
0d017f4d 560as we know that if the final quote does not match, backtracking will not
0b928c2f
FC
561help. See the independent subexpression
562L</C<< (?>pattern) >>> for more details;
b9b4dddf
YO
563possessive quantifiers are just syntactic sugar for that construct. For
564instance the above example could also be written as follows:
565
566 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
567
04838cea
RGS
568=head3 Escape sequences
569
0b928c2f 570Because patterns are processed as double-quoted strings, the following
a0d0e21e
LW
571also work:
572
f793d64a
KW
573 \t tab (HT, TAB)
574 \n newline (LF, NL)
575 \r return (CR)
576 \f form feed (FF)
577 \a alarm (bell) (BEL)
578 \e escape (think troff) (ESC)
f793d64a 579 \cK control char (example: VT)
dc0d9c48 580 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 581 \N{name} named Unicode character or character sequence
f793d64a 582 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 583 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
584 \l lowercase next char (think vi)
585 \u uppercase next char (think vi)
586 \L lowercase till \E (think vi)
587 \U uppercase till \E (think vi)
588 \Q quote (disable) pattern metacharacters till \E
589 \E end either case modification or quoted section, think vi
a0d0e21e 590
9bb1f947 591Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 592
e1d1eefb 593=head3 Character Classes and other Special Escapes
04838cea 594
a0d0e21e 595In addition, Perl defines the following:
d0b16107 596X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 597
f793d64a
KW
598 Sequence Note Description
599 [...] [1] Match a character according to the rules of the
600 bracketed character class defined by the "...".
601 Example: [a-z] matches "a" or "b" or "c" ... or "z"
602 [[:...:]] [2] Match a character according to the rules of the POSIX
603 character class "..." within the outer bracketed
604 character class. Example: [[:upper:]] matches any
605 uppercase character.
d35dd6c6
KW
606 \w [3] Match a "word" character (alphanumeric plus "_", plus
607 other connector punctuation chars plus Unicode
0b928c2f 608 marks)
f793d64a
KW
609 \W [3] Match a non-"word" character
610 \s [3] Match a whitespace character
611 \S [3] Match a non-whitespace character
612 \d [3] Match a decimal digit character
613 \D [3] Match a non-digit character
614 \pP [3] Match P, named property. Use \p{Prop} for longer names
615 \PP [3] Match non-P
616 \X [4] Match Unicode "eXtended grapheme cluster"
617 \C Match a single C-language char (octet) even if that is
618 part of a larger UTF-8 character. Thus it breaks up
619 characters into their UTF-8 bytes, so you may end up
620 with malformed pieces of UTF-8. Unsupported in
621 lookbehind.
c27a5cfe 622 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
623 '1' may actually be any positive integer.
624 \g1 [5] Backreference to a specific or previous group,
625 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 626 previous group and may optionally be wrapped in
f793d64a
KW
627 curly brackets for safer parsing.
628 \g{name} [5] Named backreference
629 \k<name> [5] Named backreference
630 \K [6] Keep the stuff left of the \K, don't include it in $&
631 \N [7] Any character but \n (experimental). Not affected by
632 /s modifier
633 \v [3] Vertical whitespace
634 \V [3] Not vertical whitespace
635 \h [3] Horizontal whitespace
636 \H [3] Not horizontal whitespace
637 \R [4] Linebreak
e1d1eefb 638
9bb1f947
KW
639=over 4
640
641=item [1]
642
643See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 644
9bb1f947 645=item [2]
b8c5462f 646
9bb1f947 647See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 648
9bb1f947 649=item [3]
5496314a 650
9bb1f947 651See L<perlrecharclass/Backslash sequences> for details.
5496314a 652
9bb1f947 653=item [4]
5496314a 654
9bb1f947 655See L<perlrebackslash/Misc> for details.
d0b16107 656
9bb1f947 657=item [5]
b8c5462f 658
c27a5cfe 659See L</Capture groups> below for details.
93733859 660
9bb1f947 661=item [6]
b8c5462f 662
9bb1f947
KW
663See L</Extended Patterns> below for details.
664
665=item [7]
666
667Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
668character or character sequence whose name is C<NAME>; and similarly
669when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
670code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947
KW
671
672=back
d0b16107 673
04838cea
RGS
674=head3 Assertions
675
a0d0e21e 676Perl defines the following zero-width assertions:
d74e8afc
ITB
677X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
678X<regexp, zero-width assertion>
679X<regular expression, zero-width assertion>
680X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 681
9bb1f947
KW
682 \b Match a word boundary
683 \B Match except at a word boundary
684 \A Match only at beginning of string
685 \Z Match only at end of string, or before newline at the end
686 \z Match only at end of string
687 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 688 of prior m//g)
a0d0e21e 689
14218588 690A word boundary (C<\b>) is a spot between two characters
19799a22
GS
691that has a C<\w> on one side of it and a C<\W> on the other side
692of it (in either order), counting the imaginary characters off the
693beginning and end of the string as matching a C<\W>. (Within
694character classes C<\b> represents backspace rather than a word
695boundary, just as it normally does in any double-quoted string.)
696The C<\A> and C<\Z> are just like "^" and "$", except that they
697won't match multiple times when the C</m> modifier is used, while
698"^" and "$" will match at every internal line boundary. To match
699the actual end of the string and not ignore an optional trailing
700newline, use C<\z>.
d74e8afc 701X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
702
703The C<\G> assertion can be used to chain global matches (using
704C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
705It is also useful when writing C<lex>-like scanners, when you have
706several patterns that you want to match against consequent substrings
0b928c2f 707of your string; see the previous reference. The actual location
19799a22 708where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d 709an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
0b928c2f
FC
710matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
711is modified somewhat, in that contents to the left of C<\G> are
58e23c8d
YO
712not counted when determining the length of the match. Thus the following
713will not match forever:
d74e8afc 714X<\G>
c47ff5f1 715
e761bb84
CO
716 my $string = 'ABC';
717 pos($string) = 1;
718 while ($string =~ /(.\G)/g) {
719 print $1;
720 }
58e23c8d
YO
721
722It will print 'A' and then terminate, as it considers the match to
723be zero-width, and thus will not match at the same position twice in a
724row.
725
726It is worth noting that C<\G> improperly used can result in an infinite
727loop. Take care when using patterns that include C<\G> in an alternation.
728
c27a5cfe 729=head3 Capture groups
04838cea 730
c27a5cfe
KW
731The bracketing construct C<( ... )> creates capture groups (also referred to as
732capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
733the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
734for the second, and so on.
735This is called a I<backreference>.
d74e8afc 736X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 737X<regex, capture group> X<regexp, capture group>
d74e8afc 738X<regular expression, capture buffer> X<backreference>
c27a5cfe 739X<regular expression, capture group> X<backreference>
1f1031fe 740X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
741X<named capture buffer> X<regular expression, named capture buffer>
742X<named capture group> X<regular expression, named capture group>
743X<%+> X<$+{name}> X<< \k<name> >>
744There is no limit to the number of captured substrings that you may use.
745Groups are numbered with the leftmost open parenthesis being number 1, etc. If
746a group did not match, the associated backreference won't match either. (This
747can happen if the group is optional, or in a different branch of an
748alternation.)
749You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
750this form, described below.
751
752You can also refer to capture groups relatively, by using a negative number, so
753that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
754group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
755example:
5624f11d
YO
756
757 /
c27a5cfe
KW
758 (Y) # group 1
759 ( # group 2
760 (X) # group 3
761 \g{-1} # backref to group 3
762 \g{-3} # backref to group 1
5624f11d
YO
763 )
764 /x
765
d8b950dc
KW
766would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
767interpolate regexes into larger regexes and not have to worry about the
768capture groups being renumbered.
769
770You can dispense with numbers altogether and create named capture groups.
771The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
772reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
773also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
774I<name> must not begin with a number, nor contain hyphens.
775When different groups within the same pattern have the same name, any reference
776to that name assumes the leftmost defined group. Named groups count in
777absolute and relative numbering, and so can also be referred to by those
778numbers.
779(It's possible to do things with named capture groups that would otherwise
780require C<(??{})>.)
781
782Capture group contents are dynamically scoped and available to you outside the
783pattern until the end of the enclosing block or until the next successful
784match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
785You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
786etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
787
788Braces are required in referring to named capture groups, but are optional for
789absolute or relative numbered ones. Braces are safer when creating a regex by
790concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
791contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
792is probably not what you intended.
793
794The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
795there were no named nor relative numbered capture groups. Absolute numbered
0b928c2f
FC
796groups were referred to using C<\1>,
797C<\2>, etc., and this notation is still
d8b950dc
KW
798accepted (and likely always will be). But it leads to some ambiguities if
799there are more than 9 capture groups, as C<\10> could mean either the tenth
800capture group, or the character whose ordinal in octal is 010 (a backspace in
801ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
802only if at least 10 left parentheses have opened before it. Likewise C<\11> is
803a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
804And so on. C<\1> through C<\9> are always interpreted as backreferences.
