This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Improve warnings about known issues in (?{...}) blocks in the regex engine
[perl5.git] / pod / perlre.pod
CommitLineData
a0d0e21e 1=head1 NAME
d74e8afc 2X<regular expression> X<regex> X<regexp>
a0d0e21e
LW
3
4perlre - Perl regular expressions
5
6=head1 DESCRIPTION
7
5d458dd8 8This page describes the syntax of regular expressions in Perl.
91e0c79e 9
cc46d5f2 10If you haven't used regular expressions before, a quick-start
91e0c79e
MJD
11introduction is available in L<perlrequick>, and a longer tutorial
12introduction is available in L<perlretut>.
13
14For reference on how regular expressions are used in matching
15operations, plus various examples of the same, see discussions of
16C<m//>, C<s///>, C<qr//> and C<??> in L<perlop/"Regexp Quote-Like
17Operators">.
cb1a09d0 18
0d017f4d
WL
19
20=head2 Modifiers
21
19799a22 22Matching operations can have various modifiers. Modifiers
5a964f20 23that relate to the interpretation of the regular expression inside
19799a22 24are listed below. Modifiers that alter the way a regular expression
5d458dd8 25is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
1e66bd83 26L<perlop/"Gory details of parsing quoted constructs">.
a0d0e21e 27
55497cff
PP
28=over 4
29
54310121 30=item m
d74e8afc 31X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff
PP
32
33Treat string as multiple lines. That is, change "^" and "$" from matching
14218588 34the start or end of the string to matching the start or end of any
7f761169 35line anywhere within the string.
55497cff 36
54310121 37=item s
d74e8afc
ITB
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
55497cff
PP
40
41Treat string as single line. That is, change "." to match any character
19799a22 42whatsoever, even a newline, which normally it would not match.
55497cff 43
f02c194e 44Used together, as /ms, they let the "." match any character whatsoever,
fb55449c 45while still allowing "^" and "$" to match, respectively, just after
19799a22 46and just before newlines within the string.
7b8d334a 47
87e95b7f
YO
48=item i
49X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
50X<regular expression, case-insensitive>
51
52Do case-insensitive pattern matching.
53
54If C<use locale> is in effect, the case map is taken from the current
55locale. See L<perllocale>.
56
54310121 57=item x
d74e8afc 58X</x>
55497cff
PP
59
60Extend your pattern's legibility by permitting whitespace and comments.
61
87e95b7f
YO
62=item p
63X</p> X<regex, preserve> X<regexp, preserve>
64
632a1772 65Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
87e95b7f
YO
66${^POSTMATCH} are available for use after matching.
67
e2e6bec7
DN
68=item g and c
69X</g> X</c>
70
71Global matching, and keep the Current position after failed matching.
72Unlike i, m, s and x, these two flags affect the way the regex is used
73rather than the regex itself. See
74L<perlretut/"Using regular expressions in Perl"> for further explanation
75of the g and c modifiers.
76
55497cff 77=back
a0d0e21e
LW
78
79These are usually written as "the C</x> modifier", even though the delimiter
14218588 80in question might not really be a slash. Any of these
a0d0e21e 81modifiers may also be embedded within the regular expression itself using
14218588 82the C<(?...)> construct. See below.
a0d0e21e 83
4633a7c4 84The C</x> modifier itself needs a little more explanation. It tells
55497cff
PP
85the regular expression parser to ignore whitespace that is neither
86backslashed nor within a character class. You can use this to break up
4633a7c4 87your regular expression into (slightly) more readable parts. The C<#>
54310121 88character is also treated as a metacharacter introducing a comment,
55497cff 89just as in ordinary Perl code. This also means that if you want real
14218588 90whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a
RGS
91class, where they are unaffected by C</x>), then you'll either have to
92escape them (using backslashes or C<\Q...\E>) or encode them using octal
8933a740
RGS
93or hex escapes. Taken together, these features go a long way towards
94making Perl's regular expressions more readable. Note that you have to
95be careful not to include the pattern delimiter in the comment--perl has
96no way of knowing you did not intend to close the pattern early. See
97the C-comment deletion code in L<perlop>. Also note that anything inside
1031e5db 98a C<\Q...\E> stays unaffected by C</x>.
