xref: /openbsd-src/gnu/usr.bin/perl/pod/perlre.pod (revision f2da64fbbbf1b03f09f390ab01267c93dfd77c4c)
1=head1 NAME
2X<regular expression> X<regex> X<regexp>
3
4perlre - Perl regular expressions
5
6=head1 DESCRIPTION
7
8This page describes the syntax of regular expressions in Perl.
9
10If you haven't used regular expressions before, a quick-start
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">.
18
19
20=head2 Modifiers
21
22Matching operations can have various modifiers.  Modifiers
23that relate to the interpretation of the regular expression inside
24are listed below.  Modifiers that alter the way a regular expression
25is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
26L<perlop/"Gory details of parsing quoted constructs">.
27
28=over 4
29
30=item m
31X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
32
33Treat string as multiple lines.  That is, change "^" and "$" from matching
34the start of the string's first line and the end of its last line to
35matching the start and end of each line within the string.
36
37=item s
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
40
41Treat string as single line.  That is, change "." to match any character
42whatsoever, even a newline, which normally it would not match.
43
44Used together, as C</ms>, they let the "." match any character whatsoever,
45while still allowing "^" and "$" to match, respectively, just after
46and just before newlines within the string.
47
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
54If locale matching rules are in effect, the case map is taken from the
55current
56locale for code points less than 255, and from Unicode rules for larger
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
75Perl doesn't match multiple characters in a bracketed
76character class unless the character that maps to them is explicitly
77mentioned, and it doesn't match them at all if the character class is
78inverted, which otherwise could be highly confusing.  See
79L<perlrecharclass/Bracketed Character Classes>, and
80L<perlrecharclass/Negation>.
81
82=item x
83X</x>
84
85Extend your pattern's legibility by permitting whitespace and comments.
86Details in L</"/x">
87
88=item p
89X</p> X<regex, preserve> X<regexp, preserve>
90
91Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
92${^POSTMATCH} are available for use after matching.
93
94In Perl 5.20 and higher this is ignored. Due to a new copy-on-write
95mechanism, ${^PREMATCH}, ${^MATCH}, and ${^POSTMATCH} will be available
96after the match regardless of the modifier.
97
98=item a, d, l and u
99X</a> X</d> X</l> X</u>
100
101These modifiers, all new in 5.14, affect which character-set rules
102(Unicode, etc.) are used, as described below in
103L</Character set modifiers>.
104
105=item Other Modifiers
106
107There are a number of flags that can be found at the end of regular
108expression constructs that are I<not> generic regular expression flags, but
109apply to the operation being performed, like matching or substitution (C<m//>
110or C<s///> respectively).
111
112Flags described further in
113L<perlretut/"Using regular expressions in Perl"> are:
114
115  c  - keep the current position during repeated matching
116  g  - globally match the pattern repeatedly in the string
117
118Substitution-specific modifiers described in
119
120L<perlop/"s/PATTERN/REPLACEMENT/msixpodualgcer"> are:
121
122  e  - evaluate the right-hand side as an expression
123  ee - evaluate the right side as a string then eval the result
124  o  - pretend to optimize your code, but actually introduce bugs
125  r  - perform non-destructive substitution and return the new value
126
127=back
128
129Regular expression modifiers are usually written in documentation
130as e.g., "the C</x> modifier", even though the delimiter
131in question might not really be a slash.  The modifiers C</imsxadlup>
132may also be embedded within the regular expression itself using
133the C<(?...)> construct, see L</Extended Patterns> below.
134
135=head3 /x
136
137C</x> tells
138the regular expression parser to ignore most whitespace that is neither
139backslashed nor within a bracketed character class.  You can use this to
140break up your regular expression into (slightly) more readable parts.
141Also, the C<#> character is treated as a metacharacter introducing a
142comment that runs up to the pattern's closing delimiter, or to the end
143of the current line if the pattern extends onto the next line.  Hence,
144this is very much like an ordinary Perl code comment.  (You can include
145the closing delimiter within the comment only if you precede it with a
146backslash, so be careful!)
147
148Use of C</x> means that if you want real
149whitespace or C<#> characters in the pattern (outside a bracketed character
150class, which is unaffected by C</x>), then you'll either have to
151escape them (using backslashes or C<\Q...\E>) or encode them using octal,
152hex, or C<\N{}> escapes.
153It is ineffective to try to continue a comment onto the next line by
154escaping the C<\n> with a backslash or C<\Q>.
155
156You can use L</(?#text)> to create a comment that ends earlier than the
157end of the current line, but C<text> also can't contain the closing
158delimiter unless escaped with a backslash.
159
160Taken together, these features go a long way towards
161making Perl's regular expressions more readable.  Here's an example:
162
163    # Delete (most) C comments.
164    $program =~ s {
165	/\*	# Match the opening delimiter.
166	.*?	# Match a minimal number of characters.
167	\*/	# Match the closing delimiter.
168    } []gsx;
169
170Note that anything inside
171a C<\Q...\E> stays unaffected by C</x>.  And note that C</x> doesn't affect
172space interpretation within a single multi-character construct.  For
173example in C<\x{...}>, regardless of the C</x> modifier, there can be no
174spaces.  Same for a L<quantifier|/Quantifiers> such as C<{3}> or
175C<{5,}>.  Similarly, C<(?:...)> can't have a space between the C<(>,
176C<?>, and C<:>.  Within any delimiters for such a
177construct, allowed spaces are not affected by C</x>, and depend on the
178construct.  For example, C<\x{...}> can't have spaces because hexadecimal
179numbers don't have spaces in them.  But, Unicode properties can have spaces, so
180in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
181L<perluniprops/Properties accessible through \p{} and \P{}>.
182X</x>
183
184=head3 Character set modifiers
185
186C</d>, C</u>, C</a>, and C</l>, available starting in 5.14, are called
187the character set modifiers; they affect the character set rules
188used for the regular expression.
189
190The C</d>, C</u>, and C</l> modifiers are not likely to be of much use
191to you, and so you need not worry about them very much.  They exist for
192Perl's internal use, so that complex regular expression data structures
193can be automatically serialized and later exactly reconstituted,
194including all their nuances.  But, since Perl can't keep a secret, and
195there may be rare instances where they are useful, they are documented
196here.
197
198The C</a> modifier, on the other hand, may be useful.  Its purpose is to
199allow code that is to work mostly on ASCII data to not have to concern
200itself with Unicode.
201
202Briefly, C</l> sets the character set to that of whatever B<L>ocale is in
203effect at the time of the execution of the pattern match.
204
205C</u> sets the character set to B<U>nicode.
206
207C</a> also sets the character set to Unicode, BUT adds several
208restrictions for B<A>SCII-safe matching.
209
210C</d> is the old, problematic, pre-5.14 B<D>efault character set
211behavior.  Its only use is to force that old behavior.
212
213At any given time, exactly one of these modifiers is in effect.  Their
214existence allows Perl to keep the originally compiled behavior of a
215regular expression, regardless of what rules are in effect when it is
216actually executed.  And if it is interpolated into a larger regex, the
217original's rules continue to apply to it, and only it.
218
219The C</l> and C</u> modifiers are automatically selected for
220regular expressions compiled within the scope of various pragmas,
221and we recommend that in general, you use those pragmas instead of
222specifying these modifiers explicitly.  For one thing, the modifiers
223affect only pattern matching, and do not extend to even any replacement
224done, whereas using the pragmas give consistent results for all
225appropriate operations within their scopes.  For example,
226
227 s/foo/\Ubar/il
228
229will match "foo" using the locale's rules for case-insensitive matching,
230but the C</l> does not affect how the C<\U> operates.  Most likely you
231want both of them to use locale rules.  To do this, instead compile the
232regular expression within the scope of C<use locale>.  This both
233implicitly adds the C</l> and applies locale rules to the C<\U>.   The
234lesson is to C<use locale> and not C</l> explicitly.
235
236Similarly, it would be better to use C<use feature 'unicode_strings'>
237instead of,
238
239 s/foo/\Lbar/iu
240
241to get Unicode rules, as the C<\L> in the former (but not necessarily
242the latter) would also use Unicode rules.
243
244More detail on each of the modifiers follows.  Most likely you don't
245need to know this detail for C</l>, C</u>, and C</d>, and can skip ahead
246to L<E<sol>a|/E<sol>a (and E<sol>aa)>.
247
248=head4 /l
249
250means to use the current locale's rules (see L<perllocale>) when pattern
251matching.  For example, C<\w> will match the "word" characters of that
252locale, and C<"/i"> case-insensitive matching will match according to
253the locale's case folding rules.  The locale used will be the one in
254effect at the time of execution of the pattern match.  This may not be
255the same as the compilation-time locale, and can differ from one match
256to another if there is an intervening call of the
257L<setlocale() function|perllocale/The setlocale function>.
258
259The only non-single-byte locale Perl supports is (starting in v5.20)
260UTF-8.  This means that code points above 255 are treated as Unicode no
261matter what locale is in effect (since UTF-8 implies Unicode).
262
263Under Unicode rules, there are a few case-insensitive matches that cross
264the 255/256 boundary.  Except for UTF-8 locales in Perls v5.20 and
265later, these are disallowed under C</l>.  For example, 0xFF (on ASCII
266platforms) does not caselessly match the character at 0x178, C<LATIN
267CAPITAL LETTER Y WITH DIAERESIS>, because 0xFF may not be C<LATIN SMALL
268LETTER Y WITH DIAERESIS> in the current locale, and Perl has no way of
269knowing if that character even exists in the locale, much less what code
270point it is.
271
272In a UTF-8 locale in v5.20 and later, the only visible difference
273between locale and non-locale in regular expressions should be tainting
274(see L<perlsec>).
275
276This modifier may be specified to be the default by C<use locale>, but
277see L</Which character set modifier is in effect?>.
278X</l>
279
280=head4 /u
281
282means to use Unicode rules when pattern matching.  On ASCII platforms,
283this means that the code points between 128 and 255 take on their
284Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's).
285(Otherwise Perl considers their meanings to be undefined.)  Thus,
286under this modifier, the ASCII platform effectively becomes a Unicode
287platform; and hence, for example, C<\w> will match any of the more than
288100_000 word characters in Unicode.
289
290Unlike most locales, which are specific to a language and country pair,
291Unicode classifies all the characters that are letters I<somewhere> in
292the world as
293C<\w>.  For example, your locale might not think that C<LATIN SMALL
294LETTER ETH> is a letter (unless you happen to speak Icelandic), but
295Unicode does.  Similarly, all the characters that are decimal digits
296somewhere in the world will match C<\d>; this is hundreds, not 10,
297possible matches.  And some of those digits look like some of the 10
298ASCII digits, but mean a different number, so a human could easily think
299a number is a different quantity than it really is.  For example,
300C<BENGALI DIGIT FOUR> (U+09EA) looks very much like an
301C<ASCII DIGIT EIGHT> (U+0038).  And, C<\d+>, may match strings of digits
302that are a mixture from different writing systems, creating a security
303issue.  L<Unicode::UCD/num()> can be used to sort
304this out.  Or the C</a> modifier can be used to force C<\d> to match
305just the ASCII 0 through 9.
306
307Also, under this modifier, case-insensitive matching works on the full
308set of Unicode
309characters.  The C<KELVIN SIGN>, for example matches the letters "k" and
310"K"; and C<LATIN SMALL LIGATURE FF> matches the sequence "ff", which,
311if you're not prepared, might make it look like a hexadecimal constant,
312presenting another potential security issue.  See
313L<http://unicode.org/reports/tr36> for a detailed discussion of Unicode
314security issues.
315
316This modifier may be specified to be the default by C<use feature
317'unicode_strings>, C<use locale ':not_characters'>, or
318C<L<use 5.012|perlfunc/use VERSION>> (or higher),
319but see L</Which character set modifier is in effect?>.
320X</u>
321
322=head4 /d
323
324This modifier means to use the "Default" native rules of the platform
325except when there is cause to use Unicode rules instead, as follows:
326
327=over 4
328
329=item 1
330
331the target string is encoded in UTF-8; or
332
333=item 2
334
335the pattern is encoded in UTF-8; or
336
337=item 3
338
339the pattern explicitly mentions a code point that is above 255 (say by
340C<\x{100}>); or
341
342=item 4
343
344the pattern uses a Unicode name (C<\N{...}>);  or
345
346=item 5
347
348the pattern uses a Unicode property (C<\p{...}>); or
349
350=item 6
351
352the pattern uses L</C<(?[ ])>>
353
354=back
355
356Another mnemonic for this modifier is "Depends", as the rules actually
357used depend on various things, and as a result you can get unexpected
358results.  See L<perlunicode/The "Unicode Bug">.  The Unicode Bug has
359become rather infamous, leading to yet another (printable) name for this
360modifier, "Dodgy".