805There are several examples below that illustrate these perils. You can avoid
806the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
807and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
808digits padded with leading zeros, since a leading zero implies an octal
809constant.
d8b950dc
KW
810
811The C<\I<digit>> notation also works in certain circumstances outside
ed7efc79 812the pattern. See L</Warning on \1 Instead of $1> below for details.
81714fb9 813
14218588 814Examples:
a0d0e21e
LW
815
816 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
817
d8b950dc 818 /(.)\g1/ # find first doubled char
81714fb9
YO
819 and print "'$1' is the first doubled character\n";
820
821 /(?<char>.)\k<char>/ # ... a different way
822 and print "'$+{char}' is the first doubled character\n";
823
d8b950dc 824 /(?'char'.)\g1/ # ... mix and match
81714fb9 825 and print "'$1' is the first doubled character\n";
c47ff5f1 826
14218588 827 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
828 $hours = $1;
829 $minutes = $2;
830 $seconds = $3;
a0d0e21e 831 }
c47ff5f1 832
9d860678
KW
833 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
834 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
835 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
836 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
837
838 $a = '(.)\1'; # Creates problems when concatenated.
839 $b = '(.)\g{1}'; # Avoids the problems.
840 "aa" =~ /${a}/; # True
841 "aa" =~ /${b}/; # True
842 "aa0" =~ /${a}0/; # False!
843 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
844 "aa\x08" =~ /${a}0/; # True!
845 "aa\x08" =~ /${b}0/; # False
9d860678 846
14218588
GS
847Several special variables also refer back to portions of the previous
848match. C<$+> returns whatever the last bracket match matched.
849C<$&> returns the entire matched string. (At one point C<$0> did
850also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
851everything before the matched string. C<$'> returns everything
852after the matched string. And C<$^N> contains whatever was matched by
853the most-recently closed group (submatch). C<$^N> can be used in
854extended patterns (see below), for example to assign a submatch to a
81714fb9 855variable.
d74e8afc 856X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 857
d8b950dc
KW
858These special variables, like the C<%+> hash and the numbered match variables
859(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
860until the end of the enclosing block or until the next successful
861match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
862X<$+> X<$^N> X<$&> X<$`> X<$'>
863X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
864
0d017f4d 865B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 866which makes it easier to write code that tests for a series of more
665e98b9
JH
867specific cases and remembers the best match.
868
14218588
GS
869B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
870C<$'> anywhere in the program, it has to provide them for every
871pattern match. This may substantially slow your program. Perl
d8b950dc 872uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
873price for each pattern that contains capturing parentheses. (To
874avoid this cost while retaining the grouping behaviour, use the
875extended regular expression C<(?: ... )> instead.) But if you never
876use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
877parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
878if you can, but if you can't (and some algorithms really appreciate
879them), once you've used them once, use them at will, because you've
880already paid the price. As of 5.005, C<$&> is not so costly as the
881other two.
d74e8afc 882X<$&> X<$`> X<$'>
68dc0745 883
99d59c4d 884As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
885C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
886and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 887successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
888The use of these variables incurs no global performance penalty, unlike
889their punctuation char equivalents, however at the trade-off that you
890have to tell perl when you want to use them.
87e95b7f 891X</p> X<p modifier>
cde0cee5 892
9d727203
KW
893=head2 Quoting metacharacters
894
19799a22
GS
895Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
896C<\w>, C<\n>. Unlike some other regular expression languages, there
897are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 898that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
899interpreted as a literal character, not a metacharacter. This was
900once used in a common idiom to disable or quote the special meanings
901of regular expression metacharacters in a string that you want to
36bbe248 902use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
903
904 $pattern =~ s/(\W)/\\$1/g;
905
f1cbbd6e 906(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
907Today it is more common to use the quotemeta() function or the C<\Q>
908metaquoting escape sequence to disable all metacharacters' special
909meanings like this:
a0d0e21e
LW
910
911 /$unquoted\Q$quoted\E$unquoted/
912
9da458fc
IZ
913Beware that if you put literal backslashes (those not inside
914interpolated variables) between C<\Q> and C<\E>, double-quotish
915backslash interpolation may lead to confusing results. If you
916I<need> to use literal backslashes within C<\Q...\E>,
917consult L<perlop/"Gory details of parsing quoted constructs">.
918
19799a22
GS
919=head2 Extended Patterns
920
14218588 921Perl also defines a consistent extension syntax for features not
0b928c2f
FC
922found in standard tools like B<awk> and
923B<lex>. The syntax for most of these is a
14218588
GS
924pair of parentheses with a question mark as the first thing within
925the parentheses. The character after the question mark indicates
926the extension.
19799a22 927
14218588
GS
928The stability of these extensions varies widely. Some have been
929part of the core language for many years. Others are experimental
930and may change without warning or be completely removed. Check
931the documentation on an individual feature to verify its current
932status.
19799a22 933
14218588
GS
934A question mark was chosen for this and for the minimal-matching
935construct because 1) question marks are rare in older regular
936expressions, and 2) whenever you see one, you should stop and
0b928c2f 937"question" exactly what is going on. That's psychology....
a0d0e21e 938
70ca8714 939=over 4
a0d0e21e 940
cc6b7395 941=item C<(?#text)>
d74e8afc 942X<(?#)>
a0d0e21e 943
14218588 944A comment. The text is ignored. If the C</x> modifier enables
19799a22 945whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
946the comment as soon as it sees a C<)>, so there is no way to put a literal
947C<)> in the comment.
a0d0e21e 948
cfaf538b 949=item C<(?adlupimsx-imsx)>
fb85c044 950
cfaf538b 951=item C<(?^alupimsx)>
fb85c044 952X<(?)> X<(?^)>
19799a22 953
0b6d1084
JH
954One or more embedded pattern-match modifiers, to be turned on (or
955turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
956the remainder of the enclosing pattern group (if any).
957
fb85c044 958This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d 959configuration file, taken from an argument, or specified in a table
0b928c2f
FC
960somewhere. Consider the case where some patterns want to be
961case-sensitive and some do not: The case-insensitive ones merely need to
0d017f4d 962include C<(?i)> at the front of the pattern. For example:
19799a22
GS
963
964 $pattern = "foobar";
5d458dd8 965 if ( /$pattern/i ) { }
19799a22
GS
966
967 # more flexible:
968
969 $pattern = "(?i)foobar";
5d458dd8 970 if ( /$pattern/ ) { }
19799a22 971
0b6d1084 972These modifiers are restored at the end of the enclosing group. For example,
19799a22 973
d8b950dc 974 ( (?i) blah ) \s+ \g1
19799a22 975
0d017f4d
WL
976will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
977repetition of the previous word, assuming the C</x> modifier, and no C</i>
978modifier outside this group.
19799a22 979
8eb5594e
DR
980These modifiers do not carry over into named subpatterns called in the
981enclosing group. In other words, a pattern such as C<((?i)(&NAME))> does not
982change the case-sensitivity of the "NAME" pattern.
983
dc925305
KW
984Any of these modifiers can be set to apply globally to all regular
985expressions compiled within the scope of a C<use re>. See
a0bbd6ff 986L<re/"'/flags' mode">.
dc925305 987
9de15fec
KW
988Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
989after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
990C<"d">) may follow the caret to override it.
991But a minus sign is not legal with it.
992
dc925305 993Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 994that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 995C<u> modifiers are mutually exclusive: specifying one de-specifies the
ed7efc79
KW
996others, and a maximum of one (or two C<a>'s) may appear in the
997construct. Thus, for
0b928c2f 998example, C<(?-p)> will warn when compiled under C<use warnings>;
b6fa137b 999C<(?-d:...)> and C<(?dl:...)> are fatal errors.
9de15fec
KW
1000
1001Note also that the C<p> modifier is special in that its presence
1002anywhere in a pattern has a global effect.
cde0cee5 1003
5a964f20 1004=item C<(?:pattern)>
d74e8afc 1005X<(?:)>
a0d0e21e 1006
cfaf538b 1007=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 1008
cfaf538b 1009=item C<(?^aluimsx:pattern)>
fb85c044
KW
1010X<(?^:)>
1011
5a964f20
TC
1012This is for clustering, not capturing; it groups subexpressions like
1013"()", but doesn't make backreferences as "()" does. So
a0d0e21e 1014
5a964f20 1015 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
1016
1017is like
1018
5a964f20 1019 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 1020
19799a22
GS
1021but doesn't spit out extra fields. It's also cheaper not to capture
1022characters if you don't need to.
a0d0e21e 1023
19799a22 1024Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 1025C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
1026
1027 /(?s-i:more.*than).*million/i
1028
14218588 1029is equivalent to the more verbose
ca9dfc88
IZ
1030
1031 /(?:(?s-i)more.*than).*million/i
1032
fb85c044 1033Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
1034after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
1035flags (except C<"d">) may follow the caret, so
fb85c044
KW
1036
1037 (?^x:foo)
1038
1039is equivalent to
1040
1041 (?x-ims:foo)
1042
1043The caret tells Perl that this cluster doesn't inherit the flags of any
0b928c2f 1044surrounding pattern, but uses the system defaults (C<d-imsx>),
fb85c044
KW
1045modified by any flags specified.