d74e8afc 99X</x>
a0d0e21e
LW
100
101=head2 Regular Expressions
102
04838cea
RGS
103=head3 Metacharacters
104
384f06ae 105The patterns used in Perl pattern matching evolved from those supplied in
14218588 106the Version 8 regex routines. (The routines are derived
19799a22
GS
107(distantly) from Henry Spencer's freely redistributable reimplementation
108of the V8 routines.) See L<Version 8 Regular Expressions> for
109details.
a0d0e21e
LW
110
111In particular the following metacharacters have their standard I<egrep>-ish
112meanings:
d74e8afc
ITB
113X<metacharacter>
114X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
115
a0d0e21e 116
54310121 117 \ Quote the next metacharacter
a0d0e21e
LW
118 ^ Match the beginning of the line
119 . Match any character (except newline)
c07a80fd 120 $ Match the end of the line (or before newline at the end)
a0d0e21e
LW
121 | Alternation
122 () Grouping
123 [] Character class
124
14218588
GS
125By default, the "^" character is guaranteed to match only the
126beginning of the string, the "$" character only the end (or before the
127newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
128assumption that the string contains only one line. Embedded newlines
129will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 130string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
131newline within the string (except if the newline is the last character in
132the string), and "$" will match before any newline. At the
a0d0e21e
LW
133cost of a little more overhead, you can do this by using the /m modifier
134on the pattern match operator. (Older programs did this by setting C<$*>,
f02c194e 135but this practice has been removed in perl 5.9.)
d74e8afc 136X<^> X<$> X</m>
a0d0e21e 137
14218588 138To simplify multi-line substitutions, the "." character never matches a
55497cff 139newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 140the string is a single line--even if it isn't.
d74e8afc 141X<.> X</s>
a0d0e21e 142
04838cea
RGS
143=head3 Quantifiers
144
a0d0e21e 145The following standard quantifiers are recognized:
d74e8afc 146X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e
LW
147
148 * Match 0 or more times
149 + Match 1 or more times
150 ? Match 1 or 0 times
151 {n} Match exactly n times
152 {n,} Match at least n times
153 {n,m} Match at least n but not more than m times
154
155(If a curly bracket occurs in any other context, it is treated
b975c076 156as a regular character. In particular, the lower bound
527e91da
BB
157is not optional.) The "*" quantifier is equivalent to C<{0,}>, the "+"
158quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
9c79236d
GS
159to integral values less than a preset limit defined when perl is built.
160This is usually 32766 on the most common platforms. The actual limit can
161be seen in the error message generated by code such as this:
162
820475bd 163 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 164
54310121
PP
165By default, a quantified subpattern is "greedy", that is, it will match as
166many times as possible (given a particular starting location) while still
167allowing the rest of the pattern to match. If you want it to match the
168minimum number of times possible, follow the quantifier with a "?". Note
169that the meanings don't change, just the "greediness":
0d017f4d 170X<metacharacter> X<greedy> X<greediness>
d74e8afc 171X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 172
0d017f4d
WL
173 *? Match 0 or more times, not greedily
174 +? Match 1 or more times, not greedily
175 ?? Match 0 or 1 time, not greedily
176 {n}? Match exactly n times, not greedily
177 {n,}? Match at least n times, not greedily
178 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 179
b9b4dddf
YO
180By default, when a quantified subpattern does not allow the rest of the
181overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 182sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
183as well.