361
362Unless the pattern or string are encoded in UTF-8, only ASCII characters
363can match positively.
364
365Here are some examples of how that works on an ASCII platform:
366
367 $str =  "\xDF";      # $str is not in UTF-8 format.
368 $str =~ /^\w/;       # No match, as $str isn't in UTF-8 format.
369 $str .= "\x{0e0b}";  # Now $str is in UTF-8 format.
370 $str =~ /^\w/;       # Match! $str is now in UTF-8 format.
371 chop $str;
372 $str =~ /^\w/;       # Still a match! $str remains in UTF-8 format.
373
374This modifier is automatically selected by default when none of the
375others are, so yet another name for it is "Default".
376
377Because of the unexpected behaviors associated with this modifier, you
378probably should only use it to maintain weird backward compatibilities.
379
380=head4 /a (and /aa)
381
382This modifier stands for ASCII-restrict (or ASCII-safe).  This modifier,
383unlike the others, may be doubled-up to increase its effect.
384
385When it appears singly, it causes the sequences C<\d>, C<\s>, C<\w>, and
386the Posix character classes to match only in the ASCII range.  They thus
387revert to their pre-5.6, pre-Unicode meanings.  Under C</a>,  C<\d>
388always means precisely the digits C<"0"> to C<"9">; C<\s> means the five
389characters C<[ \f\n\r\t]>, and starting in Perl v5.18, experimentally,
390the vertical tab; C<\w> means the 63 characters
391C<[A-Za-z0-9_]>; and likewise, all the Posix classes such as
392C<[[:print:]]> match only the appropriate ASCII-range characters.
393
394This modifier is useful for people who only incidentally use Unicode,
395and who do not wish to be burdened with its complexities and security
396concerns.
397
398With C</a>, one can write C<\d> with confidence that it will only match
399ASCII characters, and should the need arise to match beyond ASCII, you
400can instead use C<\p{Digit}> (or C<\p{Word}> for C<\w>).  There are
401similar C<\p{...}> constructs that can match beyond ASCII both white
402space (see L<perlrecharclass/Whitespace>), and Posix classes (see
403L<perlrecharclass/POSIX Character Classes>).  Thus, this modifier
404doesn't mean you can't use Unicode, it means that to get Unicode
405matching you must explicitly use a construct (C<\p{}>, C<\P{}>) that
406signals Unicode.
407
408As you would expect, this modifier causes, for example, C<\D> to mean
409the same thing as C<[^0-9]>; in fact, all non-ASCII characters match
410C<\D>, C<\S>, and C<\W>.  C<\b> still means to match at the boundary
411between C<\w> and C<\W>, using the C</a> definitions of them (similarly
412for C<\B>).
413
414Otherwise, C</a> behaves like the C</u> modifier, in that
415case-insensitive matching uses Unicode rules; for example, "k" will
416match the Unicode C<\N{KELVIN SIGN}> under C</i> matching, and code
417points in the Latin1 range, above ASCII will have Unicode rules when it
418comes to case-insensitive matching.
419
420To forbid ASCII/non-ASCII matches (like "k" with C<\N{KELVIN SIGN}>),
421specify the "a" twice, for example C</aai> or C</aia>.  (The first
422occurrence of "a" restricts the C<\d>, etc., and the second occurrence
423adds the C</i> restrictions.)  But, note that code points outside the
424ASCII range will use Unicode rules for C</i> matching, so the modifier
425doesn't really restrict things to just ASCII; it just forbids the
426intermixing of ASCII and non-ASCII.
427
428To summarize, this modifier provides protection for applications that
429don't wish to be exposed to all of Unicode.  Specifying it twice
430gives added protection.
431
432This modifier may be specified to be the default by C<use re '/a'>
433or C<use re '/aa'>.  If you do so, you may actually have occasion to use
434the C</u> modifier explicitly if there are a few regular expressions
435where you do want full Unicode rules (but even here, it's best if
436everything were under feature C<"unicode_strings">, along with the
437C<use re '/aa'>).  Also see L</Which character set modifier is in
438effect?>.
439X</a>
440X</aa>
441
442=head4 Which character set modifier is in effect?
443
444Which of these modifiers is in effect at any given point in a regular
445expression depends on a fairly complex set of interactions.  These have
446been designed so that in general you don't have to worry about it, but
447this section gives the gory details.  As
448explained below in L</Extended Patterns> it is possible to explicitly
449specify modifiers that apply only to portions of a regular expression.
450The innermost always has priority over any outer ones, and one applying
451to the whole expression has priority over any of the default settings that are
452described in the remainder of this section.
453
454The C<L<use re 'E<sol>foo'|re/"'/flags' mode">> pragma can be used to set
455default modifiers (including these) for regular expressions compiled
456within its scope.  This pragma has precedence over the other pragmas
457listed below that also change the defaults.
458
459Otherwise, C<L<use locale|perllocale>> sets the default modifier to C</l>;
460and C<L<use feature 'unicode_strings|feature>>, or
461C<L<use 5.012|perlfunc/use VERSION>> (or higher) set the default to
462C</u> when not in the same scope as either C<L<use locale|perllocale>>
463or C<L<use bytes|bytes>>.
464(C<L<use locale ':not_characters'|perllocale/Unicode and UTF-8>> also
465sets the default to C</u>, overriding any plain C<use locale>.)
466Unlike the mechanisms mentioned above, these
467affect operations besides regular expressions pattern matching, and so
468give more consistent results with other operators, including using
469C<\U>, C<\l>, etc. in substitution replacements.
470
471If none of the above apply, for backwards compatibility reasons, the
472C</d> modifier is the one in effect by default.  As this can lead to
473unexpected results, it is best to specify which other rule set should be
474used.
475
476=head4 Character set modifier behavior prior to Perl 5.14
477
478Prior to 5.14, there were no explicit modifiers, but C</l> was implied
479for regexes compiled within the scope of C<use locale>, and C</d> was
480implied otherwise.  However, interpolating a regex into a larger regex
481would ignore the original compilation in favor of whatever was in effect
482at the time of the second compilation.  There were a number of
483inconsistencies (bugs) with the C</d> modifier, where Unicode rules
484would be used when inappropriate, and vice versa.  C<\p{}> did not imply
485Unicode rules, and neither did all occurrences of C<\N{}>, until 5.12.
486
487=head2 Regular Expressions
488
489=head3 Metacharacters
490
491The patterns used in Perl pattern matching evolved from those supplied in
492the Version 8 regex routines.  (The routines are derived
493(distantly) from Henry Spencer's freely redistributable reimplementation
494of the V8 routines.)  See L<Version 8 Regular Expressions> for
495details.
496
497In particular the following metacharacters have their standard I<egrep>-ish
498meanings:
499X<metacharacter>
500X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
501
502
503    \        Quote the next metacharacter
504    ^        Match the beginning of the line
505    .        Match any character (except newline)
506    $        Match the end of the string (or before newline at the end
507             of the string)
508    |        Alternation
509    ()       Grouping
510    []       Bracketed Character class
511
512By default, the "^" character is guaranteed to match only the
513beginning of the string, the "$" character only the end (or before the
514newline at the end), and Perl does certain optimizations with the
515assumption that the string contains only one line.  Embedded newlines
516will not be matched by "^" or "$".  You may, however, wish to treat a
517string as a multi-line buffer, such that the "^" will match after any
518newline within the string (except if the newline is the last character in
519the string), and "$" will match before any newline.  At the
520cost of a little more overhead, you can do this by using the /m modifier
521on the pattern match operator.  (Older programs did this by setting C<$*>,
522but this option was removed in perl 5.10.)
523X<^> X<$> X</m>
524
525To simplify multi-line substitutions, the "." character never matches a
526newline unless you use the C</s> modifier, which in effect tells Perl to pretend
527the string is a single line--even if it isn't.
528X<.> X</s>
529
530=head3 Quantifiers
531
532The following standard quantifiers are recognized:
533X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
534
535    *           Match 0 or more times
536    +           Match 1 or more times
537    ?           Match 1 or 0 times
538    {n}         Match exactly n times
539    {n,}        Match at least n times
540    {n,m}       Match at least n but not more than m times
541
542(If a curly bracket occurs in any other context and does not form part of
543a backslashed sequence like C<\x{...}>, it is treated as a regular
544character.  In particular, the lower quantifier bound is not optional,
545and a typo in a quantifier silently causes it to be treated as the
546literal characters.  For example,
547
548    /o{4,a}/
549
550compiles to match the sequence of six characters
551S<C<"o { 4 , a }">>.  It is planned to eventually require literal uses
552of curly brackets to be escaped, say by preceding them with a backslash
553or enclosing them within square brackets, (C<"\{"> or C<"[{]">).  This
554change will allow for future syntax extensions (like making the lower
555bound of a quantifier optional), and better error checking.  In the
556meantime, you should get in the habit of escaping all instances where
557you mean a literal "{".)
558
559The "*" quantifier is equivalent to C<{0,}>, the "+"
560quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>.  n and m are limited
561to non-negative integral values less than a preset limit defined when perl is built.
562This is usually 32766 on the most common platforms.  The actual limit can
563be seen in the error message generated by code such as this:
564
565    $_ **= $_ , / {$_} / for 2 .. 42;
566
567By default, a quantified subpattern is "greedy", that is, it will match as
568many times as possible (given a particular starting location) while still
569allowing the rest of the pattern to match.  If you want it to match the
570minimum number of times possible, follow the quantifier with a "?".  Note
571that the meanings don't change, just the "greediness":
572X<metacharacter> X<greedy> X<greediness>
573X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
574
575    *?        Match 0 or more times, not greedily
576    +?        Match 1 or more times, not greedily
577    ??        Match 0 or 1 time, not greedily
578    {n}?      Match exactly n times, not greedily (redundant)
579    {n,}?     Match at least n times, not greedily
580    {n,m}?    Match at least n but not more than m times, not greedily
581
582Normally when a quantified subpattern does not allow the rest of the
583overall pattern to match, Perl will backtrack. However, this behaviour is
584sometimes undesirable. Thus Perl provides the "possessive" quantifier form
585as well.
586
587 *+     Match 0 or more times and give nothing back
588 ++     Match 1 or more times and give nothing back
589 ?+     Match 0 or 1 time and give nothing back
590 {n}+   Match exactly n times and give nothing back (redundant)
591 {n,}+  Match at least n times and give nothing back
592 {n,m}+ Match at least n but not more than m times and give nothing back
593
594For instance,
595
596   'aaaa' =~ /a++a/
597
598will never match, as the C<a++> will gobble up all the C<a>'s in the
599string and won't leave any for the remaining part of the pattern. This
600feature can be extremely useful to give perl hints about where it
601shouldn't backtrack. For instance, the typical "match a double-quoted
602string" problem can be most efficiently performed when written as:
603
604   /"(?:[^"\\]++|\\.)*+"/
605
606as we know that if the final quote does not match, backtracking will not
607help. See the independent subexpression
608L</C<< (?>pattern) >>> for more details;
609possessive quantifiers are just syntactic sugar for that construct. For
610instance the above example could also be written as follows:
611
612   /"(?>(?:(?>[^"\\]+)|\\.)*)"/
613
614Note that the possessive quantifier modifier can not be be combined
615with the non-greedy modifier. This is because it would make no sense.
616Consider the follow equivalency table:
617
618    Illegal         Legal
619    ------------    ------
620    X??+            X{0}
621    X+?+            X{1}
622    X{min,max}?+    X{min}
623
624=head3 Escape sequences
625
626Because patterns are processed as double-quoted strings, the following
627also work:
628
629 \t          tab                   (HT, TAB)
630 \n          newline               (LF, NL)
631 \r          return                (CR)
632 \f          form feed             (FF)
633 \a          alarm (bell)          (BEL)
634 \e          escape (think troff)  (ESC)
635 \cK         control char          (example: VT)
636 \x{}, \x00  character whose ordinal is the given hexadecimal number
637 \N{name}    named Unicode character or character sequence
638 \N{U+263D}  Unicode character     (example: FIRST QUARTER MOON)
639 \o{}, \000  character whose ordinal is the given octal number
640 \l          lowercase next char (think vi)
641 \u          uppercase next char (think vi)
642 \L          lowercase until \E (think vi)
643 \U          uppercase until \E (think vi)
644 \Q          quote (disable) pattern metacharacters until \E
645 \E          end either case modification or quoted section, think vi
646
647Details are in L<perlop/Quote and Quote-like Operators>.