1046
1047The caret allows for simpler stringification of compiled regular
1048expressions. These look like
1049
1050 (?^:pattern)
1051
1052with any non-default flags appearing between the caret and the colon.
1053A test that looks at such stringification thus doesn't need to have the
1054system default flags hard-coded in it, just the caret. If new flags are
1055added to Perl, the meaning of the caret's expansion will change to include
1056the default for those flags, so the test will still work, unchanged.
1057
1058Specifying a negative flag after the caret is an error, as the flag is
1059redundant.
1060
1061Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
1062to match at the beginning.
1063
594d7033
YO
1064=item C<(?|pattern)>
1065X<(?|)> X<Branch reset>
1066
1067This is the "branch reset" pattern, which has the special property
c27a5cfe 1068that the capture groups are numbered from the same starting point
99d59c4d 1069in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 1070
c27a5cfe 1071Capture groups are numbered from left to right, but inside this
693596a8 1072construct the numbering is restarted for each branch.
4deaaa80 1073
c27a5cfe 1074The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
1075following this construct will be numbered as though the construct
1076contained only one branch, that being the one with the most capture
c27a5cfe 1077groups in it.
4deaaa80 1078
0b928c2f 1079This construct is useful when you want to capture one of a
4deaaa80
PJ
1080number of alternative matches.
1081
1082Consider the following pattern. The numbers underneath show in
c27a5cfe 1083which group the captured content will be stored.
594d7033
YO
1084
1085
1086 # before ---------------branch-reset----------- after
1087 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1088 # 1 2 2 3 2 3 4
1089
ab106183
A
1090Be careful when using the branch reset pattern in combination with
1091named captures. Named captures are implemented as being aliases to
c27a5cfe 1092numbered groups holding the captures, and that interferes with the
ab106183
A
1093implementation of the branch reset pattern. If you are using named
1094captures in a branch reset pattern, it's best to use the same names,
1095in the same order, in each of the alternations:
1096
1097 /(?| (?<a> x ) (?<b> y )
1098 | (?<a> z ) (?<b> w )) /x
1099
1100Not doing so may lead to surprises:
1101
1102 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1103 say $+ {a}; # Prints '12'
1104 say $+ {b}; # *Also* prints '12'.
1105
c27a5cfe
KW
1106The problem here is that both the group named C<< a >> and the group
1107named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 1108
ee9b8eae
YO
1109=item Look-Around Assertions
1110X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1111
0b928c2f 1112Look-around assertions are zero-width patterns which match a specific
ee9b8eae
YO
1113pattern without including it in C<$&>. Positive assertions match when
1114their subpattern matches, negative assertions match when their subpattern
1115fails. Look-behind matches text up to the current match position,
1116look-ahead matches text following the current match position.
1117
1118=over 4
1119
5a964f20 1120=item C<(?=pattern)>
d74e8afc 1121X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 1122
19799a22 1123A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
1124matches a word followed by a tab, without including the tab in C<$&>.
1125
5a964f20 1126=item C<(?!pattern)>
d74e8afc 1127X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 1128
19799a22 1129A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 1130matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
1131however that look-ahead and look-behind are NOT the same thing. You cannot
1132use this for look-behind.
7b8d334a 1133
5a964f20 1134If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
1135will not do what you want. That's because the C<(?!foo)> is just saying that
1136the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
0b928c2f 1137match. Use look-behind instead (see below).
c277df42 1138
ee9b8eae
YO
1139=item C<(?<=pattern)> C<\K>
1140X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 1141
c47ff5f1 1142A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
1143matches a word that follows a tab, without including the tab in C<$&>.
1144Works only for fixed-width look-behind.
c277df42 1145
ee9b8eae
YO
1146There is a special form of this construct, called C<\K>, which causes the
1147regex engine to "keep" everything it had matched prior to the C<\K> and
0b928c2f 1148not include it in C<$&>. This effectively provides variable-length
ee9b8eae
YO
1149look-behind. The use of C<\K> inside of another look-around assertion
1150is allowed, but the behaviour is currently not well defined.
1151
c62285ac 1152For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
1153equivalent C<< (?<=...) >> construct, and it is especially useful in
1154situations where you want to efficiently remove something following
1155something else in a string. For instance
1156
1157 s/(foo)bar/$1/g;
1158
1159can be rewritten as the much more efficient
1160
1161 s/foo\Kbar//g;
1162
5a964f20 1163=item C<(?<!pattern)>
d74e8afc 1164X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 1165
19799a22
GS
1166A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
1167matches any occurrence of "foo" that does not follow "bar". Works
1168only for fixed-width look-behind.
c277df42 1169
ee9b8eae
YO
1170=back
1171
81714fb9
YO
1172=item C<(?'NAME'pattern)>
1173
1174=item C<< (?<NAME>pattern) >>
1175X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1176
c27a5cfe 1177A named capture group. Identical in every respect to normal capturing
0b928c2f
FC
1178parentheses C<()> but for the additional fact that the group
1179can be referred to by name in various regular expression
1180constructs (like C<\g{NAME}>) and can be accessed by name
1181after a successful match via C<%+> or C<%->. See L<perlvar>
90a18110 1182for more details on the C<%+> and C<%-> hashes.
81714fb9 1183
c27a5cfe
KW
1184If multiple distinct capture groups have the same name then the
1185$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 1186
0d017f4d 1187The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
1188
1189B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 1190function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
1191numbered sequentially regardless of being named or not. Thus in the
1192pattern
1193
1194 /(x)(?<foo>y)(z)/
1195
1196$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1197the opposite which is what a .NET regex hacker might expect.
1198
1f1031fe
YO
1199Currently NAME is restricted to simple identifiers only.
1200In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1201its Unicode extension (see L<utf8>),
1202though it isn't extended by the locale (see L<perllocale>).
81714fb9 1203
1f1031fe 1204B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 1205with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 1206may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 1207support the use of single quotes as a delimiter for the name.
81714fb9 1208
1f1031fe
YO
1209=item C<< \k<NAME> >>
1210
1211=item C<< \k'NAME' >>
81714fb9
YO
1212
1213Named backreference. Similar to numeric backreferences, except that
1214the group is designated by name and not number. If multiple groups
1215have the same name then it refers to the leftmost defined group in
1216the current match.
1217
0d017f4d 1218It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
1219earlier in the pattern.
1220
1221Both forms are equivalent.
1222
1f1031fe 1223B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 1224with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 1225may be used instead of C<< \k<NAME> >>.
1f1031fe 1226
cc6b7395 1227=item C<(?{ code })>
d74e8afc 1228X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 1229
19799a22 1230B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1231experimental, and may be changed without notice. Code executed that
1232has side effects may not perform identically from version to version
1233due to the effect of future optimisations in the regex engine.
c277df42 1234
cc46d5f2 1235This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
1236always succeeds, and its C<code> is not interpolated. Currently,
1237the rules to determine where the C<code> ends are somewhat convoluted.
1238
77ea4f6d
JV
1239This feature can be used together with the special variable C<$^N> to
1240capture the results of submatches in variables without having to keep
1241track of the number of nested parentheses. For example:
1242
1243 $_ = "The brown fox jumps over the lazy dog";
1244 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1245 print "color = $color, animal = $animal\n";
1246
754091cb
RGS
1247Inside the C<(?{...})> block, C<$_> refers to the string the regular
1248expression is matching against. You can also use C<pos()> to know what is
fa11829f 1249the current position of matching within this string.
754091cb 1250
19799a22
GS
1251The C<code> is properly scoped in the following sense: If the assertion
1252is backtracked (compare L<"Backtracking">), all changes introduced after
1253C<local>ization are undone, so that
b9ac3b5b
GS
1254
1255 $_ = 'a' x 8;
5d458dd8 1256 m<
d1fbf752 1257 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 1258 (
5d458dd8 1259 a
b9ac3b5b 1260 (?{
d1fbf752
KW
1261 local $cnt = $cnt + 1; # Update $cnt,
1262 # backtracking-safe.
b9ac3b5b 1263 })
5d458dd8 1264 )*
b9ac3b5b 1265 aaaa
d1fbf752
KW
1266 (?{ $res = $cnt }) # On success copy to
1267 # non-localized location.
b9ac3b5b
GS
1268 >x;
1269
0d017f4d 1270will set C<$res = 4>. Note that after the match, C<$cnt> returns to the globally
14218588 1271introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
1272are unwound.
1273
19799a22
GS
1274This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
1275switch. If I<not> used in this way, the result of evaluation of
1276C<code> is put into the special variable C<$^R>. This happens
1277immediately, so C<$^R> can be used from other C<(?{ code })> assertions
1278inside the same regular expression.
b9ac3b5b 1279
19799a22
GS
1280The assignment to C<$^R> above is properly localized, so the old
1281value of C<$^R> is restored if the assertion is backtracked; compare
1282L<"Backtracking">.
b9ac3b5b 1283
19799a22
GS
1284For reasons of security, this construct is forbidden if the regular
1285expression involves run-time interpolation of variables, unless the
1286perilous C<use re 'eval'> pragma has been used (see L<re>), or the
0b928c2f 1287variables contain results of the C<qr//> operator (see
b6fa137b 1288L<perlop/"qr/STRINGE<sol>msixpodual">).