184
0d017f4d
WL
185 *+ Match 0 or more times and give nothing back
186 ++ Match 1 or more times and give nothing back
187 ?+ Match 0 or 1 time and give nothing back
b9b4dddf 188 {n}+ Match exactly n times and give nothing back (redundant)
04838cea
RGS
189 {n,}+ Match at least n times and give nothing back
190 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
191
192For instance,
193
194 'aaaa' =~ /a++a/
195
196will never match, as the C<a++> will gobble up all the C<a>'s in the
197string and won't leave any for the remaining part of the pattern. This
198feature can be extremely useful to give perl hints about where it
199shouldn't backtrack. For instance, the typical "match a double-quoted
200string" problem can be most efficiently performed when written as:
201
202 /"(?:[^"\\]++|\\.)*+"/
203
0d017f4d 204as we know that if the final quote does not match, backtracking will not
b9b4dddf
YO
205help. See the independent subexpression C<< (?>...) >> for more details;
206possessive quantifiers are just syntactic sugar for that construct. For
207instance the above example could also be written as follows:
208
209 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
210
04838cea
RGS
211=head3 Escape sequences
212
5f05dabc 213Because patterns are processed as double quoted strings, the following
a0d0e21e 214also work:
0d017f4d 215X<\t> X<\n> X<\r> X<\f> X<\e> X<\a> X<\l> X<\u> X<\L> X<\U> X<\E> X<\Q>
d74e8afc 216X<\0> X<\c> X<\N> X<\x>
a0d0e21e 217
0f36ee90
PP
218 \t tab (HT, TAB)
219 \n newline (LF, NL)
220 \r return (CR)
221 \f form feed (FF)
222 \a alarm (bell) (BEL)
223 \e escape (think troff) (ESC)
0d017f4d
WL
224 \033 octal char (example: ESC)
225 \x1B hex char (example: ESC)
196ac2fc 226 \x{263a} long hex char (example: Unicode SMILEY)
0d017f4d 227 \cK control char (example: VT)
196ac2fc 228 \N{name} named Unicode character
cb1a09d0
AD
229 \l lowercase next char (think vi)
230 \u uppercase next char (think vi)
231 \L lowercase till \E (think vi)
232 \U uppercase till \E (think vi)
233 \E end case modification (think vi)
5a964f20 234 \Q quote (disable) pattern metacharacters till \E
a0d0e21e 235
a034a98d 236If C<use locale> is in effect, the case map used by C<\l>, C<\L>, C<\u>
423cee85 237and C<\U> is taken from the current locale. See L<perllocale>. For
4a2d328f 238documentation of C<\N{name}>, see L<charnames>.
a034a98d 239
1d2dff63
GS
240You cannot include a literal C<$> or C<@> within a C<\Q> sequence.
241An unescaped C<$> or C<@> interpolates the corresponding variable,
242while escaping will cause the literal string C<\$> to be matched.
243You'll need to write something like C<m/\Quser\E\@\Qhost/>.
244
e1d1eefb 245=head3 Character Classes and other Special Escapes
04838cea 246
a0d0e21e 247In addition, Perl defines the following:
d74e8afc 248X<\w> X<\W> X<\s> X<\S> X<\d> X<\D> X<\X> X<\p> X<\P> X<\C>
f7819f85 249X<\g> X<\k> X<\N> X<\K> X<\v> X<\V> X<\h> X<\H>
0d017f4d 250X<word> X<whitespace> X<character class> X<backreference>
a0d0e21e 251
81714fb9
YO
252 \w Match a "word" character (alphanumeric plus "_")
253 \W Match a non-"word" character
254 \s Match a whitespace character
255 \S Match a non-whitespace character
256 \d Match a digit character
257 \D Match a non-digit character
258 \pP Match P, named property. Use \p{Prop} for longer names.
259 \PP Match non-P
260 \X Match eXtended Unicode "combining character sequence",
e1f17637 261 equivalent to (?>\PM\pM*)
81714fb9
YO
262 \C Match a single C char (octet) even under Unicode.
263 NOTE: breaks up characters into their UTF-8 bytes,
264 so you may end up with malformed pieces of UTF-8.
265 Unsupported in lookbehind.
5d458dd8 266 \1 Backreference to a specific group.
c74340f9 267 '1' may actually be any positive integer.
2bf803e2
YO
268 \g1 Backreference to a specific or previous group,
269 \g{-1} number may be negative indicating a previous buffer and may
270 optionally be wrapped in curly brackets for safer parsing.
1f1031fe 271 \g{name} Named backreference
81714fb9 272 \k<name> Named backreference
ee9b8eae 273 \K Keep the stuff left of the \K, don't include it in $&
c741660a 274 \N Any character but \n
e1d1eefb
YO
275 \v Vertical whitespace
276 \V Not vertical whitespace
277 \h Horizontal whitespace
278 \H Not horizontal whitespace
2ddf2931 279 \R Linebreak
a0d0e21e 280
08ce8fc6
JH
281A C<\w> matches a single alphanumeric character (an alphabetic
282character, or a decimal digit) or C<_>, not a whole word. Use C<\w+>
283to match a string of Perl-identifier characters (which isn't the same
284as matching an English word). If C<use locale> is in effect, the list
285of alphabetic characters generated by C<\w> is taken from the current
286locale. See L<perllocale>. You may use C<\w>, C<\W>, C<\s>, C<\S>,
0d017f4d
WL
287C<\d>, and C<\D> within character classes, but they aren't usable
288as either end of a range. If any of them precedes or follows a "-",
289the "-" is understood literally. If Unicode is in effect, C<\s> matches
c62285ac 290also "\x{85}", "\x{2028}", and "\x{2029}". See L<perlunicode> for more
0d017f4d
WL
291details about C<\pP>, C<\PP>, C<\X> and the possibility of defining
292your own C<\p> and C<\P> properties, and L<perluniintro> about Unicode
293in general.