648
649=head3 Character Classes and other Special Escapes
650
651In addition, Perl defines the following:
652X<\g> X<\k> X<\K> X<backreference>
653
654 Sequence   Note    Description
655  [...]     [1]  Match a character according to the rules of the
656                   bracketed character class defined by the "...".
657                   Example: [a-z] matches "a" or "b" or "c" ... or "z"
658  [[:...:]] [2]  Match a character according to the rules of the POSIX
659                   character class "..." within the outer bracketed
660                   character class.  Example: [[:upper:]] matches any
661                   uppercase character.
662  (?[...])  [8]  Extended bracketed character class
663  \w        [3]  Match a "word" character (alphanumeric plus "_", plus
664                   other connector punctuation chars plus Unicode
665                   marks)
666  \W        [3]  Match a non-"word" character
667  \s        [3]  Match a whitespace character
668  \S        [3]  Match a non-whitespace character
669  \d        [3]  Match a decimal digit character
670  \D        [3]  Match a non-digit character
671  \pP       [3]  Match P, named property.  Use \p{Prop} for longer names
672  \PP       [3]  Match non-P
673  \X        [4]  Match Unicode "eXtended grapheme cluster"
674  \C             Match a single C-language char (octet) even if that is
675                   part of a larger UTF-8 character.  Thus it breaks up
676                   characters into their UTF-8 bytes, so you may end up
677                   with malformed pieces of UTF-8.  Unsupported in
678                   lookbehind. (Deprecated.)
679  \1        [5]  Backreference to a specific capture group or buffer.
680                   '1' may actually be any positive integer.
681  \g1       [5]  Backreference to a specific or previous group,
682  \g{-1}    [5]  The number may be negative indicating a relative
683                   previous group and may optionally be wrapped in
684                   curly brackets for safer parsing.
685  \g{name}  [5]  Named backreference
686  \k<name>  [5]  Named backreference
687  \K        [6]  Keep the stuff left of the \K, don't include it in $&
688  \N        [7]  Any character but \n.  Not affected by /s modifier
689  \v        [3]  Vertical whitespace
690  \V        [3]  Not vertical whitespace
691  \h        [3]  Horizontal whitespace
692  \H        [3]  Not horizontal whitespace
693  \R        [4]  Linebreak
694
695=over 4
696
697=item [1]
698
699See L<perlrecharclass/Bracketed Character Classes> for details.
700
701=item [2]
702
703See L<perlrecharclass/POSIX Character Classes> for details.
704
705=item [3]
706
707See L<perlrecharclass/Backslash sequences> for details.
708
709=item [4]
710
711See L<perlrebackslash/Misc> for details.
712
713=item [5]
714
715See L</Capture groups> below for details.
716
717=item [6]
718
719See L</Extended Patterns> below for details.
720
721=item [7]
722
723Note that C<\N> has two meanings.  When of the form C<\N{NAME}>, it matches the
724character or character sequence whose name is C<NAME>; and similarly
725when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
726code point is I<hex>.  Otherwise it matches any character but C<\n>.
727
728=item [8]
729
730See L<perlrecharclass/Extended Bracketed Character Classes> for details.
731
732=back
733
734=head3 Assertions
735
736Perl defines the following zero-width assertions:
737X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
738X<regexp, zero-width assertion>
739X<regular expression, zero-width assertion>
740X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
741
742    \b  Match a word boundary
743    \B  Match except at a word boundary
744    \A  Match only at beginning of string
745    \Z  Match only at end of string, or before newline at the end
746    \z  Match only at end of string
747    \G  Match only at pos() (e.g. at the end-of-match position
748        of prior m//g)
749
750A word boundary (C<\b>) is a spot between two characters
751that has a C<\w> on one side of it and a C<\W> on the other side
752of it (in either order), counting the imaginary characters off the
753beginning and end of the string as matching a C<\W>.  (Within
754character classes C<\b> represents backspace rather than a word
755boundary, just as it normally does in any double-quoted string.)
756The C<\A> and C<\Z> are just like "^" and "$", except that they
757won't match multiple times when the C</m> modifier is used, while
758"^" and "$" will match at every internal line boundary.  To match
759the actual end of the string and not ignore an optional trailing
760newline, use C<\z>.
761X<\b> X<\A> X<\Z> X<\z> X</m>
762
763The C<\G> assertion can be used to chain global matches (using
764C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
765It is also useful when writing C<lex>-like scanners, when you have
766several patterns that you want to match against consequent substrings
767of your string; see the previous reference.  The actual location
768where C<\G> will match can also be influenced by using C<pos()> as
769an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
770matches (see L</"Repeated Patterns Matching a Zero-length Substring">)
771is modified somewhat, in that contents to the left of C<\G> are
772not counted when determining the length of the match. Thus the following
773will not match forever:
774X<\G>
775
776     my $string = 'ABC';
777     pos($string) = 1;
778     while ($string =~ /(.\G)/g) {
779         print $1;
780     }
781
782It will print 'A' and then terminate, as it considers the match to
783be zero-width, and thus will not match at the same position twice in a
784row.
785
786It is worth noting that C<\G> improperly used can result in an infinite
787loop. Take care when using patterns that include C<\G> in an alternation.
788
789Note also that C<s///> will refuse to overwrite part of a substitution
790that has already been replaced; so for example this will stop after the
791first iteration, rather than iterating its way backwards through the
792string:
793
794    $_ = "123456789";
795    pos = 6;
796    s/.(?=.\G)/X/g;
797    print; 	# prints 1234X6789, not XXXXX6789
798
799
800=head3 Capture groups
801
802The bracketing construct C<( ... )> creates capture groups (also referred to as
803capture buffers). To refer to the current contents of a group later on, within
804the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
805for the second, and so on.
806This is called a I<backreference>.
807X<regex, capture buffer> X<regexp, capture buffer>
808X<regex, capture group> X<regexp, capture group>
809X<regular expression, capture buffer> X<backreference>
810X<regular expression, capture group> X<backreference>
811X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
812X<named capture buffer> X<regular expression, named capture buffer>
813X<named capture group> X<regular expression, named capture group>
814X<%+> X<$+{name}> X<< \k<name> >>
815There is no limit to the number of captured substrings that you may use.
816Groups are numbered with the leftmost open parenthesis being number 1, etc.  If
817a group did not match, the associated backreference won't match either. (This
818can happen if the group is optional, or in a different branch of an
819alternation.)
820You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
821this form, described below.
822
823You can also refer to capture groups relatively, by using a negative number, so
824that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
825group, and C<\g-2> and C<\g{-2}> both refer to the group before it.  For
826example:
827
828        /
829         (Y)            # group 1
830         (              # group 2
831            (X)         # group 3
832            \g{-1}      # backref to group 3
833            \g{-3}      # backref to group 1
834         )
835        /x
836
837would match the same as C</(Y) ( (X) \g3 \g1 )/x>.  This allows you to
838interpolate regexes into larger regexes and not have to worry about the
839capture groups being renumbered.
840
841You can dispense with numbers altogether and create named capture groups.
842The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
843reference.  (To be compatible with .Net regular expressions, C<\g{I<name>}> may
844also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
845I<name> must not begin with a number, nor contain hyphens.
846When different groups within the same pattern have the same name, any reference
847to that name assumes the leftmost defined group.  Named groups count in
848absolute and relative numbering, and so can also be referred to by those
849numbers.
850(It's possible to do things with named capture groups that would otherwise
851require C<(??{})>.)
852
853Capture group contents are dynamically scoped and available to you outside the
854pattern until the end of the enclosing block or until the next successful
855match, whichever comes first.  (See L<perlsyn/"Compound Statements">.)
856You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
857etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
858
859Braces are required in referring to named capture groups, but are optional for
860absolute or relative numbered ones.  Braces are safer when creating a regex by
861concatenating smaller strings.  For example if you have C<qr/$a$b/>, and C<$a>
862contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
863is probably not what you intended.
864
865The C<\g> and C<\k> notations were introduced in Perl 5.10.0.  Prior to that
866there were no named nor relative numbered capture groups.  Absolute numbered
867groups were referred to using C<\1>,
868C<\2>, etc., and this notation is still
869accepted (and likely always will be).  But it leads to some ambiguities if
870there are more than 9 capture groups, as C<\10> could mean either the tenth
871capture group, or the character whose ordinal in octal is 010 (a backspace in
872ASCII).  Perl resolves this ambiguity by interpreting C<\10> as a backreference
873only if at least 10 left parentheses have opened before it.  Likewise C<\11> is
874a backreference only if at least 11 left parentheses have opened before it.
875And so on.  C<\1> through C<\9> are always interpreted as backreferences.
876There are several examples below that illustrate these perils.  You can avoid
877the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
878and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
879digits padded with leading zeros, since a leading zero implies an octal
880constant.
881
882The C<\I<digit>> notation also works in certain circumstances outside
883the pattern.  See L</Warning on \1 Instead of $1> below for details.
884
885Examples:
886
887    s/^([^ ]*) *([^ ]*)/$2 $1/;     # swap first two words
888
889    /(.)\g1/                        # find first doubled char
890         and print "'$1' is the first doubled character\n";
891
892    /(?<char>.)\k<char>/            # ... a different way
893         and print "'$+{char}' is the first doubled character\n";
894
895    /(?'char'.)\g1/                 # ... mix and match
896         and print "'$1' is the first doubled character\n";
897
898    if (/Time: (..):(..):(..)/) {   # parse out values
899        $hours = $1;
900        $minutes = $2;
901        $seconds = $3;
902    }
903
904    /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/   # \g10 is a backreference
905    /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/    # \10 is octal
906    /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/  # \10 is a backreference
907    /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
908
909    $a = '(.)\1';        # Creates problems when concatenated.
910    $b = '(.)\g{1}';     # Avoids the problems.
911    "aa" =~ /${a}/;      # True
912    "aa" =~ /${b}/;      # True
913    "aa0" =~ /${a}0/;    # False!
914    "aa0" =~ /${b}0/;    # True
915    "aa\x08" =~ /${a}0/;  # True!
916    "aa\x08" =~ /${b}0/;  # False
917
918Several special variables also refer back to portions of the previous
919match.  C<$+> returns whatever the last bracket match matched.
920C<$&> returns the entire matched string.  (At one point C<$0> did
921also, but now it returns the name of the program.)  C<$`> returns
922everything before the matched string.  C<$'> returns everything
923after the matched string. And C<$^N> contains whatever was matched by
924the most-recently closed group (submatch). C<$^N> can be used in
925extended patterns (see below), for example to assign a submatch to a
926variable.
927X<$+> X<$^N> X<$&> X<$`> X<$'>
928
929These special variables, like the C<%+> hash and the numbered match variables
930(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
931until the end of the enclosing block or until the next successful
932match, whichever comes first.  (See L<perlsyn/"Compound Statements">.)
933X<$+> X<$^N> X<$&> X<$`> X<$'>
934X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
935
936B<NOTE>: Failed matches in Perl do not reset the match variables,
937which makes it easier to write code that tests for a series of more
938specific cases and remembers the best match.
939
940B<WARNING>: If your code is to run on Perl 5.16 or earlier,
941beware that once Perl sees that you need one of C<$&>, C<$`>, or
942C<$'> anywhere in the program, it has to provide them for every
943pattern match.  This may substantially slow your program.
944
945Perl uses the same mechanism to produce C<$1>, C<$2>, etc, so you also
946pay a price for each pattern that contains capturing parentheses.
947(To avoid this cost while retaining the grouping behaviour, use the
948extended regular expression C<(?: ... )> instead.)  But if you never
949use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
950parentheses will not be penalized.  So avoid C<$&>, C<$'>, and C<$`>
951if you can, but if you can't (and some algorithms really appreciate
952them), once you've used them once, use them at will, because you've
953already paid the price.
954X<$&> X<$`> X<$'>
955
956Perl 5.16 introduced a slightly more efficient mechanism that notes
957separately whether each of C<$`>, C<$&>, and C<$'> have been seen, and
958thus may only need to copy part of the string.  Perl 5.20 introduced a
959much more efficient copy-on-write mechanism which eliminates any slowdown.
960
961As another workaround for this problem, Perl 5.10.0 introduced C<${^PREMATCH}>,
962C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
963and C<$'>, B<except> that they are only guaranteed to be defined after a
964successful match that was executed with the C</p> (preserve) modifier.
965The use of these variables incurs no global performance penalty, unlike
966their punctuation char equivalents, however at the trade-off that you
967have to tell perl when you want to use them.  As of Perl 5.20, these three
968variables are equivalent to C<$`>, C<$&> and C<$'>, and C</p> is ignored.