871b0233 1289
0d017f4d 1290This restriction is due to the wide-spread and remarkably convenient
19799a22 1291custom of using run-time determined strings as patterns. For example:
871b0233
IZ
1292
1293 $re = <>;
1294 chomp $re;
1295 $string =~ /$re/;
1296
14218588
GS
1297Before Perl knew how to execute interpolated code within a pattern,
1298this operation was completely safe from a security point of view,
1299although it could raise an exception from an illegal pattern. If
1300you turn on the C<use re 'eval'>, though, it is no longer secure,
1301so you should only do so if you are also using taint checking.
1302Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 1303compartment. See L<perlsec> for details about both these mechanisms.
871b0233 1304
e95d7314
GG
1305B<WARNING>: Use of lexical (C<my>) variables in these blocks is
1306broken. The result is unpredictable and will make perl unstable. The
1307workaround is to use global (C<our>) variables.
1308
8525cfae
FC
1309B<WARNING>: In perl 5.12.x and earlier, the regex engine
1310was not re-entrant, so interpolated code could not
1311safely invoke the regex engine either directly with
e95d7314 1312C<m//> or C<s///>), or indirectly with functions such as
8525cfae 1313C<split>. Invoking the regex engine in these blocks would make perl
e95d7314 1314unstable.
8988a1bb 1315
14455d6c 1316=item C<(??{ code })>
d74e8afc
ITB
1317X<(??{})>
1318X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1319
19799a22 1320B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1321experimental, and may be changed without notice. Code executed that
1322has side effects may not perform identically from version to version
1323due to the effect of future optimisations in the regex engine.
0f5d15d6 1324
19799a22
GS
1325This is a "postponed" regular subexpression. The C<code> is evaluated
1326at run time, at the moment this subexpression may match. The result
0b928c2f 1327of evaluation is considered a regular expression and matched as
61528107 1328if it were inserted instead of this construct. Note that this means
c27a5cfe 1329that the contents of capture groups defined inside an eval'ed pattern
6bda09f9 1330are not available outside of the pattern, and vice versa, there is no
b2711834
FC
1331way for the inner pattern returned from the code block to refer to a
1332capture group defined outside. (The code block itself can use C<$1>, etc.,
6cf90346 1333to refer to the enclosing pattern's capture groups.) Thus,
6bda09f9
YO
1334
1335 ('a' x 100)=~/(??{'(.)' x 100})/
1336
81714fb9 1337B<will> match, it will B<not> set $1.
0f5d15d6 1338
428594d9 1339The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
1340where the C<code> ends are currently somewhat convoluted.
1341
1342The following pattern matches a parenthesized group:
0f5d15d6 1343
d1fbf752
KW
1344 $re = qr{
1345 \(
1346 (?:
1347 (?> [^()]+ ) # Non-parens without backtracking
1348 |
1349 (??{ $re }) # Group with matching parens
1350 )*
1351 \)
1352 }x;
0f5d15d6 1353
6bda09f9
YO
1354See also C<(?PARNO)> for a different, more efficient way to accomplish
1355the same task.
1356
0b370c0a
A
1357For reasons of security, this construct is forbidden if the regular
1358expression involves run-time interpolation of variables, unless the
1359perilous C<use re 'eval'> pragma has been used (see L<re>), or the
0b928c2f 1360variables contain results of the C<qr//> operator (see
b6fa137b 1361L<perlop/"qrE<sol>STRINGE<sol>msixpodual">).
0b370c0a 1362
8525cfae
FC
1363In perl 5.12.x and earlier, because the regex engine was not re-entrant,
1364delayed code could not safely invoke the regex engine either directly with
1365C<m//> or C<s///>), or indirectly with functions such as C<split>.
8988a1bb 1366
5d458dd8
YO
1367Recursing deeper than 50 times without consuming any input string will
1368result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
1369changing it requires a custom build.
1370
542fa716
YO
1371=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1372X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1373X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1374X<regex, relative recursion>
6bda09f9 1375
81714fb9 1376Similar to C<(??{ code })> except it does not involve compiling any code,
c27a5cfe
KW
1377instead it treats the contents of a capture group as an independent
1378pattern that must match at the current position. Capture groups
81714fb9 1379contained by the pattern will have the value as determined by the
6bda09f9
YO
1380outermost recursion.
1381
894be9b7 1382PARNO is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1383the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1384the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716 1385C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
c27a5cfe 1386to be relative, with negative numbers indicating preceding capture groups
542fa716 1387and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1388declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1389Note that the counting for relative recursion differs from that of
c27a5cfe 1390relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1391included.
6bda09f9 1392
81714fb9 1393The following pattern matches a function foo() which may contain
f145b7e9 1394balanced parentheses as the argument.
6bda09f9 1395
d1fbf752 1396 $re = qr{ ( # paren group 1 (full function)
81714fb9 1397 foo
d1fbf752 1398 ( # paren group 2 (parens)
6bda09f9 1399 \(
d1fbf752 1400 ( # paren group 3 (contents of parens)
6bda09f9 1401 (?:
d1fbf752 1402 (?> [^()]+ ) # Non-parens without backtracking
6bda09f9 1403 |
d1fbf752 1404 (?2) # Recurse to start of paren group 2
6bda09f9
YO
1405 )*
1406 )
1407 \)
1408 )
1409 )
1410 }x;
1411
1412If the pattern was used as follows
1413
1414 'foo(bar(baz)+baz(bop))'=~/$re/
1415 and print "\$1 = $1\n",
1416 "\$2 = $2\n",
1417 "\$3 = $3\n";
1418
1419the output produced should be the following:
1420
1421 $1 = foo(bar(baz)+baz(bop))
1422 $2 = (bar(baz)+baz(bop))
81714fb9 1423 $3 = bar(baz)+baz(bop)
6bda09f9 1424
c27a5cfe 1425If there is no corresponding capture group defined, then it is a
61528107 1426fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1427string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1428into perl, so changing it requires a custom build.
1429
542fa716
YO
1430The following shows how using negative indexing can make it
1431easier to embed recursive patterns inside of a C<qr//> construct
1432for later use:
1433
1434 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1435 if (/foo $parens \s+ + \s+ bar $parens/x) {
1436 # do something here...
1437 }
1438
81714fb9 1439B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1440PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1441a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1442as atomic. Also, modifiers are resolved at compile time, so constructs
1443like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1444be processed.
6bda09f9 1445
894be9b7
YO
1446=item C<(?&NAME)>
1447X<(?&NAME)>
1448
0d017f4d
WL
1449Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1450parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1451the same name, then it recurses to the leftmost.
1452
1453It is an error to refer to a name that is not declared somewhere in the
1454pattern.
1455
1f1031fe
YO
1456B<NOTE:> In order to make things easier for programmers with experience
1457with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1458may be used instead of C<< (?&NAME) >>.
1f1031fe 1459
e2e6a0f1
YO
1460=item C<(?(condition)yes-pattern|no-pattern)>
1461X<(?()>
286f584a 1462
e2e6a0f1 1463=item C<(?(condition)yes-pattern)>
286f584a 1464
41ef34de
ML
1465Conditional expression. Matches C<yes-pattern> if C<condition> yields
1466a true value, matches C<no-pattern> otherwise. A missing pattern always
1467matches.
1468
1469C<(condition)> should be either an integer in
e2e6a0f1
YO
1470parentheses (which is valid if the corresponding pair of parentheses
1471matched), a look-ahead/look-behind/evaluate zero-width assertion, a
c27a5cfe 1472name in angle brackets or single quotes (which is valid if a group
e2e6a0f1
YO
1473with the given name matched), or the special symbol (R) (true when
1474evaluated inside of recursion or eval). Additionally the R may be
1475followed by a number, (which will be true when evaluated when recursing
1476inside of the appropriate group), or by C<&NAME>, in which case it will
1477be true only when evaluated during recursion in the named group.
1478
1479Here's a summary of the possible predicates:
1480
1481=over 4
1482
1483=item (1) (2) ...
1484
c27a5cfe 1485Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1486
1487=item (<NAME>) ('NAME')
1488
c27a5cfe 1489Checks if a group with the given name has matched something.
e2e6a0f1 1490
f01cd190
FC
1491=item (?=...) (?!...) (?<=...) (?<!...)
1492
1493Checks whether the pattern matches (or does not match, for the '!'
1494variants).
1495
e2e6a0f1
YO
1496=item (?{ CODE })
1497
f01cd190 1498Treats the return value of the code block as the condition.
e2e6a0f1
YO
1499
1500=item (R)
1501
1502Checks if the expression has been evaluated inside of recursion.
1503
1504=item (R1) (R2) ...
1505
1506Checks if the expression has been evaluated while executing directly
1507inside of the n-th capture group. This check is the regex equivalent of
1508
1509 if ((caller(0))[3] eq 'subname') { ... }
1510
1511In other words, it does not check the full recursion stack.