d74e8afc 294X<\w> X<\W> X<word>
a0d0e21e 295
e1d1eefb 296C<\R> will atomically match a linebreak, including the network line-ending
e2cb52ee 297"\x0D\x0A". Specifically, X<\R> is exactly equivalent to
e1d1eefb
YO
298
299 (?>\x0D\x0A?|[\x0A-\x0C\x85\x{2028}\x{2029}])
300
301B<Note:> C<\R> has no special meaning inside of a character class;
302use C<\v> instead (vertical whitespace).
303X<\R>
304
b8c5462f 305The POSIX character class syntax
d74e8afc 306X<character class>
b8c5462f 307
820475bd 308 [:class:]
b8c5462f 309
0d017f4d 310is also available. Note that the C<[> and C<]> brackets are I<literal>;
5496314a
SP
311they must always be used within a character class expression.
312
313 # this is correct:
314 $string =~ /[[:alpha:]]/;
315
316 # this is not, and will generate a warning:
317 $string =~ /[:alpha:]/;
318
6fa80ea2
YO
319The following table shows the mapping of POSIX character class
320names, common escapes, literal escape sequences and their equivalent
321Unicode style property names.
322X<character class> X<\p> X<\p{}>
d74e8afc
ITB
323X<alpha> X<alnum> X<ascii> X<blank> X<cntrl> X<digit> X<graph>
324X<lower> X<print> X<punct> X<space> X<upper> X<word> X<xdigit>
b8c5462f 325
6fa80ea2
YO
326B<Note:> up to Perl 5.10 the property names used were shared with
327standard Unicode properties, this was changed in Perl 5.11, see
328L<perl5110delta> for details.
329
330 POSIX Esc Class Property Note
331 --------------------------------------------------------
332 alnum [0-9A-Za-z] IsPosixAlnum
333 alpha [A-Za-z] IsPosixAlpha
334 ascii [\000-\177] IsASCII
335 blank [\011 ] IsPosixBlank [1]
336 cntrl [\0-\37\177] IsPosixCntrl
337 digit \d [0-9] IsPosixDigit
338 graph [!-~] IsPosixGraph
339 lower [a-z] IsPosixLower
340 print [ -~] IsPosixPrint
341 punct [!-/:-@[-`{-~] IsPosixPunct
342 space [\11-\15 ] IsPosixSpace [2]
343 \s [\11\12\14\15 ] IsPerlSpace [2]
344 upper [A-Z] IsPosixUpper
345 word \w [0-9A-Z_a-z] IsPerlWord [3]
346 xdigit [0-9A-Fa-f] IsXDigit
b8c5462f 347
07698885
RGS
348=over
349
350=item [1]
351
b432a672 352A GNU extension equivalent to C<[ \t]>, "all horizontal whitespace".
07698885
RGS
353
354=item [2]
355
6fa80ea2
YO
356Note that C<\s> and C<[[:space:]]> are B<not> equivalent as C<[[:space:]]>
357includes also the (very rare) "vertical tabulator", "\cK" or chr(11) in
358ASCII.
07698885
RGS
359
360=item [3]
361
08ce8fc6 362A Perl extension, see above.