969X</p> X<p modifier>
970
971=head2 Quoting metacharacters
972
973Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
974C<\w>, C<\n>.  Unlike some other regular expression languages, there
975are no backslashed symbols that aren't alphanumeric.  So anything
976that looks like \\, \(, \), \[, \], \{, or \} is always
977interpreted as a literal character, not a metacharacter.  This was
978once used in a common idiom to disable or quote the special meanings
979of regular expression metacharacters in a string that you want to
980use for a pattern. Simply quote all non-"word" characters:
981
982    $pattern =~ s/(\W)/\\$1/g;
983
984(If C<use locale> is set, then this depends on the current locale.)
985Today it is more common to use the quotemeta() function or the C<\Q>
986metaquoting escape sequence to disable all metacharacters' special
987meanings like this:
988
989    /$unquoted\Q$quoted\E$unquoted/
990
991Beware that if you put literal backslashes (those not inside
992interpolated variables) between C<\Q> and C<\E>, double-quotish
993backslash interpolation may lead to confusing results.  If you
994I<need> to use literal backslashes within C<\Q...\E>,
995consult L<perlop/"Gory details of parsing quoted constructs">.
996
997C<quotemeta()> and C<\Q> are fully described in L<perlfunc/quotemeta>.
998
999=head2 Extended Patterns
1000
1001Perl also defines a consistent extension syntax for features not
1002found in standard tools like B<awk> and
1003B<lex>.  The syntax for most of these is a
1004pair of parentheses with a question mark as the first thing within
1005the parentheses.  The character after the question mark indicates
1006the extension.
1007
1008The stability of these extensions varies widely.  Some have been
1009part of the core language for many years.  Others are experimental
1010and may change without warning or be completely removed.  Check
1011the documentation on an individual feature to verify its current
1012status.
1013
1014A question mark was chosen for this and for the minimal-matching
1015construct because 1) question marks are rare in older regular
1016expressions, and 2) whenever you see one, you should stop and
1017"question" exactly what is going on.  That's psychology....
1018
1019=over 4
1020
1021=item C<(?#text)>
1022X<(?#)>
1023
1024A comment.  The text is ignored.
1025Note that Perl closes
1026the comment as soon as it sees a C<)>, so there is no way to put a literal
1027C<)> in the comment.  The pattern's closing delimiter must be escaped by
1028a backslash if it appears in the comment.
1029
1030See L</E<sol>x> for another way to have comments in patterns.
1031
1032=item C<(?adlupimsx-imsx)>
1033
1034=item C<(?^alupimsx)>
1035X<(?)> X<(?^)>
1036
1037One or more embedded pattern-match modifiers, to be turned on (or
1038turned off, if preceded by C<->) for the remainder of the pattern or
1039the remainder of the enclosing pattern group (if any).
1040
1041This is particularly useful for dynamic patterns, such as those read in from a
1042configuration file, taken from an argument, or specified in a table
1043somewhere.  Consider the case where some patterns want to be
1044case-sensitive and some do not:  The case-insensitive ones merely need to
1045include C<(?i)> at the front of the pattern.  For example:
1046
1047    $pattern = "foobar";
1048    if ( /$pattern/i ) { }
1049
1050    # more flexible:
1051
1052    $pattern = "(?i)foobar";
1053    if ( /$pattern/ ) { }
1054
1055These modifiers are restored at the end of the enclosing group. For example,
1056
1057    ( (?i) blah ) \s+ \g1
1058
1059will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
1060repetition of the previous word, assuming the C</x> modifier, and no C</i>
1061modifier outside this group.
1062
1063These modifiers do not carry over into named subpatterns called in the
1064enclosing group. In other words, a pattern such as C<((?i)(?&NAME))> does not
1065change the case-sensitivity of the "NAME" pattern.
1066
1067Any of these modifiers can be set to apply globally to all regular
1068expressions compiled within the scope of a C<use re>.  See
1069L<re/"'/flags' mode">.
1070
1071Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1072after the C<"?"> is a shorthand equivalent to C<d-imsx>.  Flags (except
1073C<"d">) may follow the caret to override it.
1074But a minus sign is not legal with it.
1075
1076Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
1077that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
1078C<u> modifiers are mutually exclusive: specifying one de-specifies the
1079others, and a maximum of one (or two C<a>'s) may appear in the
1080construct.  Thus, for
1081example, C<(?-p)> will warn when compiled under C<use warnings>;
1082C<(?-d:...)> and C<(?dl:...)> are fatal errors.
1083
1084Note also that the C<p> modifier is special in that its presence
1085anywhere in a pattern has a global effect.
1086
1087=item C<(?:pattern)>
1088X<(?:)>
1089
1090=item C<(?adluimsx-imsx:pattern)>
1091
1092=item C<(?^aluimsx:pattern)>
1093X<(?^:)>
1094
1095This is for clustering, not capturing; it groups subexpressions like
1096"()", but doesn't make backreferences as "()" does.  So
1097
1098    @fields = split(/\b(?:a|b|c)\b/)
1099
1100is like
1101
1102    @fields = split(/\b(a|b|c)\b/)
1103
1104but doesn't spit out extra fields.  It's also cheaper not to capture
1105characters if you don't need to.
1106
1107Any letters between C<?> and C<:> act as flags modifiers as with
1108C<(?adluimsx-imsx)>.  For example,
1109
1110    /(?s-i:more.*than).*million/i
1111
1112is equivalent to the more verbose
1113
1114    /(?:(?s-i)more.*than).*million/i
1115
1116Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
1117after the C<"?"> is a shorthand equivalent to C<d-imsx>.  Any positive
1118flags (except C<"d">) may follow the caret, so
1119
1120    (?^x:foo)
1121
1122is equivalent to
1123
1124    (?x-ims:foo)
1125
1126The caret tells Perl that this cluster doesn't inherit the flags of any
1127surrounding pattern, but uses the system defaults (C<d-imsx>),
1128modified by any flags specified.
1129
1130The caret allows for simpler stringification of compiled regular
1131expressions.  These look like
1132
1133    (?^:pattern)
1134
1135with any non-default flags appearing between the caret and the colon.
1136A test that looks at such stringification thus doesn't need to have the
1137system default flags hard-coded in it, just the caret.  If new flags are
1138added to Perl, the meaning of the caret's expansion will change to include
1139the default for those flags, so the test will still work, unchanged.
1140
1141Specifying a negative flag after the caret is an error, as the flag is
1142redundant.
1143
1144Mnemonic for C<(?^...)>:  A fresh beginning since the usual use of a caret is
1145to match at the beginning.
1146
1147=item C<(?|pattern)>
1148X<(?|)> X<Branch reset>
1149
1150This is the "branch reset" pattern, which has the special property
1151that the capture groups are numbered from the same starting point
1152in each alternation branch. It is available starting from perl 5.10.0.
1153
1154Capture groups are numbered from left to right, but inside this
1155construct the numbering is restarted for each branch.
1156
1157The numbering within each branch will be as normal, and any groups
1158following this construct will be numbered as though the construct
1159contained only one branch, that being the one with the most capture
1160groups in it.
1161
1162This construct is useful when you want to capture one of a
1163number of alternative matches.
1164
1165Consider the following pattern.  The numbers underneath show in
1166which group the captured content will be stored.
1167
1168
1169    # before  ---------------branch-reset----------- after
1170    / ( a )  (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1171    # 1            2         2  3        2     3     4
1172
1173Be careful when using the branch reset pattern in combination with
1174named captures. Named captures are implemented as being aliases to
1175numbered groups holding the captures, and that interferes with the
1176implementation of the branch reset pattern. If you are using named
1177captures in a branch reset pattern, it's best to use the same names,
1178in the same order, in each of the alternations:
1179
1180   /(?|  (?<a> x ) (?<b> y )
1181      |  (?<a> z ) (?<b> w )) /x
1182
1183Not doing so may lead to surprises:
1184
1185  "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
1186  say $+ {a};   # Prints '12'
1187  say $+ {b};   # *Also* prints '12'.
1188
1189The problem here is that both the group named C<< a >> and the group
1190named C<< b >> are aliases for the group belonging to C<< $1 >>.
1191
1192=item Look-Around Assertions
1193X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
1194
1195Look-around assertions are zero-width patterns which match a specific
1196pattern without including it in C<$&>. Positive assertions match when
1197their subpattern matches, negative assertions match when their subpattern
1198fails. Look-behind matches text up to the current match position,
1199look-ahead matches text following the current match position.
1200
1201=over 4
1202
1203=item C<(?=pattern)>
1204X<(?=)> X<look-ahead, positive> X<lookahead, positive>
1205
1206A zero-width positive look-ahead assertion.  For example, C</\w+(?=\t)/>
1207matches a word followed by a tab, without including the tab in C<$&>.
1208
1209=item C<(?!pattern)>
1210X<(?!)> X<look-ahead, negative> X<lookahead, negative>
1211
1212A zero-width negative look-ahead assertion.  For example C</foo(?!bar)/>
1213matches any occurrence of "foo" that isn't followed by "bar".  Note
1214however that look-ahead and look-behind are NOT the same thing.  You cannot
1215use this for look-behind.
1216
1217If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
1218will not do what you want.  That's because the C<(?!foo)> is just saying that
1219the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
1220match.  Use look-behind instead (see below).
1221
1222=item C<(?<=pattern)> C<\K>
1223X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
1224
1225A zero-width positive look-behind assertion.  For example, C</(?<=\t)\w+/>
1226matches a word that follows a tab, without including the tab in C<$&>.
1227Works only for fixed-width look-behind.
1228
1229There is a special form of this construct, called C<\K> (available since
1230Perl 5.10.0), which causes the
1231regex engine to "keep" everything it had matched prior to the C<\K> and
1232not include it in C<$&>. This effectively provides variable-length
1233look-behind. The use of C<\K> inside of another look-around assertion
1234is allowed, but the behaviour is currently not well defined.
1235
1236For various reasons C<\K> may be significantly more efficient than the
1237equivalent C<< (?<=...) >> construct, and it is especially useful in
1238situations where you want to efficiently remove something following
1239something else in a string. For instance
1240
1241  s/(foo)bar/$1/g;
1242
1243can be rewritten as the much more efficient
1244
1245  s/foo\Kbar//g;
1246
1247=item C<(?<!pattern)>
1248X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
1249
1250A zero-width negative look-behind assertion.  For example C</(?<!bar)foo/>
1251matches any occurrence of "foo" that does not follow "bar".  Works
1252only for fixed-width look-behind.
1253
1254=back
1255
1256=item C<(?'NAME'pattern)>
1257
1258=item C<< (?<NAME>pattern) >>
1259X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
1260
1261A named capture group. Identical in every respect to normal capturing
1262parentheses C<()> but for the additional fact that the group
1263can be referred to by name in various regular expression
1264constructs (like C<\g{NAME}>) and can be accessed by name
1265after a successful match via C<%+> or C<%->. See L<perlvar>
1266for more details on the C<%+> and C<%-> hashes.
1267
1268If multiple distinct capture groups have the same name then the
1269$+{NAME} will refer to the leftmost defined group in the match.
1270
1271The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
1272
1273B<NOTE:> While the notation of this construct is the same as the similar
1274function in .NET regexes, the behavior is not. In Perl the groups are
1275numbered sequentially regardless of being named or not. Thus in the
1276pattern
1277
1278  /(x)(?<foo>y)(z)/
1279
1280$+{foo} will be the same as $2, and $3 will contain 'z' instead of
1281the opposite which is what a .NET regex hacker might expect.
1282
1283Currently NAME is restricted to simple identifiers only.
1284In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
1285its Unicode extension (see L<utf8>),
1286though it isn't extended by the locale (see L<perllocale>).
1287
1288B<NOTE:> In order to make things easier for programmers with experience
1289with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
1290may be used instead of C<< (?<NAME>pattern) >>; however this form does not
1291support the use of single quotes as a delimiter for the name.
1292
1293=item C<< \k<NAME> >>
1294
1295=item C<< \k'NAME' >>
1296
1297Named backreference. Similar to numeric backreferences, except that
1298the group is designated by name and not number. If multiple groups
1299have the same name then it refers to the leftmost defined group in
1300the current match.
1301
1302It is an error to refer to a name not defined by a C<< (?<NAME>) >>
1303earlier in the pattern.
1304
1305Both forms are equivalent.
1306
1307B<NOTE:> In order to make things easier for programmers with experience
1308with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
1309may be used instead of C<< \k<NAME> >>.
1310
1311=item C<(?{ code })>
1312X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
1313
1314B<WARNING>: Using this feature safely requires that you understand its
1315limitations.  Code executed that has side effects may not perform identically
1316from version to version due to the effect of future optimisations in the regex
1317engine.  For more information on this, see L</Embedded Code Execution
1318Frequency>.