1512
1513=item (R&NAME)
1514
1515Similar to C<(R1)>, this predicate checks to see if we're executing
1516directly inside of the leftmost group with a given name (this is the same
1517logic used by C<(?&NAME)> to disambiguate). It does not check the full
1518stack, but only the name of the innermost active recursion.
1519
1520=item (DEFINE)
1521
1522In this case, the yes-pattern is never directly executed, and no
1523no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1524See below for details.
1525
1526=back
1527
1528For example:
1529
1530 m{ ( \( )?
1531 [^()]+
1532 (?(1) \) )
1533 }x
1534
1535matches a chunk of non-parentheses, possibly included in parentheses
1536themselves.
1537
0b928c2f
FC
1538A special form is the C<(DEFINE)> predicate, which never executes its
1539yes-pattern directly, and does not allow a no-pattern. This allows one to
1540define subpatterns which will be executed only by the recursion mechanism.
e2e6a0f1
YO
1541This way, you can define a set of regular expression rules that can be
1542bundled into any pattern you choose.
1543
1544It is recommended that for this usage you put the DEFINE block at the
1545end of the pattern, and that you name any subpatterns defined within it.
1546
1547Also, it's worth noting that patterns defined this way probably will
1548not be as efficient, as the optimiser is not very clever about
1549handling them.
1550
1551An example of how this might be used is as follows:
1552
2bf803e2 1553 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1554 (?(DEFINE)
2bf803e2
YO
1555 (?<NAME_PAT>....)
1556 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1557 )/x
1558
c27a5cfe
KW
1559Note that capture groups matched inside of recursion are not accessible
1560after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1561necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1562C<$+{NAME}> would be.
286f584a 1563
51a1303c
BF
1564Finally, keep in mind that subpatterns created inside a DEFINE block
1565count towards the absolute and relative number of captures, so this:
1566
1567 my @captures = "a" =~ /(.) # First capture
1568 (?(DEFINE)
1569 (?<EXAMPLE> 1 ) # Second capture
1570 )/x;
1571 say scalar @captures;
1572
1573Will output 2, not 1. This is particularly important if you intend to
1574compile the definitions with the C<qr//> operator, and later
1575interpolate them in another pattern.
1576
c47ff5f1 1577=item C<< (?>pattern) >>
6bda09f9 1578X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1579
19799a22
GS
1580An "independent" subexpression, one which matches the substring
1581that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1582position, and it matches I<nothing other than this substring>. This
19799a22
GS
1583construct is useful for optimizations of what would otherwise be
1584"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1585It may also be useful in places where the "grab all you can, and do not
1586give anything back" semantic is desirable.
19799a22 1587
c47ff5f1 1588For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1589(anchored at the beginning of string, as above) will match I<all>
1590characters C<a> at the beginning of string, leaving no C<a> for
1591C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1592since the match of the subgroup C<a*> is influenced by the following
1593group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1594C<a*ab> will match fewer characters than a standalone C<a*>, since
1595this makes the tail match.
1596
0b928c2f
FC
1597C<< (?>pattern) >> does not disable backtracking altogether once it has
1598matched. It is still possible to backtrack past the construct, but not
1599into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1600
c47ff5f1 1601An effect similar to C<< (?>pattern) >> may be achieved by writing
0b928c2f
FC
1602C<(?=(pattern))\g{-1}>. This matches the same substring as a standalone
1603C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
c47ff5f1 1604makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1605(The difference between these two constructs is that the second one
1606uses a capturing group, thus shifting ordinals of backreferences
1607in the rest of a regular expression.)
1608
1609Consider this pattern:
c277df42 1610
871b0233 1611 m{ \(
e2e6a0f1 1612 (
f793d64a 1613 [^()]+ # x+
e2e6a0f1 1614 |
871b0233
IZ
1615 \( [^()]* \)
1616 )+
e2e6a0f1 1617 \)
871b0233 1618 }x
5a964f20 1619
19799a22
GS
1620That will efficiently match a nonempty group with matching parentheses
1621two levels deep or less. However, if there is no such group, it
1622will take virtually forever on a long string. That's because there
1623are so many different ways to split a long string into several
1624substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1625to a subpattern of the above pattern. Consider how the pattern
1626above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1627seconds, but that each extra letter doubles this time. This
1628exponential performance will make it appear that your program has
14218588 1629hung. However, a tiny change to this pattern
5a964f20 1630
e2e6a0f1
YO
1631 m{ \(
1632 (
f793d64a 1633 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1634 |
871b0233
IZ
1635 \( [^()]* \)
1636 )+
e2e6a0f1 1637 \)
871b0233 1638 }x
c277df42 1639
c47ff5f1 1640which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1641this yourself would be a productive exercise), but finishes in a fourth
1642the time when used on a similar string with 1000000 C<a>s. Be aware,
0b928c2f
FC
1643however, that, when this construct is followed by a
1644quantifier, it currently triggers a warning message under
9f1b1f2d 1645the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1646C<"matches null string many times in regex">.
c277df42 1647
c47ff5f1 1648On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1649effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1650This was only 4 times slower on a string with 1000000 C<a>s.
1651
9da458fc
IZ
1652The "grab all you can, and do not give anything back" semantic is desirable
1653in many situations where on the first sight a simple C<()*> looks like
1654the correct solution. Suppose we parse text with comments being delimited
1655by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1656its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1657the comment delimiter, because it may "give up" some whitespace if
1658the remainder of the pattern can be made to match that way. The correct
1659answer is either one of these:
1660
1661 (?>#[ \t]*)
1662 #[ \t]*(?![ \t])
1663
1664For example, to grab non-empty comments into $1, one should use either
1665one of these:
1666
1667 / (?> \# [ \t]* ) ( .+ ) /x;
1668 / \# [ \t]* ( [^ \t] .* ) /x;
1669
1670Which one you pick depends on which of these expressions better reflects
1671the above specification of comments.
1672
6bda09f9
YO
1673In some literature this construct is called "atomic matching" or
1674"possessive matching".
1675
b9b4dddf
YO
1676Possessive quantifiers are equivalent to putting the item they are applied
1677to inside of one of these constructs. The following equivalences apply:
1678
1679 Quantifier Form Bracketing Form
1680 --------------- ---------------
1681 PAT*+ (?>PAT*)
1682 PAT++ (?>PAT+)
1683 PAT?+ (?>PAT?)
1684 PAT{min,max}+ (?>PAT{min,max})
1685
e2e6a0f1
YO
1686=back
1687
1688=head2 Special Backtracking Control Verbs
1689
1690B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1691removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1692be noted to avoid problems during upgrades.
1693
1694These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1695otherwise stated the ARG argument is optional; in some cases, it is
1696forbidden.
1697
1698Any pattern containing a special backtracking verb that allows an argument
e1020413 1699has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1700C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1701rules apply:
e2e6a0f1 1702
5d458dd8
YO
1703On failure, the C<$REGERROR> variable will be set to the ARG value of the
1704verb pattern, if the verb was involved in the failure of the match. If the
1705ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1706name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1707none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1708
5d458dd8
YO
1709On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1710the C<$REGMARK> variable will be set to the name of the last
1711C<(*MARK:NAME)> pattern executed. See the explanation for the
1712C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1713
5d458dd8 1714B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
0b928c2f 1715and most other regex-related variables. They are not local to a scope, nor
5d458dd8
YO
1716readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1717Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1718
1719If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1720argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1 1721
70ca8714 1722=over 3
e2e6a0f1
YO
1723
1724=item Verbs that take an argument
1725
1726=over 4
1727
5d458dd8 1728=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1729X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1730
5d458dd8
YO
1731This zero-width pattern prunes the backtracking tree at the current point
1732when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1733where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1734A may backtrack as necessary to match. Once it is reached, matching
1735continues in B, which may also backtrack as necessary; however, should B
1736not match, then no further backtracking will take place, and the pattern
1737will fail outright at the current starting position.
54612592
YO
1738
1739The following example counts all the possible matching strings in a
1740pattern (without actually matching any of them).
1741
e2e6a0f1 1742 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1743 print "Count=$count\n";
1744
1745which produces:
1746
1747 aaab
1748 aaa
1749 aa
1750 a
1751 aab
1752 aa
1753 a
1754 ab
1755 a
1756 Count=9
1757
5d458dd8 1758If we add a C<(*PRUNE)> before the count like the following
54612592 1759
5d458dd8 1760 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1761 print "Count=$count\n";
1762
0b928c2f 1763we prevent backtracking and find the count of the longest matching string
353c6505 1764at each matching starting point like so:
54612592
YO
1765
1766 aaab
1767 aab
1768 ab
1769 Count=3
1770
5d458dd8 1771Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1772
5d458dd8
YO
1773See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1774control backtracking. In some cases, the use of C<(*PRUNE)> can be
1775replaced with a C<< (?>pattern) >> with no functional difference; however,
1776C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1777C<< (?>pattern) >> alone.