1319
1320This zero-width assertion executes any embedded Perl code.  It always
1321succeeds, and its return value is set as C<$^R>.
1322
1323In literal patterns, the code is parsed at the same time as the
1324surrounding code. While within the pattern, control is passed temporarily
1325back to the perl parser, until the logically-balancing closing brace is
1326encountered. This is similar to the way that an array index expression in
1327a literal string is handled, for example
1328
1329    "abc$array[ 1 + f('[') + g()]def"
1330
1331In particular, braces do not need to be balanced:
1332
1333    s/abc(?{ f('{'); })/def/
1334
1335Even in a pattern that is interpolated and compiled at run-time, literal
1336code blocks will be compiled once, at perl compile time; the following
1337prints "ABCD":
1338
1339    print "D";
1340    my $qr = qr/(?{ BEGIN { print "A" } })/;
1341    my $foo = "foo";
1342    /$foo$qr(?{ BEGIN { print "B" } })/;
1343    BEGIN { print "C" }
1344
1345In patterns where the text of the code is derived from run-time
1346information rather than appearing literally in a source code /pattern/,
1347the code is compiled at the same time that the pattern is compiled, and
1348for reasons of security, C<use re 'eval'> must be in scope. This is to
1349stop user-supplied patterns containing code snippets from being
1350executable.
1351
1352In situations where you need to enable this with C<use re 'eval'>, you should
1353also have taint checking enabled.  Better yet, use the carefully
1354constrained evaluation within a Safe compartment.  See L<perlsec> for
1355details about both these mechanisms.
1356
1357From the viewpoint of parsing, lexical variable scope and closures,
1358
1359    /AAA(?{ BBB })CCC/
1360
1361behaves approximately like
1362
1363    /AAA/ && do { BBB } && /CCC/
1364
1365Similarly,
1366
1367    qr/AAA(?{ BBB })CCC/
1368
1369behaves approximately like
1370
1371    sub { /AAA/ && do { BBB } && /CCC/ }
1372
1373In particular:
1374
1375    { my $i = 1; $r = qr/(?{ print $i })/ }
1376    my $i = 2;
1377    /$r/; # prints "1"
1378
1379Inside a C<(?{...})> block, C<$_> refers to the string the regular
1380expression is matching against. You can also use C<pos()> to know what is
1381the current position of matching within this string.
1382
1383The code block introduces a new scope from the perspective of lexical
1384variable declarations, but B<not> from the perspective of C<local> and
1385similar localizing behaviours. So later code blocks within the same
1386pattern will still see the values which were localized in earlier blocks.
1387These accumulated localizations are undone either at the end of a
1388successful match, or if the assertion is backtracked (compare
1389L<"Backtracking">). For example,
1390
1391  $_ = 'a' x 8;
1392  m<
1393     (?{ $cnt = 0 })               # Initialize $cnt.
1394     (
1395       a
1396       (?{
1397           local $cnt = $cnt + 1;  # Update $cnt,
1398                                   # backtracking-safe.
1399       })
1400     )*
1401     aaaa
1402     (?{ $res = $cnt })            # On success copy to
1403                                   # non-localized location.
1404   >x;
1405
1406will initially increment C<$cnt> up to 8; then during backtracking, its
1407value will be unwound back to 4, which is the value assigned to C<$res>.
1408At the end of the regex execution, $cnt will be wound back to its initial
1409value of 0.
1410
1411This assertion may be used as the condition in a
1412
1413    (?(condition)yes-pattern|no-pattern)
1414
1415switch.  If I<not> used in this way, the result of evaluation of C<code>
1416is put into the special variable C<$^R>.  This happens immediately, so
1417C<$^R> can be used from other C<(?{ code })> assertions inside the same
1418regular expression.
1419
1420The assignment to C<$^R> above is properly localized, so the old
1421value of C<$^R> is restored if the assertion is backtracked; compare
1422L<"Backtracking">.
1423
1424Note that the special variable C<$^N>  is particularly useful with code
1425blocks to capture the results of submatches in variables without having to
1426keep track of the number of nested parentheses. For example:
1427
1428  $_ = "The brown fox jumps over the lazy dog";
1429  /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
1430  print "color = $color, animal = $animal\n";
1431
1432
1433=item C<(??{ code })>
1434X<(??{})>
1435X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
1436
1437B<WARNING>: Using this feature safely requires that you understand its
1438limitations.  Code executed that has side effects may not perform
1439identically from version to version due to the effect of future
1440optimisations in the regex engine.  For more information on this, see
1441L</Embedded Code Execution Frequency>.
1442
1443This is a "postponed" regular subexpression.  It behaves in I<exactly> the
1444same way as a C<(?{ code })> code block as described above, except that
1445its return value, rather than being assigned to C<$^R>, is treated as a
1446pattern, compiled if it's a string (or used as-is if its a qr// object),
1447then matched as if it were inserted instead of this construct.
1448
1449During the matching of this sub-pattern, it has its own set of
1450captures which are valid during the sub-match, but are discarded once
1451control returns to the main pattern. For example, the following matches,
1452with the inner pattern capturing "B" and matching "BB", while the outer
1453pattern captures "A";
1454
1455    my $inner = '(.)\1';
1456    "ABBA" =~ /^(.)(??{ $inner })\1/;
1457    print $1; # prints "A";
1458
1459Note that this means that  there is no way for the inner pattern to refer
1460to a capture group defined outside.  (The code block itself can use C<$1>,
1461etc., to refer to the enclosing pattern's capture groups.)  Thus, although
1462
1463    ('a' x 100)=~/(??{'(.)' x 100})/
1464
1465I<will> match, it will I<not> set $1 on exit.
1466
1467The following pattern matches a parenthesized group:
1468
1469 $re = qr{
1470            \(
1471            (?:
1472               (?> [^()]+ )  # Non-parens without backtracking
1473             |
1474               (??{ $re })   # Group with matching parens
1475            )*
1476            \)
1477         }x;
1478
1479See also
1480L<C<(?I<PARNO>)>|/(?PARNO) (?-PARNO) (?+PARNO) (?R) (?0)>
1481for a different, more efficient way to accomplish
1482the same task.
1483
1484Executing a postponed regular expression 50 times without consuming any
1485input string will result in a fatal error.  The maximum depth is compiled
1486into perl, so changing it requires a custom build.
1487
1488=item C<(?I<PARNO>)> C<(?-I<PARNO>)> C<(?+I<PARNO>)> C<(?R)> C<(?0)>
1489X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
1490X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
1491X<regex, relative recursion> X<GOSUB> X<GOSTART>
1492
1493Recursive subpattern. Treat the contents of a given capture buffer in the
1494current pattern as an independent subpattern and attempt to match it at
1495the current position in the string. Information about capture state from
1496the caller for things like backreferences is available to the subpattern,
1497but capture buffers set by the subpattern are not visible to the caller.
1498
1499Similar to C<(??{ code })> except that it does not involve executing any
1500code or potentially compiling a returned pattern string; instead it treats
1501the part of the current pattern contained within a specified capture group
1502as an independent pattern that must match at the current position. Also
1503different is the treatment of capture buffers, unlike C<(??{ code })>
1504recursive patterns have access to their callers match state, so one can
1505use backreferences safely.
1506
1507I<PARNO> is a sequence of digits (not starting with 0) whose value reflects
1508the paren-number of the capture group to recurse to. C<(?R)> recurses to
1509the beginning of the whole pattern. C<(?0)> is an alternate syntax for
1510C<(?R)>. If I<PARNO> is preceded by a plus or minus sign then it is assumed
1511to be relative, with negative numbers indicating preceding capture groups
1512and positive ones following. Thus C<(?-1)> refers to the most recently
1513declared group, and C<(?+1)> indicates the next group to be declared.
1514Note that the counting for relative recursion differs from that of
1515relative backreferences, in that with recursion unclosed groups B<are>
1516included.
1517
1518The following pattern matches a function foo() which may contain
1519balanced parentheses as the argument.
1520
1521  $re = qr{ (                   # paren group 1 (full function)
1522              foo
1523              (                 # paren group 2 (parens)
1524                \(
1525                  (             # paren group 3 (contents of parens)
1526                  (?:
1527                   (?> [^()]+ ) # Non-parens without backtracking
1528                  |
1529                   (?2)         # Recurse to start of paren group 2
1530                  )*
1531                  )
1532                \)
1533              )
1534            )
1535          }x;
1536
1537If the pattern was used as follows
1538
1539    'foo(bar(baz)+baz(bop))'=~/$re/
1540        and print "\$1 = $1\n",
1541                  "\$2 = $2\n",
1542                  "\$3 = $3\n";
1543
1544the output produced should be the following:
1545
1546    $1 = foo(bar(baz)+baz(bop))
1547    $2 = (bar(baz)+baz(bop))
1548    $3 = bar(baz)+baz(bop)
1549
1550If there is no corresponding capture group defined, then it is a
1551fatal error.  Recursing deeper than 50 times without consuming any input
1552string will also result in a fatal error.  The maximum depth is compiled
1553into perl, so changing it requires a custom build.
1554
1555The following shows how using negative indexing can make it
1556easier to embed recursive patterns inside of a C<qr//> construct
1557for later use:
1558
1559    my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1560    if (/foo $parens \s+ \+ \s+ bar $parens/x) {
1561       # do something here...
1562    }
1563
1564B<Note> that this pattern does not behave the same way as the equivalent
1565PCRE or Python construct of the same form. In Perl you can backtrack into
1566a recursed group, in PCRE and Python the recursed into group is treated
1567as atomic. Also, modifiers are resolved at compile time, so constructs
1568like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1569be processed.
1570
1571=item C<(?&NAME)>
1572X<(?&NAME)>
1573
1574Recurse to a named subpattern. Identical to C<(?I<PARNO>)> except that the
1575parenthesis to recurse to is determined by name. If multiple parentheses have
1576the same name, then it recurses to the leftmost.
1577
1578It is an error to refer to a name that is not declared somewhere in the
1579pattern.
1580
1581B<NOTE:> In order to make things easier for programmers with experience
1582with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
1583may be used instead of C<< (?&NAME) >>.
1584
1585=item C<(?(condition)yes-pattern|no-pattern)>
1586X<(?()>
1587
1588=item C<(?(condition)yes-pattern)>
1589
1590Conditional expression. Matches C<yes-pattern> if C<condition> yields
1591a true value, matches C<no-pattern> otherwise. A missing pattern always
1592matches.
1593
1594C<(condition)> should be one of: 1) an integer in
1595parentheses (which is valid if the corresponding pair of parentheses
1596matched); 2) a look-ahead/look-behind/evaluate zero-width assertion; 3) a
1597name in angle brackets or single quotes (which is valid if a group
1598with the given name matched); or 4) the special symbol (R) (true when
1599evaluated inside of recursion or eval). Additionally the R may be
1600followed by a number, (which will be true when evaluated when recursing
1601inside of the appropriate group), or by C<&NAME>, in which case it will
1602be true only when evaluated during recursion in the named group.
1603
1604Here's a summary of the possible predicates:
1605
1606=over 4
1607
1608=item (1) (2) ...
1609
1610Checks if the numbered capturing group has matched something.
1611
1612=item (<NAME>) ('NAME')
1613
1614Checks if a group with the given name has matched something.
1615
1616=item (?=...) (?!...) (?<=...) (?<!...)
1617
1618Checks whether the pattern matches (or does not match, for the '!'
1619variants).
1620
1621=item (?{ CODE })
1622
1623Treats the return value of the code block as the condition.
1624
1625=item (R)
1626
1627Checks if the expression has been evaluated inside of recursion.
1628
1629=item (R1) (R2) ...
1630
1631Checks if the expression has been evaluated while executing directly
1632inside of the n-th capture group. This check is the regex equivalent of
1633
1634  if ((caller(0))[3] eq 'subname') { ... }
1635
1636In other words, it does not check the full recursion stack.
1637
1638=item (R&NAME)
1639
1640Similar to C<(R1)>, this predicate checks to see if we're executing
1641directly inside of the leftmost group with a given name (this is the same
1642logic used by C<(?&NAME)> to disambiguate). It does not check the full
1643stack, but only the name of the innermost active recursion.
1644
1645=item (DEFINE)
1646
1647In this case, the yes-pattern is never directly executed, and no
1648no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1649See below for details.
1650
1651=back
1652
1653For example:
1654
1655    m{ ( \( )?
1656       [^()]+
1657       (?(1) \) )
1658     }x
1659
1660matches a chunk of non-parentheses, possibly included in parentheses
1661themselves.
1662
1663A special form is the C<(DEFINE)> predicate, which never executes its
1664yes-pattern directly, and does not allow a no-pattern. This allows one to
1665define subpatterns which will be executed only by the recursion mechanism.
1666This way, you can define a set of regular expression rules that can be
1667bundled into any pattern you choose.
1668
1669It is recommended that for this usage you put the DEFINE block at the
1670end of the pattern, and that you name any subpatterns defined within it.
1671
1672Also, it's worth noting that patterns defined this way probably will
1673not be as efficient, as the optimizer is not very clever about
1674handling them.
1675
1676An example of how this might be used is as follows:
1677
1678  /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
1679   (?(DEFINE)
1680     (?<NAME_PAT>....)
1681     (?<ADDRESS_PAT>....)
1682   )/x
1683
1684Note that capture groups matched inside of recursion are not accessible
1685after the recursion returns, so the extra layer of capturing groups is
1686necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1687C<$+{NAME}> would be.
1688
1689Finally, keep in mind that subpatterns created inside a DEFINE block
1690count towards the absolute and relative number of captures, so this:
1691
1692    my @captures = "a" =~ /(.)                  # First capture
1693                           (?(DEFINE)
1694                               (?<EXAMPLE> 1 )  # Second capture
1695                           )/x;
1696    say scalar @captures;
1697
1698Will output 2, not 1. This is particularly important if you intend to
1699compile the definitions with the C<qr//> operator, and later
1700interpolate them in another pattern.
1701
1702=item C<< (?>pattern) >>
1703X<backtrack> X<backtracking> X<atomic> X<possessive>
1704
1705An "independent" subexpression, one which matches the substring
1706that a I<standalone> C<pattern> would match if anchored at the given
1707position, and it matches I<nothing other than this substring>.  This
1708construct is useful for optimizations of what would otherwise be
1709"eternal" matches, because it will not backtrack (see L<"Backtracking">).
1710It may also be useful in places where the "grab all you can, and do not
1711give anything back" semantic is desirable.
1712
1713For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
1714(anchored at the beginning of string, as above) will match I<all>
1715characters C<a> at the beginning of string, leaving no C<a> for
1716C<ab> to match.  In contrast, C<a*ab> will match the same as C<a+b>,
1717since the match of the subgroup C<a*> is influenced by the following
1718group C<ab> (see L<"Backtracking">).  In particular, C<a*> inside
1719C<a*ab> will match fewer characters than a standalone C<a*>, since
1720this makes the tail match.
1721
1722C<< (?>pattern) >> does not disable backtracking altogether once it has
1723matched. It is still possible to backtrack past the construct, but not
1724into it. So C<< ((?>a*)|(?>b*))ar >> will still match "bar".
1725
1726An effect similar to C<< (?>pattern) >> may be achieved by writing
1727C<(?=(pattern))\g{-1}>.  This matches the same substring as a standalone
1728C<a+>, and the following C<\g{-1}> eats the matched string; it therefore
1729makes a zero-length assertion into an analogue of C<< (?>...) >>.
1730(The difference between these two constructs is that the second one
1731uses a capturing group, thus shifting ordinals of backreferences
1732in the rest of a regular expression.)
1733
1734Consider this pattern:
1735
1736    m{ \(
1737          (
1738            [^()]+           # x+
1739          |
1740            \( [^()]* \)
1741          )+
1742       \)
1743     }x
1744
1745That will efficiently match a nonempty group with matching parentheses
1746two levels deep or less.  However, if there is no such group, it
1747will take virtually forever on a long string.  That's because there
1748are so many different ways to split a long string into several
1749substrings.  This is what C<(.+)+> is doing, and C<(.+)+> is similar
1750to a subpattern of the above pattern.  Consider how the pattern
1751above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1752seconds, but that each extra letter doubles this time.  This
1753exponential performance will make it appear that your program has
1754hung.  However, a tiny change to this pattern
1755
1756    m{ \(
1757          (
1758            (?> [^()]+ )        # change x+ above to (?> x+ )
1759          |
1760            \( [^()]* \)
1761          )+
1762       \)
1763     }x
1764
1765which uses C<< (?>...) >> matches exactly when the one above does (verifying
1766this yourself would be a productive exercise), but finishes in a fourth
1767the time when used on a similar string with 1000000 C<a>s.  Be aware,
1768however, that, when this construct is followed by a
1769quantifier, it currently triggers a warning message under
1770the C<use warnings> pragma or B<-w> switch saying it
1771C<"matches null string many times in regex">.
1772
1773On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
1774effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
1775This was only 4 times slower on a string with 1000000 C<a>s.
1776
1777The "grab all you can, and do not give anything back" semantic is desirable
1778in many situations where on the first sight a simple C<()*> looks like
1779the correct solution.  Suppose we parse text with comments being delimited
1780by C<#> followed by some optional (horizontal) whitespace.  Contrary to
1781its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
1782the comment delimiter, because it may "give up" some whitespace if
1783the remainder of the pattern can be made to match that way.  The correct
1784answer is either one of these:
1785
1786    (?>#[ \t]*)
1787    #[ \t]*(?![ \t])
1788
1789For example, to grab non-empty comments into $1, one should use either
1790one of these:
1791
1792    / (?> \# [ \t]* ) (        .+ ) /x;
1793    /     \# [ \t]*   ( [^ \t] .* ) /x;
1794
1795Which one you pick depends on which of these expressions better reflects
1796the above specification of comments.
1797
1798In some literature this construct is called "atomic matching" or
1799"possessive matching".
1800
1801Possessive quantifiers are equivalent to putting the item they are applied
1802to inside of one of these constructs. The following equivalences apply:
1803
1804    Quantifier Form     Bracketing Form
1805    ---------------     ---------------
1806    PAT*+               (?>PAT*)
1807    PAT++               (?>PAT+)
1808    PAT?+               (?>PAT?)
1809    PAT{min,max}+       (?>PAT{min,max})
1810
1811=item C<(?[ ])>
1812
1813See L<perlrecharclass/Extended Bracketed Character Classes>.
1814
1815=back
1816
1817=head2 Special Backtracking Control Verbs
1818
1819These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1820otherwise stated the ARG argument is optional; in some cases, it is
1821forbidden.
1822
1823Any pattern containing a special backtracking verb that allows an argument
1824has the special behaviour that when executed it sets the current package's
1825C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1826rules apply:
1827
1828On failure, the C<$REGERROR> variable will be set to the ARG value of the
1829verb pattern, if the verb was involved in the failure of the match. If the
1830ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1831name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1832none. Also, the C<$REGMARK> variable will be set to FALSE.
1833
1834On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1835the C<$REGMARK> variable will be set to the name of the last
1836C<(*MARK:NAME)> pattern executed.  See the explanation for the
1837C<(*MARK:NAME)> verb below for more details.
1838
1839B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1840and most other regex-related variables. They are not local to a scope, nor
1841readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1842Use C<local> to localize changes to them to a specific scope if necessary.
1843
1844If a pattern does not contain a special backtracking verb that allows an
1845argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
1846
1847=over 3
1848
1849=item Verbs that take an argument
1850
1851=over 4
1852
1853=item C<(*PRUNE)> C<(*PRUNE:NAME)>
1854X<(*PRUNE)> X<(*PRUNE:NAME)>
1855
1856This zero-width pattern prunes the backtracking tree at the current point
1857when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1858where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1859A may backtrack as necessary to match. Once it is reached, matching
1860continues in B, which may also backtrack as necessary; however, should B
1861not match, then no further backtracking will take place, and the pattern
1862will fail outright at the current starting position.
1863
1864The following example counts all the possible matching strings in a
1865pattern (without actually matching any of them).
1866
1867    'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
1868    print "Count=$count\n";
1869
1870which produces:
1871
1872    aaab
1873    aaa
1874    aa
1875    a
1876    aab
1877    aa
1878    a
1879    ab
1880    a
1881    Count=9
1882
1883If we add a C<(*PRUNE)> before the count like the following
1884
1885    'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
1886    print "Count=$count\n";
1887
1888we prevent backtracking and find the count of the longest matching string
1889at each matching starting point like so:
1890
1891    aaab
1892    aab
1893    ab
1894    Count=3
1895
1896Any number of C<(*PRUNE)> assertions may be used in a pattern.
1897
1898See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1899control backtracking. In some cases, the use of C<(*PRUNE)> can be
1900replaced with a C<< (?>pattern) >> with no functional difference; however,
1901C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1902C<< (?>pattern) >> alone.
1903
1904=item C<(*SKIP)> C<(*SKIP:NAME)>
1905X<(*SKIP)>
1906
1907This zero-width pattern is similar to C<(*PRUNE)>, except that on
1908failure it also signifies that whatever text that was matched leading up
1909to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1910of this pattern. This effectively means that the regex engine "skips" forward
1911to this position on failure and tries to match again, (assuming that
1912there is sufficient room to match).
1913
1914The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1915C<(*MARK:NAME)> was encountered while matching, then it is that position
1916which is used as the "skip point". If no C<(*MARK)> of that name was
1917encountered, then the C<(*SKIP)> operator has no effect. When used
1918without a name the "skip point" is where the match point was when
1919executing the (*SKIP) pattern.
1920
1921Compare the following to the examples in C<(*PRUNE)>; note the string
1922is twice as long:
1923
1924 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
1925 print "Count=$count\n";
1926
1927outputs
1928
1929    aaab
1930    aaab
1931    Count=2
1932
1933Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
1934executed, the next starting point will be where the cursor was when the
1935C<(*SKIP)> was executed.
1936
1937=item C<(*MARK:NAME)> C<(*:NAME)>
1938X<(*MARK)> X<(*MARK:NAME)> X<(*:NAME)>
1939
1940This zero-width pattern can be used to mark the point reached in a string
1941when a certain part of the pattern has been successfully matched. This
1942mark may be given a name. A later C<(*SKIP)> pattern will then skip
1943forward to that point if backtracked into on failure. Any number of
1944C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
1945
1946In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1947can be used to "label" a pattern branch, so that after matching, the
1948program can determine which branches of the pattern were involved in the
1949match.
1950
1951When a match is successful, the C<$REGMARK> variable will be set to the
1952name of the most recently executed C<(*MARK:NAME)> that was involved
1953in the match.
1954
1955This can be used to determine which branch of a pattern was matched
1956without using a separate capture group for each branch, which in turn
1957can result in a performance improvement, as perl cannot optimize
1958C</(?:(x)|(y)|(z))/> as efficiently as something like
1959C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1960
1961When a match has failed, and unless another verb has been involved in
1962failing the match and has provided its own name to use, the C<$REGERROR>
1963variable will be set to the name of the most recently executed
1964C<(*MARK:NAME)>.
1965
1966See L</(*SKIP)> for more details.
1967
1968As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1969
1970=item C<(*THEN)> C<(*THEN:NAME)>
1971
1972This is similar to the "cut group" operator C<::> from Perl 6.  Like
1973C<(*PRUNE)>, this verb always matches, and when backtracked into on
1974failure, it causes the regex engine to try the next alternation in the
1975innermost enclosing group (capturing or otherwise) that has alternations.
1976The two branches of a C<(?(condition)yes-pattern|no-pattern)> do not
1977count as an alternation, as far as C<(*THEN)> is concerned.
1978
1979Its name comes from the observation that this operation combined with the
1980alternation operator (C<|>) can be used to create what is essentially a
1981pattern-based if/then/else block:
1982
1983  ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1984
1985Note that if this operator is used and NOT inside of an alternation then
1986it acts exactly like the C<(*PRUNE)> operator.
1987
1988  / A (*PRUNE) B /
1989
1990is the same as
1991
1992  / A (*THEN) B /
1993
1994but
1995
1996  / ( A (*THEN) B | C ) /
1997
1998is not the same as
1999
2000  / ( A (*PRUNE) B | C ) /
2001
2002as after matching the A but failing on the B the C<(*THEN)> verb will
2003backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
2004
2005=back
2006
2007=item Verbs without an argument
2008
2009=over 4
2010
2011=item C<(*COMMIT)>
2012X<(*COMMIT)>
2013
2014This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
2015zero-width pattern similar to C<(*SKIP)>, except that when backtracked
2016into on failure it causes the match to fail outright. No further attempts
2017to find a valid match by advancing the start pointer will occur again.
2018For example,
2019
2020 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
2021 print "Count=$count\n";
2022
2023outputs
2024
2025    aaab
2026    Count=1
2027
2028In other words, once the C<(*COMMIT)> has been entered, and if the pattern
2029does not match, the regex engine will not try any further matching on the
2030rest of the string.
2031
2032=item C<(*FAIL)> C<(*F)>
2033X<(*FAIL)> X<(*F)>
2034
2035This pattern matches nothing and always fails. It can be used to force the
2036engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
2037fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
2038
2039It is probably useful only when combined with C<(?{})> or C<(??{})>.