54612592 1778
5d458dd8
YO
1779=item C<(*SKIP)> C<(*SKIP:NAME)>
1780X<(*SKIP)>
e2e6a0f1 1781
5d458dd8 1782This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1783failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1784to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1785of this pattern. This effectively means that the regex engine "skips" forward
1786to this position on failure and tries to match again, (assuming that
1787there is sufficient room to match).
1788
1789The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1790C<(*MARK:NAME)> was encountered while matching, then it is that position
1791which is used as the "skip point". If no C<(*MARK)> of that name was
1792encountered, then the C<(*SKIP)> operator has no effect. When used
1793without a name the "skip point" is where the match point was when
1794executing the (*SKIP) pattern.
1795
0b928c2f 1796Compare the following to the examples in C<(*PRUNE)>; note the string
24b23f37
YO
1797is twice as long:
1798
d1fbf752
KW
1799 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1800 print "Count=$count\n";
24b23f37
YO
1801
1802outputs
1803
1804 aaab
1805 aaab
1806 Count=2
1807
5d458dd8 1808Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1809executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1810C<(*SKIP)> was executed.
1811
5d458dd8 1812=item C<(*MARK:NAME)> C<(*:NAME)>
b16db30f 1813X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
5d458dd8
YO
1814
1815This zero-width pattern can be used to mark the point reached in a string
1816when a certain part of the pattern has been successfully matched. This
1817mark may be given a name. A later C<(*SKIP)> pattern will then skip
1818forward to that point if backtracked into on failure. Any number of
b4222fa9 1819C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1820
1821In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1822can be used to "label" a pattern branch, so that after matching, the
1823program can determine which branches of the pattern were involved in the
1824match.
1825
1826When a match is successful, the C<$REGMARK> variable will be set to the
1827name of the most recently executed C<(*MARK:NAME)> that was involved
1828in the match.
1829
1830This can be used to determine which branch of a pattern was matched
c27a5cfe 1831without using a separate capture group for each branch, which in turn
5d458dd8
YO
1832can result in a performance improvement, as perl cannot optimize
1833C</(?:(x)|(y)|(z))/> as efficiently as something like
1834C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1835
1836When a match has failed, and unless another verb has been involved in
1837failing the match and has provided its own name to use, the C<$REGERROR>
1838variable will be set to the name of the most recently executed
1839C<(*MARK:NAME)>.
1840
42ac7c82 1841See L</(*SKIP)> for more details.
5d458dd8 1842
b62d2d15
YO
1843As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1844
5d458dd8
YO
1845=item C<(*THEN)> C<(*THEN:NAME)>
1846
ac9d8485 1847This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1848C<(*PRUNE)>, this verb always matches, and when backtracked into on
1849failure, it causes the regex engine to try the next alternation in the
ac9d8485
FC
1850innermost enclosing group (capturing or otherwise) that has alternations.
1851The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1852count as an alternation, as far as C<(*THEN)> is concerned.
5d458dd8
YO
1853
1854Its name comes from the observation that this operation combined with the
1855alternation operator (C<|>) can be used to create what is essentially a
1856pattern-based if/then/else block:
1857
1858 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1859
1860Note that if this operator is used and NOT inside of an alternation then
1861it acts exactly like the C<(*PRUNE)> operator.
1862
1863 / A (*PRUNE) B /
1864
1865is the same as
1866
1867 / A (*THEN) B /
1868
1869but
1870
1871 / ( A (*THEN) B | C (*THEN) D ) /
1872
1873is not the same as
1874
1875 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1876
1877as after matching the A but failing on the B the C<(*THEN)> verb will
1878backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1879
cbeadc21
JV
1880=back
1881
1882=item Verbs without an argument
1883
1884=over 4
1885
e2e6a0f1
YO
1886=item C<(*COMMIT)>
1887X<(*COMMIT)>
24b23f37 1888
241e7389 1889This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1890zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1891into on failure it causes the match to fail outright. No further attempts
1892to find a valid match by advancing the start pointer will occur again.
1893For example,
24b23f37 1894
d1fbf752
KW
1895 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
1896 print "Count=$count\n";
24b23f37
YO
1897
1898outputs
1899
1900 aaab
1901 Count=1
1902
e2e6a0f1
YO
1903In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1904does not match, the regex engine will not try any further matching on the
1905rest of the string.
c277df42 1906
e2e6a0f1
YO
1907=item C<(*FAIL)> C<(*F)>
1908X<(*FAIL)> X<(*F)>
9af228c6 1909
e2e6a0f1
YO
1910This pattern matches nothing and always fails. It can be used to force the
1911engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1912fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1913
e2e6a0f1 1914It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1915
e2e6a0f1
YO
1916=item C<(*ACCEPT)>
1917X<(*ACCEPT)>
9af228c6 1918
e2e6a0f1
YO
1919B<WARNING:> This feature is highly experimental. It is not recommended
1920for production code.
9af228c6 1921
e2e6a0f1
YO
1922This pattern matches nothing and causes the end of successful matching at
1923the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1924whether there is actually more to match in the string. When inside of a
0d017f4d 1925nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1926via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1927
c27a5cfe 1928If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1929marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1930For instance:
9af228c6 1931
e2e6a0f1 1932 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1933
e2e6a0f1 1934will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0b928c2f 1935be set. If another branch in the inner parentheses was matched, such as in the
e2e6a0f1 1936string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1937
1938=back
c277df42 1939
a0d0e21e
LW
1940=back
1941
c07a80fd 1942=head2 Backtracking
d74e8afc 1943X<backtrack> X<backtracking>
c07a80fd 1944
35a734be
IZ
1945NOTE: This section presents an abstract approximation of regular
1946expression behavior. For a more rigorous (and complicated) view of
1947the rules involved in selecting a match among possible alternatives,
0d017f4d 1948see L<Combining RE Pieces>.
35a734be 1949
c277df42 1950A fundamental feature of regular expression matching involves the
5a964f20 1951notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1952by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1953C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1954internally, but the general principle outlined here is valid.
c07a80fd
PP
1955
1956For a regular expression to match, the I<entire> regular expression must
1957match, not just part of it. So if the beginning of a pattern containing a
1958quantifier succeeds in a way that causes later parts in the pattern to
1959fail, the matching engine backs up and recalculates the beginning
1960part--that's why it's called backtracking.
1961
1962Here is an example of backtracking: Let's say you want to find the
1963word following "foo" in the string "Food is on the foo table.":
1964
1965 $_ = "Food is on the foo table.";
1966 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 1967 print "$2 follows $1.\n";
c07a80fd
PP
1968 }
1969
1970When the match runs, the first part of the regular expression (C<\b(foo)>)
1971finds a possible match right at the beginning of the string, and loads up
1972$1 with "Foo". However, as soon as the matching engine sees that there's
1973no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1974mistake and starts over again one character after where it had the
c07a80fd
PP
1975tentative match. This time it goes all the way until the next occurrence
1976of "foo". The complete regular expression matches this time, and you get
1977the expected output of "table follows foo."
1978
1979Sometimes minimal matching can help a lot. Imagine you'd like to match
1980everything between "foo" and "bar". Initially, you write something
1981like this:
1982
1983 $_ = "The food is under the bar in the barn.";
1984 if ( /foo(.*)bar/ ) {
f793d64a 1985 print "got <$1>\n";
c07a80fd
PP
1986 }
1987
1988Which perhaps unexpectedly yields:
1989
1990 got <d is under the bar in the >
1991
1992That's because C<.*> was greedy, so you get everything between the
14218588 1993I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1994to use minimal matching to make sure you get the text between a "foo"
1995and the first "bar" thereafter.
1996
1997 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1998 got <d is under the >
1999
0d017f4d 2000Here's another example. Let's say you'd like to match a number at the end
b6e13d97 2001of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
2002So you write this:
2003
2004 $_ = "I have 2 numbers: 53147";
f793d64a
KW
2005 if ( /(.*)(\d*)/ ) { # Wrong!
2006 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
2007 }
2008
2009That won't work at all, because C<.*> was greedy and gobbled up the
2010whole string. As C<\d*> can match on an empty string the complete
2011regular expression matched successfully.
2012
8e1088bc 2013 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
2014
2015Here are some variants, most of which don't work:
2016
2017 $_ = "I have 2 numbers: 53147";
2018 @pats = qw{
f793d64a
KW
2019 (.*)(\d*)
2020 (.*)(\d+)
2021 (.*?)(\d*)
2022 (.*?)(\d+)
2023 (.*)(\d+)$
2024 (.*?)(\d+)$
2025 (.*)\b(\d+)$
2026 (.*\D)(\d+)$
c07a80fd
PP
2027 };
2028
2029 for $pat (@pats) {
f793d64a
KW
2030 printf "%-12s ", $pat;
2031 if ( /$pat/ ) {
2032 print "<$1> <$2>\n";
2033 } else {
2034 print "FAIL\n";
2035 }
c07a80fd
PP
2036 }
2037
2038That will print out:
2039
2040 (.*)(\d*) <I have 2 numbers: 53147> <>
2041 (.*)(\d+) <I have 2 numbers: 5314> <7>
2042 (.*?)(\d*) <> <>
2043 (.*?)(\d+) <I have > <2>
2044 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
2045 (.*?)(\d+)$ <I have 2 numbers: > <53147>
2046 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2047 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2048
2049As you see, this can be a bit tricky. It's important to realize that a
2050regular expression is merely a set of assertions that gives a definition
2051of success. There may be 0, 1, or several different ways that the
2052definition might succeed against a particular string. And if there are
5a964f20
TC
2053multiple ways it might succeed, you need to understand backtracking to
2054know which variety of success you will achieve.
c07a80fd 2055
19799a22 2056When using look-ahead assertions and negations, this can all get even
8b19b778 2057trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
2058followed by "123". You might try to write that as
2059
871b0233 2060 $_ = "ABC123";
f793d64a
KW
2061 if ( /^\D*(?!123)/ ) { # Wrong!