2040
2041=item C<(*ACCEPT)>
2042X<(*ACCEPT)>
2043
2044This pattern matches nothing and causes the end of successful matching at
2045the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
2046whether there is actually more to match in the string. When inside of a
2047nested pattern, such as recursion, or in a subpattern dynamically generated
2048via C<(??{})>, only the innermost pattern is ended immediately.
2049
2050If the C<(*ACCEPT)> is inside of capturing groups then the groups are
2051marked as ended at the point at which the C<(*ACCEPT)> was encountered.
2052For instance:
2053
2054  'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
2055
2056will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
2057be set. If another branch in the inner parentheses was matched, such as in the
2058string 'ACDE', then the C<D> and C<E> would have to be matched as well.
2059
2060=back
2061
2062=back
2063
2064=head2 Backtracking
2065X<backtrack> X<backtracking>
2066
2067NOTE: This section presents an abstract approximation of regular
2068expression behavior.  For a more rigorous (and complicated) view of
2069the rules involved in selecting a match among possible alternatives,
2070see L<Combining RE Pieces>.
2071
2072A fundamental feature of regular expression matching involves the
2073notion called I<backtracking>, which is currently used (when needed)
2074by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
2075C<+?>, C<{n,m}>, and C<{n,m}?>.  Backtracking is often optimized
2076internally, but the general principle outlined here is valid.
2077
2078For a regular expression to match, the I<entire> regular expression must
2079match, not just part of it.  So if the beginning of a pattern containing a
2080quantifier succeeds in a way that causes later parts in the pattern to
2081fail, the matching engine backs up and recalculates the beginning
2082part--that's why it's called backtracking.
2083
2084Here is an example of backtracking:  Let's say you want to find the
2085word following "foo" in the string "Food is on the foo table.":
2086
2087    $_ = "Food is on the foo table.";
2088    if ( /\b(foo)\s+(\w+)/i ) {
2089        print "$2 follows $1.\n";
2090    }
2091
2092When the match runs, the first part of the regular expression (C<\b(foo)>)
2093finds a possible match right at the beginning of the string, and loads up
2094$1 with "Foo".  However, as soon as the matching engine sees that there's
2095no whitespace following the "Foo" that it had saved in $1, it realizes its
2096mistake and starts over again one character after where it had the
2097tentative match.  This time it goes all the way until the next occurrence
2098of "foo". The complete regular expression matches this time, and you get
2099the expected output of "table follows foo."
2100
2101Sometimes minimal matching can help a lot.  Imagine you'd like to match
2102everything between "foo" and "bar".  Initially, you write something
2103like this:
2104
2105    $_ =  "The food is under the bar in the barn.";
2106    if ( /foo(.*)bar/ ) {
2107        print "got <$1>\n";
2108    }
2109
2110Which perhaps unexpectedly yields:
2111
2112  got <d is under the bar in the >
2113
2114That's because C<.*> was greedy, so you get everything between the
2115I<first> "foo" and the I<last> "bar".  Here it's more effective
2116to use minimal matching to make sure you get the text between a "foo"
2117and the first "bar" thereafter.
2118
2119    if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
2120  got <d is under the >
2121
2122Here's another example. Let's say you'd like to match a number at the end
2123of a string, and you also want to keep the preceding part of the match.
2124So you write this:
2125
2126    $_ = "I have 2 numbers: 53147";
2127    if ( /(.*)(\d*)/ ) {                                # Wrong!
2128        print "Beginning is <$1>, number is <$2>.\n";
2129    }
2130
2131That won't work at all, because C<.*> was greedy and gobbled up the
2132whole string. As C<\d*> can match on an empty string the complete
2133regular expression matched successfully.
2134
2135    Beginning is <I have 2 numbers: 53147>, number is <>.
2136
2137Here are some variants, most of which don't work:
2138
2139    $_ = "I have 2 numbers: 53147";
2140    @pats = qw{
2141        (.*)(\d*)
2142        (.*)(\d+)
2143        (.*?)(\d*)
2144        (.*?)(\d+)
2145        (.*)(\d+)$
2146        (.*?)(\d+)$
2147        (.*)\b(\d+)$
2148        (.*\D)(\d+)$
2149    };
2150
2151    for $pat (@pats) {
2152        printf "%-12s ", $pat;
2153        if ( /$pat/ ) {
2154            print "<$1> <$2>\n";
2155        } else {
2156            print "FAIL\n";
2157        }
2158    }
2159
2160That will print out:
2161
2162    (.*)(\d*)    <I have 2 numbers: 53147> <>
2163    (.*)(\d+)    <I have 2 numbers: 5314> <7>
2164    (.*?)(\d*)   <> <>
2165    (.*?)(\d+)   <I have > <2>
2166    (.*)(\d+)$   <I have 2 numbers: 5314> <7>
2167    (.*?)(\d+)$  <I have 2 numbers: > <53147>
2168    (.*)\b(\d+)$ <I have 2 numbers: > <53147>
2169    (.*\D)(\d+)$ <I have 2 numbers: > <53147>
2170
2171As you see, this can be a bit tricky.  It's important to realize that a
2172regular expression is merely a set of assertions that gives a definition
2173of success.  There may be 0, 1, or several different ways that the
2174definition might succeed against a particular string.  And if there are
2175multiple ways it might succeed, you need to understand backtracking to
2176know which variety of success you will achieve.
2177
2178When using look-ahead assertions and negations, this can all get even
2179trickier.  Imagine you'd like to find a sequence of non-digits not
2180followed by "123".  You might try to write that as
2181
2182    $_ = "ABC123";
2183    if ( /^\D*(?!123)/ ) {                # Wrong!
2184        print "Yup, no 123 in $_\n";
2185    }
2186
2187But that isn't going to match; at least, not the way you're hoping.  It
2188claims that there is no 123 in the string.  Here's a clearer picture of
2189why that pattern matches, contrary to popular expectations:
2190
2191    $x = 'ABC123';
2192    $y = 'ABC445';
2193
2194    print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
2195    print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
2196
2197    print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
2198    print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
2199
2200This prints
2201
2202    2: got ABC
2203    3: got AB
2204    4: got ABC
2205
2206You might have expected test 3 to fail because it seems to a more
2207general purpose version of test 1.  The important difference between
2208them is that test 3 contains a quantifier (C<\D*>) and so can use
2209backtracking, whereas test 1 will not.  What's happening is
2210that you've asked "Is it true that at the start of $x, following 0 or more
2211non-digits, you have something that's not 123?"  If the pattern matcher had
2212let C<\D*> expand to "ABC", this would have caused the whole pattern to
2213fail.
2214
2215The search engine will initially match C<\D*> with "ABC".  Then it will
2216try to match C<(?!123)> with "123", which fails.  But because
2217a quantifier (C<\D*>) has been used in the regular expression, the
2218search engine can backtrack and retry the match differently
2219in the hope of matching the complete regular expression.
2220
2221The pattern really, I<really> wants to succeed, so it uses the
2222standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
2223time.  Now there's indeed something following "AB" that is not
2224"123".  It's "C123", which suffices.
2225
2226We can deal with this by using both an assertion and a negation.
2227We'll say that the first part in $1 must be followed both by a digit
2228and by something that's not "123".  Remember that the look-aheads
2229are zero-width expressions--they only look, but don't consume any
2230of the string in their match.  So rewriting this way produces what
2231you'd expect; that is, case 5 will fail, but case 6 succeeds:
2232
2233    print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
2234    print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
2235
2236    6: got ABC
2237
2238In other words, the two zero-width assertions next to each other work as though
2239they're ANDed together, just as you'd use any built-in assertions:  C</^$/>
2240matches only if you're at the beginning of the line AND the end of the
2241line simultaneously.  The deeper underlying truth is that juxtaposition in
2242regular expressions always means AND, except when you write an explicit OR
2243using the vertical bar.  C</ab/> means match "a" AND (then) match "b",
2244although the attempted matches are made at different positions because "a"
2245is not a zero-width assertion, but a one-width assertion.
2246
2247B<WARNING>: Particularly complicated regular expressions can take
2248exponential time to solve because of the immense number of possible
2249ways they can use backtracking to try for a match.  For example, without
2250internal optimizations done by the regular expression engine, this will
2251take a painfully long time to run:
2252
2253    'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
2254
2255And if you used C<*>'s in the internal groups instead of limiting them
2256to 0 through 5 matches, then it would take forever--or until you ran
2257out of stack space.  Moreover, these internal optimizations are not
2258always applicable.  For example, if you put C<{0,5}> instead of C<*>
2259on the external group, no current optimization is applicable, and the
2260match takes a long time to finish.
2261
2262A powerful tool for optimizing such beasts is what is known as an
2263"independent group",
2264which does not backtrack (see L</C<< (?>pattern) >>>).  Note also that
2265zero-length look-ahead/look-behind assertions will not backtrack to make
2266the tail match, since they are in "logical" context: only
2267whether they match is considered relevant.  For an example
2268where side-effects of look-ahead I<might> have influenced the
2269following match, see L</C<< (?>pattern) >>>.
2270
2271=head2 Version 8 Regular Expressions
2272X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
2273
2274In case you're not familiar with the "regular" Version 8 regex
2275routines, here are the pattern-matching rules not described above.
2276
2277Any single character matches itself, unless it is a I<metacharacter>
2278with a special meaning described here or above.  You can cause
2279characters that normally function as metacharacters to be interpreted
2280literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
2281character; "\\" matches a "\"). This escape mechanism is also required
2282for the character used as the pattern delimiter.
2283
2284A series of characters matches that series of characters in the target
2285string, so the pattern C<blurfl> would match "blurfl" in the target
2286string.
2287
2288You can specify a character class, by enclosing a list of characters
2289in C<[]>, which will match any character from the list.  If the
2290first character after the "[" is "^", the class matches any character not
2291in the list.  Within a list, the "-" character specifies a
2292range, so that C<a-z> represents all characters between "a" and "z",
2293inclusive.  If you want either "-" or "]" itself to be a member of a
2294class, put it at the start of the list (possibly after a "^"), or
2295escape it with a backslash.  "-" is also taken literally when it is
2296at the end of the list, just before the closing "]".  (The
2297following all specify the same class of three characters: C<[-az]>,
2298C<[az-]>, and C<[a\-z]>.  All are different from C<[a-z]>, which
2299specifies a class containing twenty-six characters, even on EBCDIC-based
2300character sets.)  Also, if you try to use the character
2301classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
2302a range, the "-" is understood literally.
2303
2304Note also that the whole range idea is rather unportable between
2305character sets--and even within character sets they may cause results
2306you probably didn't expect.  A sound principle is to use only ranges
2307that begin from and end at either alphabetics of equal case ([a-e],
2308[A-E]), or digits ([0-9]).  Anything else is unsafe.  If in doubt,
2309spell out the character sets in full.
2310
2311Characters may be specified using a metacharacter syntax much like that
2312used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
2313"\f" a form feed, etc.  More generally, \I<nnn>, where I<nnn> is a string
2314of three octal digits, matches the character whose coded character set value
2315is I<nnn>.  Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
2316matches the character whose ordinal is I<nn>. The expression \cI<x>
2317matches the character control-I<x>.  Finally, the "." metacharacter
2318matches any character except "\n" (unless you use C</s>).
2319
2320You can specify a series of alternatives for a pattern using "|" to
2321separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
2322or "foe" in the target string (as would C<f(e|i|o)e>).  The
2323first alternative includes everything from the last pattern delimiter
2324("(", "(?:", etc. or the beginning of the pattern) up to the first "|", and
2325the last alternative contains everything from the last "|" to the next
2326closing pattern delimiter.  That's why it's common practice to include
2327alternatives in parentheses: to minimize confusion about where they
2328start and end.
2329
2330Alternatives are tried from left to right, so the first
2331alternative found for which the entire expression matches, is the one that
2332is chosen. This means that alternatives are not necessarily greedy. For
2333example: when matching C<foo|foot> against "barefoot", only the "foo"
2334part will match, as that is the first alternative tried, and it successfully
2335matches the target string. (This might not seem important, but it is
2336important when you are capturing matched text using parentheses.)
2337
2338Also remember that "|" is interpreted as a literal within square brackets,
2339so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
2340
2341Within a pattern, you may designate subpatterns for later reference
2342by enclosing them in parentheses, and you may refer back to the
2343I<n>th subpattern later in the pattern using the metacharacter
2344\I<n> or \gI<n>.  Subpatterns are numbered based on the left to right order
2345of their opening parenthesis.  A backreference matches whatever
2346actually matched the subpattern in the string being examined, not
2347the rules for that subpattern.  Therefore, C<(0|0x)\d*\s\g1\d*> will
2348match "0x1234 0x4321", but not "0x1234 01234", because subpattern
23491 matched "0x", even though the rule C<0|0x> could potentially match
2350the leading 0 in the second number.
2351
2352=head2 Warning on \1 Instead of $1
2353
2354Some people get too used to writing things like:
2355
2356    $pattern =~ s/(\W)/\\\1/g;
2357
2358This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
2359shocking the
2360B<sed> addicts, but it's a dirty habit to get into.  That's because in
2361PerlThink, the righthand side of an C<s///> is a double-quoted string.  C<\1> in
2362the usual double-quoted string means a control-A.  The customary Unix
2363meaning of C<\1> is kludged in for C<s///>.  However, if you get into the habit
2364of doing that, you get yourself into trouble if you then add an C</e>
2365modifier.
2366
2367    s/(\d+)/ \1 + 1 /eg;            # causes warning under -w
2368
2369Or if you try to do
2370
2371    s/(\d+)/\1000/;
2372
2373You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
2374C<${1}000>.  The operation of interpolation should not be confused
2375with the operation of matching a backreference.  Certainly they mean two
2376different things on the I<left> side of the C<s///>.
2377
2378=head2 Repeated Patterns Matching a Zero-length Substring
2379
2380B<WARNING>: Difficult material (and prose) ahead.  This section needs a rewrite.
2381
2382Regular expressions provide a terse and powerful programming language.  As
2383with most other power tools, power comes together with the ability
2384to wreak havoc.
2385
2386A common abuse of this power stems from the ability to make infinite
2387loops using regular expressions, with something as innocuous as:
2388
2389    'foo' =~ m{ ( o? )* }x;
2390
2391The C<o?> matches at the beginning of C<'foo'>, and since the position
2392in the string is not moved by the match, C<o?> would match again and again
2393because of the C<*> quantifier.  Another common way to create a similar cycle
2394is with the looping modifier C<//g>:
2395
2396    @matches = ( 'foo' =~ m{ o? }xg );
2397
2398or
2399
2400    print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
2401
2402or the loop implied by split().
2403
2404However, long experience has shown that many programming tasks may
2405be significantly simplified by using repeated subexpressions that
2406may match zero-length substrings.  Here's a simple example being:
2407
2408    @chars = split //, $string;           # // is not magic in split
2409    ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
2410
2411Thus Perl allows such constructs, by I<forcefully breaking
2412the infinite loop>.  The rules for this are different for lower-level
2413loops given by the greedy quantifiers C<*+{}>, and for higher-level
2414ones like the C</g> modifier or split() operator.
2415
2416The lower-level loops are I<interrupted> (that is, the loop is
2417broken) when Perl detects that a repeated expression matched a
2418zero-length substring.   Thus
2419
2420   m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2421
2422is made equivalent to
2423
2424   m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
2425
2426For example, this program
2427
2428   #!perl -l
2429   "aaaaab" =~ /
2430     (?:
2431        a                 # non-zero
2432        |                 # or
2433       (?{print "hello"}) # print hello whenever this
2434                          #    branch is tried
2435       (?=(b))            # zero-width assertion
2436     )*  # any number of times
2437    /x;
2438   print $&;
2439   print $1;
2440
2441prints
2442
2443   hello
2444   aaaaa
2445   b
2446
2447Notice that "hello" is only printed once, as when Perl sees that the sixth
2448iteration of the outermost C<(?:)*> matches a zero-length string, it stops
2449the C<*>.
2450
2451The higher-level loops preserve an additional state between iterations:
2452whether the last match was zero-length.  To break the loop, the following
2453match after a zero-length match is prohibited to have a length of zero.
2454This prohibition interacts with backtracking (see L<"Backtracking">),
2455and so the I<second best> match is chosen if the I<best> match is of
2456zero length.
2457
2458For example:
2459
2460    $_ = 'bar';
2461    s/\w??/<$&>/g;
2462
2463results in C<< <><b><><a><><r><> >>.  At each position of the string the best
2464match given by non-greedy C<??> is the zero-length match, and the I<second
2465best> match is what is matched by C<\w>.  Thus zero-length matches
2466alternate with one-character-long matches.
2467
2468Similarly, for repeated C<m/()/g> the second-best match is the match at the
2469position one notch further in the string.
2470
2471The additional state of being I<matched with zero-length> is associated with
2472the matched string, and is reset by each assignment to pos().
2473Zero-length matches at the end of the previous match are ignored
2474during C<split>.
2475
2476=head2 Combining RE Pieces
2477
2478Each of the elementary pieces of regular expressions which were described
2479before (such as C<ab> or C<\Z>) could match at most one substring
2480at the given position of the input string.  However, in a typical regular
2481expression these elementary pieces are combined into more complicated
2482patterns using combining operators C<ST>, C<S|T>, C<S*> etc.
2483(in these examples C<S> and C<T> are regular subexpressions).
2484
2485Such combinations can include alternatives, leading to a problem of choice:
2486if we match a regular expression C<a|ab> against C<"abc">, will it match
2487substring C<"a"> or C<"ab">?  One way to describe which substring is
2488actually matched is the concept of backtracking (see L<"Backtracking">).
2489However, this description is too low-level and makes you think
2490in terms of a particular implementation.
2491
2492Another description starts with notions of "better"/"worse".  All the
2493substrings which may be matched by the given regular expression can be
2494sorted from the "best" match to the "worst" match, and it is the "best"
2495match which is chosen.  This substitutes the question of "what is chosen?"
2496by the question of "which matches are better, and which are worse?".
2497
2498Again, for elementary pieces there is no such question, since at most
2499one match at a given position is possible.  This section describes the
2500notion of better/worse for combining operators.  In the description
2501below C<S> and C<T> are regular subexpressions.
2502
2503=over 4
2504
2505=item C<ST>
2506
2507Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2508substrings which can be matched by C<S>, C<B> and C<B'> are substrings
2509which can be matched by C<T>.
2510
2511If C<A> is a better match for C<S> than C<A'>, C<AB> is a better
2512match than C<A'B'>.
2513
2514If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
2515C<B> is a better match for C<T> than C<B'>.
2516
2517=item C<S|T>
2518
2519When C<S> can match, it is a better match than when only C<T> can match.
2520
2521Ordering of two matches for C<S> is the same as for C<S>.  Similar for
2522two matches for C<T>.
2523
2524=item C<S{REPEAT_COUNT}>
2525
2526Matches as C<SSS...S> (repeated as many times as necessary).
2527
2528=item C<S{min,max}>
2529
2530Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2531
2532=item C<S{min,max}?>
2533
2534Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2535
2536=item C<S?>, C<S*>, C<S+>
2537
2538Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2539
2540=item C<S??>, C<S*?>, C<S+?>
2541
2542Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2543
2544=item C<< (?>S) >>
2545
2546Matches the best match for C<S> and only that.
2547
2548=item C<(?=S)>, C<(?<=S)>
2549
2550Only the best match for C<S> is considered.  (This is important only if
2551C<S> has capturing parentheses, and backreferences are used somewhere
2552else in the whole regular expression.)
2553
2554=item C<(?!S)>, C<(?<!S)>
2555
2556For this grouping operator there is no need to describe the ordering, since
2557only whether or not C<S> can match is important.
2558
2559=item C<(??{ EXPR })>, C<(?I<PARNO>)>
2560
2561The ordering is the same as for the regular expression which is
2562the result of EXPR, or the pattern contained by capture group I<PARNO>.
2563
2564=item C<(?(condition)yes-pattern|no-pattern)>
2565
2566Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2567already determined.  The ordering of the matches is the same as for the
2568chosen subexpression.
2569
2570=back
2571
2572The above recipes describe the ordering of matches I<at a given position>.
2573One more rule is needed to understand how a match is determined for the
2574whole regular expression: a match at an earlier position is always better
2575than a match at a later position.
2576
2577=head2 Creating Custom RE Engines
2578
2579As of Perl 5.10.0, one can create custom regular expression engines.  This
2580is not for the faint of heart, as they have to plug in at the C level.  See
2581L<perlreapi> for more details.
2582
2583As an alternative, overloaded constants (see L<overload>) provide a simple
2584way to extend the functionality of the RE engine, by substituting one
2585pattern for another.
2586
2587Suppose that we want to enable a new RE escape-sequence C<\Y|> which
2588matches at a boundary between whitespace characters and non-whitespace
2589characters.  Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2590at these positions, so we want to have each C<\Y|> in the place of the
2591more complicated version.  We can create a module C<customre> to do
2592this:
2593
2594    package customre;
2595    use overload;
2596
2597    sub import {
2598      shift;
2599      die "No argument to customre::import allowed" if @_;
2600      overload::constant 'qr' => \&convert;
2601    }
2602
2603    sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2604
2605    # We must also take care of not escaping the legitimate \\Y|
2606    # sequence, hence the presence of '\\' in the conversion rules.
2607    my %rules = ( '\\' => '\\\\',
2608                  'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
2609    sub convert {
2610      my $re = shift;
2611      $re =~ s{
2612                \\ ( \\ | Y . )
2613              }
2614              { $rules{$1} or invalid($re,$1) }sgex;
2615      return $re;
2616    }
2617
2618Now C<use customre> enables the new escape in constant regular
2619expressions, i.e., those without any runtime variable interpolations.
2620As documented in L<overload>, this conversion will work only over
2621literal parts of regular expressions.  For C<\Y|$re\Y|> the variable
2622part of this regular expression needs to be converted explicitly
2623(but only if the special meaning of C<\Y|> should be enabled inside $re):
2624
2625    use customre;
2626    $re = <>;
2627    chomp $re;
2628    $re = customre::convert $re;
2629    /\Y|$re\Y|/;
2630
2631=head2 Embedded Code Execution Frequency
2632
2633The exact rules for how often (??{}) and (?{}) are executed in a pattern
2634are unspecified.  In the case of a successful match you can assume that
2635they DWIM and will be executed in left to right order the appropriate
2636number of times in the accepting path of the pattern as would any other
2637meta-pattern.  How non-accepting pathways and match failures affect the
2638number of times a pattern is executed is specifically unspecified and
2639may vary depending on what optimizations can be applied to the pattern
2640and is likely to change from version to version.
2641
2642For instance in
2643
2644  "aaabcdeeeee"=~/a(?{print "a"})b(?{print "b"})cde/;
2645
2646the exact number of times "a" or "b" are printed out is unspecified for
2647failure, but you may assume they will be printed at least once during
2648a successful match, additionally you may assume that if "b" is printed,
2649it will be preceded by at least one "a".
2650
2651In the case of branching constructs like the following:
2652
2653  /a(b|(?{ print "a" }))c(?{ print "c" })/;
2654
2655you can assume that the input "ac" will output "ac", and that "abc"
2656will output only "c".
2657
2658When embedded code is quantified, successful matches will call the
2659code once for each matched iteration of the quantifier.  For
2660example:
2661
2662  "good" =~ /g(?:o(?{print "o"}))*d/;
2663
2664will output "o" twice.
2665
2666=head2 PCRE/Python Support
2667
2668As of Perl 5.10.0, Perl supports several Python/PCRE-specific extensions
2669to the regex syntax. While Perl programmers are encouraged to use the
2670Perl-specific syntax, the following are also accepted:
2671
2672=over 4
2673
2674=item C<< (?PE<lt>NAMEE<gt>pattern) >>
2675
2676Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
2677
2678=item C<< (?P=NAME) >>
2679
2680Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
2681
2682=item C<< (?P>NAME) >>
2683
2684Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
2685
2686=back
2687
2688=head1 BUGS
2689
2690Many regular expression constructs don't work on EBCDIC platforms.
2691
2692There are a number of issues with regard to case-insensitive matching
2693in Unicode rules.  See C<i> under L</Modifiers> above.
2694
2695This document varies from difficult to understand to completely
2696and utterly opaque.  The wandering prose riddled with jargon is
2697hard to fathom in several places.
2698
2699This document needs a rewrite that separates the tutorial content
2700from the reference content.
2701
2702=head1 SEE ALSO
2703
2704L<perlrequick>.
2705
2706L<perlretut>.
2707
2708L<perlop/"Regexp Quote-Like Operators">.
2709
2710L<perlop/"Gory details of parsing quoted constructs">.
2711
2712L<perlfaq6>.
2713
2714L<perlfunc/pos>.
2715
2716L<perllocale>.
2717
2718L<perlebcdic>.
2719
2720I<Mastering Regular Expressions> by Jeffrey Friedl, published
2721by O'Reilly and Associates.
2722