2062 print "Yup, no 123 in $_\n";
871b0233 2063 }
c07a80fd
PP
2064
2065But that isn't going to match; at least, not the way you're hoping. It
2066claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 2067why that pattern matches, contrary to popular expectations:
c07a80fd 2068
4358a253
SS
2069 $x = 'ABC123';
2070 $y = 'ABC445';
c07a80fd 2071
4358a253
SS
2072 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2073 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 2074
4358a253
SS
2075 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2076 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
2077
2078This prints
2079
2080 2: got ABC
2081 3: got AB
2082 4: got ABC
2083
5f05dabc 2084You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
2085general purpose version of test 1. The important difference between
2086them is that test 3 contains a quantifier (C<\D*>) and so can use
2087backtracking, whereas test 1 will not. What's happening is
2088that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 2089non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 2090let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 2091fail.
14218588 2092
c07a80fd 2093The search engine will initially match C<\D*> with "ABC". Then it will
0b928c2f 2094try to match C<(?!123)> with "123", which fails. But because
c07a80fd
PP
2095a quantifier (C<\D*>) has been used in the regular expression, the
2096search engine can backtrack and retry the match differently
54310121 2097in the hope of matching the complete regular expression.
c07a80fd 2098
5a964f20
TC
2099The pattern really, I<really> wants to succeed, so it uses the
2100standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 2101time. Now there's indeed something following "AB" that is not
14218588 2102"123". It's "C123", which suffices.
c07a80fd 2103
14218588
GS
2104We can deal with this by using both an assertion and a negation.
2105We'll say that the first part in $1 must be followed both by a digit
2106and by something that's not "123". Remember that the look-aheads
2107are zero-width expressions--they only look, but don't consume any
2108of the string in their match. So rewriting this way produces what
c07a80fd
PP
2109you'd expect; that is, case 5 will fail, but case 6 succeeds:
2110
4358a253
SS
2111 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2112 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
2113
2114 6: got ABC
2115
5a964f20 2116In other words, the two zero-width assertions next to each other work as though
19799a22 2117they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
2118matches only if you're at the beginning of the line AND the end of the
2119line simultaneously. The deeper underlying truth is that juxtaposition in
2120regular expressions always means AND, except when you write an explicit OR
2121using the vertical bar. C</ab/> means match "a" AND (then) match "b",
2122although the attempted matches are made at different positions because "a"
2123is not a zero-width assertion, but a one-width assertion.
2124
0d017f4d 2125B<WARNING>: Particularly complicated regular expressions can take
14218588 2126exponential time to solve because of the immense number of possible
0d017f4d 2127ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
2128internal optimizations done by the regular expression engine, this will
2129take a painfully long time to run:
c07a80fd 2130
e1901655
IZ
2131 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2132
2133And if you used C<*>'s in the internal groups instead of limiting them
2134to 0 through 5 matches, then it would take forever--or until you ran
2135out of stack space. Moreover, these internal optimizations are not
2136always applicable. For example, if you put C<{0,5}> instead of C<*>
2137on the external group, no current optimization is applicable, and the
2138match takes a long time to finish.
c07a80fd 2139
9da458fc
IZ
2140A powerful tool for optimizing such beasts is what is known as an
2141"independent group",
96090e4f 2142which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 2143zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 2144the tail match, since they are in "logical" context: only
14218588 2145whether they match is considered relevant. For an example
9da458fc 2146where side-effects of look-ahead I<might> have influenced the
96090e4f 2147following match, see L</C<< (?>pattern) >>>.
c277df42 2148
a0d0e21e 2149=head2 Version 8 Regular Expressions
d74e8afc 2150X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 2151
5a964f20 2152In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
2153routines, here are the pattern-matching rules not described above.
2154
54310121 2155Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 2156with a special meaning described here or above. You can cause
5a964f20 2157characters that normally function as metacharacters to be interpreted
5f05dabc 2158literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
2159character; "\\" matches a "\"). This escape mechanism is also required
2160for the character used as the pattern delimiter.
2161
2162A series of characters matches that series of characters in the target
0b928c2f 2163string, so the pattern C<blurfl> would match "blurfl" in the target
0d017f4d 2164string.
a0d0e21e
LW
2165
2166You can specify a character class, by enclosing a list of characters
5d458dd8 2167in C<[]>, which will match any character from the list. If the
a0d0e21e 2168first character after the "[" is "^", the class matches any character not
14218588 2169in the list. Within a list, the "-" character specifies a
5a964f20 2170range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
2171inclusive. If you want either "-" or "]" itself to be a member of a
2172class, put it at the start of the list (possibly after a "^"), or
2173escape it with a backslash. "-" is also taken literally when it is
2174at the end of the list, just before the closing "]". (The
84850974
DD
2175following all specify the same class of three characters: C<[-az]>,
2176C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
2177specifies a class containing twenty-six characters, even on EBCDIC-based
2178character sets.) Also, if you try to use the character
2179classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2180a range, the "-" is understood literally.
a0d0e21e 2181
8ada0baa
JH
2182Note also that the whole range idea is rather unportable between
2183character sets--and even within character sets they may cause results
2184you probably didn't expect. A sound principle is to use only ranges
0d017f4d 2185that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
2186[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
2187spell out the character sets in full.
2188
54310121 2189Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
2190used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2191"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 2192of three octal digits, matches the character whose coded character set value
5d458dd8 2193is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 2194matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 2195matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 2196matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
2197
2198You can specify a series of alternatives for a pattern using "|" to
2199separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 2200or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e 2201first alternative includes everything from the last pattern delimiter
0b928c2f 2202("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
a0d0e21e 2203the last alternative contains everything from the last "|" to the next
0b928c2f 2204closing pattern delimiter. That's why it's common practice to include
14218588 2205alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
2206start and end.
2207
5a964f20 2208Alternatives are tried from left to right, so the first
a3cb178b
GS
2209alternative found for which the entire expression matches, is the one that
2210is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 2211example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
2212part will match, as that is the first alternative tried, and it successfully
2213matches the target string. (This might not seem important, but it is
2214important when you are capturing matched text using parentheses.)
2215
5a964f20 2216Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 2217so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 2218
14218588
GS
2219Within a pattern, you may designate subpatterns for later reference
2220by enclosing them in parentheses, and you may refer back to the
2221I<n>th subpattern later in the pattern using the metacharacter
0b928c2f 2222\I<n> or \gI<n>. Subpatterns are numbered based on the left to right order
14218588
GS
2223of their opening parenthesis. A backreference matches whatever
2224actually matched the subpattern in the string being examined, not
d8b950dc 2225the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
2226match "0x1234 0x4321", but not "0x1234 01234", because subpattern
22271 matched "0x", even though the rule C<0|0x> could potentially match
2228the leading 0 in the second number.
cb1a09d0 2229
0d017f4d 2230=head2 Warning on \1 Instead of $1
cb1a09d0 2231
5a964f20 2232Some people get too used to writing things like:
cb1a09d0
AD
2233
2234 $pattern =~ s/(\W)/\\\1/g;
2235
3ff1c45a
KW
2236This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2237shocking the
cb1a09d0 2238B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 2239PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
2240the usual double-quoted string means a control-A. The customary Unix
2241meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
2242of doing that, you get yourself into trouble if you then add an C</e>
2243modifier.
2244
f793d64a 2245 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
2246
2247Or if you try to do
2248
2249 s/(\d+)/\1000/;
2250
2251You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 2252C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
2253with the operation of matching a backreference. Certainly they mean two
2254different things on the I<left> side of the C<s///>.
9fa51da4 2255
0d017f4d 2256=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 2257
19799a22 2258B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
2259
2260Regular expressions provide a terse and powerful programming language. As
2261with most other power tools, power comes together with the ability
2262to wreak havoc.
2263
2264A common abuse of this power stems from the ability to make infinite
628afcb5 2265loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
2266
2267 'foo' =~ m{ ( o? )* }x;
2268
0d017f4d 2269The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 2270in the string is not moved by the match, C<o?> would match again and again
527e91da 2271because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
2272is with the looping modifier C<//g>:
2273
2274 @matches = ( 'foo' =~ m{ o? }xg );
2275
2276or
2277
2278 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2279
2280or the loop implied by split().
2281
2282However, long experience has shown that many programming tasks may
14218588
GS
2283be significantly simplified by using repeated subexpressions that
2284may match zero-length substrings. Here's a simple example being:
c84d73f1 2285
d1fbf752 2286 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
2287 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2288
9da458fc 2289Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 2290the infinite loop>. The rules for this are different for lower-level
527e91da 2291loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2292ones like the C</g> modifier or split() operator.
2293
19799a22
GS
2294The lower-level loops are I<interrupted> (that is, the loop is
2295broken) when Perl detects that a repeated expression matched a
2296zero-length substring. Thus
c84d73f1
IZ
2297
2298 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2299
5d458dd8 2300is made equivalent to
c84d73f1 2301
0b928c2f
FC
2302 m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2303
2304For example, this program
2305
2306 #!perl -l
2307 "aaaaab" =~ /
2308 (?:
2309 a # non-zero
2310 | # or
2311 (?{print "hello"}) # print hello whenever this
2312 # branch is tried
2313 (?=(b)) # zero-width assertion
2314 )* # any number of times
2315 /x;
2316 print $&;
2317 print $1;
c84d73f1 2318
0b928c2f
FC
2319prints
2320
2321 hello
2322 aaaaa
2323 b
2324
2325Notice that "hello" is only printed once, as when Perl sees that the sixth
2326iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2327the C<*>.
2328
2329The higher-level loops preserve an additional state between iterations:
5d458dd8 2330whether the last match was zero-length. To break the loop, the following
c84d73f1 2331match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2332This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2333and so the I<second best> match is chosen if the I<best> match is of
2334zero length.
2335
19799a22 2336For example:
c84d73f1
IZ
2337
2338 $_ = 'bar';
2339 s/\w??/<$&>/g;
2340
20fb949f 2341results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2342match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2343best> match is what is matched by C<\w>. Thus zero-length matches
2344alternate with one-character-long matches.
2345
5d458dd8 2346Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2347position one notch further in the string.
2348
19799a22 2349The additional state of being I<matched with zero-length> is associated with
c84d73f1 2350the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2351Zero-length matches at the end of the previous match are ignored
2352during C<split>.
c84d73f1 2353
0d017f4d 2354=head2 Combining RE Pieces
35a734be
IZ
2355
2356Each of the elementary pieces of regular expressions which were described
2357before (such as C<ab> or C<\Z>) could match at most one substring
2358at the given position of the input string. However, in a typical regular
2359expression these elementary pieces are combined into more complicated
0b928c2f 2360patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
35a734be
IZ
2361(in these examples C<S> and C<T> are regular subexpressions).
2362
2363Such combinations can include alternatives, leading to a problem of choice:
2364if we match a regular expression C<a|ab> against C<"abc">, will it match
2365substring C<"a"> or C<"ab">? One way to describe which substring is
2366actually matched is the concept of backtracking (see L<"Backtracking">).
2367However, this description is too low-level and makes you think
2368in terms of a particular implementation.
2369
2370Another description starts with notions of "better"/"worse". All the
2371substrings which may be matched by the given regular expression can be
2372sorted from the "best" match to the "worst" match, and it is the "best"
2373match which is chosen. This substitutes the question of "what is chosen?"
2374by the question of "which matches are better, and which are worse?".
2375
2376Again, for elementary pieces there is no such question, since at most
2377one match at a given position is possible. This section describes the
2378notion of better/worse for combining operators. In the description
2379below C<S> and C<T> are regular subexpressions.
2380
13a2d996 2381=over 4
35a734be
IZ
2382
2383=item C<ST>
2384
2385Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2386substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2387which can be matched by C<T>.
35a734be 2388
0b928c2f 2389If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
35a734be
IZ
2390match than C<A'B'>.
2391
2392If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
0b928c2f 2393C<B> is a better match for C<T> than C<B'>.
35a734be
IZ
2394
2395=item C<S|T>
2396
2397When C<S> can match, it is a better match than when only C<T> can match.
2398
2399Ordering of two matches for C<S> is the same as for C<S>. Similar for
2400two matches for C<T>.
2401
2402=item C<S{REPEAT_COUNT}>
2403
2404Matches as C<SSS...S> (repeated as many times as necessary).
2405
2406=item C<S{min,max}>
2407
2408Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2409
2410=item C<S{min,max}?>
2411
2412Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2413
2414=item C<S?>, C<S*>, C<S+>
2415
2416Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2417
2418=item C<S??>, C<S*?>, C<S+?>
2419
2420Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2421
c47ff5f1 2422=item C<< (?>S) >>
35a734be
IZ
2423
2424Matches the best match for C<S> and only that.
2425
2426=item C<(?=S)>, C<(?<=S)>
2427
2428Only the best match for C<S> is considered. (This is important only if
2429C<S> has capturing parentheses, and backreferences are used somewhere
2430else in the whole regular expression.)
2431
2432=item C<(?!S)>, C<(?<!S)>
2433
2434For this grouping operator there is no need to describe the ordering, since
2435only whether or not C<S> can match is important.
2436
6bda09f9 2437=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2438
2439The ordering is the same as for the regular expression which is
c27a5cfe 2440the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
2441
2442=item C<(?(condition)yes-pattern|no-pattern)>
2443
2444Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2445already determined. The ordering of the matches is the same as for the
2446chosen subexpression.
2447
2448=back
2449
2450The above recipes describe the ordering of matches I<at a given position>.
2451One more rule is needed to understand how a match is determined for the
2452whole regular expression: a match at an earlier position is always better
2453than a match at a later position.
2454
0d017f4d 2455=head2 Creating Custom RE Engines
c84d73f1 2456
0b928c2f
FC
2457As of Perl 5.10.0, one can create custom regular expression engines. This
2458is not for the faint of heart, as they have to plug in at the C level. See
2459L<perlreapi> for more details.
2460
2461As an alternative, overloaded constants (see L<overload>) provide a simple
2462way to extend the functionality of the RE engine, by substituting one
2463pattern for another.
c84d73f1
IZ
2464
2465Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2466matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2467characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2468at these positions, so we want to have each C<\Y|> in the place of the
2469more complicated version. We can create a module C<customre> to do
2470this:
2471
2472 package customre;
2473 use overload;
2474
2475 sub import {
2476 shift;
2477 die "No argument to customre::import allowed" if @_;
2478 overload::constant 'qr' => \&convert;
2479 }
2480
2481 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2482
580a9fe1
RGS
2483 # We must also take care of not escaping the legitimate \\Y|
2484 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2485 my %rules = ( '\\' => '\\\\',
f793d64a 2486 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2487 sub convert {
2488 my $re = shift;
5d458dd8 2489 $re =~ s{
c84d73f1
IZ
2490 \\ ( \\ | Y . )
2491 }
5d458dd8 2492 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2493 return $re;
2494 }
2495
2496Now C<use customre> enables the new escape in constant regular
2497expressions, i.e., those without any runtime variable interpolations.
2498As documented in L<overload>, this conversion will work only over
2499literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2500part of this regular expression needs to be converted explicitly
2501(but only if the special meaning of C<\Y|> should be enabled inside $re):
2502
2503 use customre;
2504 $re = <>;
2505 chomp $re;
2506 $re = customre::convert $re;
2507 /\Y|$re\Y|/;
2508
0b928c2f 2509=head2 PCRE/Python Support
1f1031fe 2510
0b928c2f 2511As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
1f1031fe 2512to the regex syntax. While Perl programmers are encouraged to use the
0b928c2f 2513Perl-specific syntax, the following are also accepted:
1f1031fe
YO
2514
2515=over 4
2516
ae5648b3 2517=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2518
c27a5cfe 2519Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2520
2521=item C<< (?P=NAME) >>
2522
c27a5cfe 2523Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2524
2525=item C<< (?P>NAME) >>
2526
c27a5cfe 2527Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2528
ee9b8eae 2529=back
1f1031fe 2530
19799a22
GS
2531=head1 BUGS
2532
88c9975e
KW
2533Many regular expression constructs don't work on EBCDIC platforms.
2534
ed7efc79
KW
2535There are a number of issues with regard to case-insensitive matching
2536in Unicode rules. See C<i> under L</Modifiers> above.
2537
9da458fc
IZ
2538This document varies from difficult to understand to completely
2539and utterly opaque. The wandering prose riddled with jargon is
2540hard to fathom in several places.
2541
2542This document needs a rewrite that separates the tutorial content
2543from the reference content.
19799a22
GS
2544
2545=head1 SEE ALSO
9fa51da4 2546
91e0c79e
MJD
2547L<perlrequick>.
2548
2549L<perlretut>.
2550
9b599b2a
GS
2551L<perlop/"Regexp Quote-Like Operators">.
2552
1e66bd83
PP
2553L<perlop/"Gory details of parsing quoted constructs">.
2554
14218588
GS
2555L<perlfaq6>.
2556
9b599b2a
GS
2557L<perlfunc/pos>.
2558
2559L<perllocale>.
2560
fb55449c
JH
2561L<perlebcdic>.
2562
14218588
GS
2563I<Mastering Regular Expressions> by Jeffrey Friedl, published
2564by O'Reilly and Associates.