07698885
RGS
363
364=back
aaa51d5e 365
26b44a0a 366For example use C<[:upper:]> to match all the uppercase characters.
aaa51d5e
JF
367Note that the C<[]> are part of the C<[::]> construct, not part of the
368whole character class. For example:
b8c5462f 369
820475bd 370 [01[:alpha:]%]
b8c5462f 371
0d017f4d 372matches zero, one, any alphabetic character, and the percent sign.
b8c5462f 373
345e2394
JV
374=over 4
375
fdf0a293
RB
376=item C<$>
377
378Currency symbol
379
380=item C<+> C<< < >> C<=> C<< > >> C<|> C<~>
381
382Mathematical symbols
383
384=item C<^> C<`>
385
386Modifier symbols (accents)
387
fdf0a293
RB
388
389=back
390
353c6505 391The other named classes are:
b8c5462f
JH
392
393=over 4
394
395=item cntrl
d74e8afc 396X<cntrl>
b8c5462f 397
820475bd
GS
398Any control character. Usually characters that don't produce output as
399such but instead control the terminal somehow: for example newline and
400backspace are control characters. All characters with ord() less than
0d017f4d 40132 are usually classified as control characters (assuming ASCII,
7be5a6cf
JF
402the ISO Latin character sets, and Unicode), as is the character with
403the ord() value of 127 (C<DEL>).
b8c5462f
JH
404
405=item graph
d74e8afc 406X<graph>
b8c5462f 407
f1cbbd6e 408Any alphanumeric or punctuation (special) character.
b8c5462f
JH
409
410=item print
d74e8afc 411X<print>
b8c5462f 412
f79b3095 413Any alphanumeric or punctuation (special) character or the space character.
b8c5462f
JH
414
415=item punct
d74e8afc 416X<punct>
b8c5462f 417
f1cbbd6e 418Any punctuation (special) character.
b8c5462f
JH
419
420=item xdigit
d74e8afc 421X<xdigit>
b8c5462f 422
593df60c 423Any hexadecimal digit. Though this may feel silly ([0-9A-Fa-f] would
820475bd 424work just fine) it is included for completeness.
b8c5462f 425
b8c5462f
JH
426=back
427
428You can negate the [::] character classes by prefixing the class name
429with a '^'. This is a Perl extension. For example:
d74e8afc 430X<character class, negation>
b8c5462f 431
5496314a 432 POSIX traditional Unicode
93733859 433
6fa80ea2
YO
434 [[:^digit:]] \D \P{IsPosixDigit}
435 [[:^space:]] \S \P{IsPosixSpace}
436 [[:^word:]] \W \P{IsPerlWord}
b8c5462f 437
54c18d04
MK
438Perl respects the POSIX standard in that POSIX character classes are
439only supported within a character class. The POSIX character classes
440[.cc.] and [=cc=] are recognized but B<not> supported and trying to
441use them will cause an error.
b8c5462f 442
04838cea
RGS
443=head3 Assertions
444
a0d0e21e 445Perl defines the following zero-width assertions:
d74e8afc
ITB
446X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
447X<regexp, zero-width assertion>
448X<regular expression, zero-width assertion>
449X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e
LW
450
451 \b Match a word boundary
0d017f4d 452 \B Match except at a word boundary
b85d18e9
IZ
453 \A Match only at beginning of string
454 \Z Match only at end of string, or before newline at the end
455 \z Match only at end of string
9da458fc
IZ
456 \G Match only at pos() (e.g. at the end-of-match position
457 of prior m//g)
a0d0e21e 458
14218588 459A word boundary (C<\b>) is a spot between two characters
19799a22
GS
460that has a C<\w> on one side of it and a C<\W> on the other side
461of it (in either order), counting the imaginary characters off the
462beginning and end of the string as matching a C<\W>. (Within
463character classes C<\b> represents backspace rather than a word
464boundary, just as it normally does in any double-quoted string.)
465The C<\A> and C<\Z> are just like "^" and "$", except that they
466won't match multiple times when the C</m> modifier is used, while
467"^" and "$" will match at every internal line boundary. To match
468the actual end of the string and not ignore an optional trailing
469newline, use C<\z>.
d74e8afc 470X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
471
472The C<\G> assertion can be used to chain global matches (using
473C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
474It is also useful when writing C<lex>-like scanners, when you have
475several patterns that you want to match against consequent substrings
476of your string, see the previous reference. The actual location
477where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d
YO
478an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
479matches is modified somewhat, in that contents to the left of C<\G> is
480not counted when determining the length of the match. Thus the following
481will not match forever:
d74e8afc 482X<\G>
c47ff5f1 483
58e23c8d
YO
484 $str = 'ABC';
485 pos($str) = 1;
486 while (/.\G/g) {
487 print $&;
488 }
489
490It will print 'A' and then terminate, as it considers the match to
491be zero-width, and thus will not match at the same position twice in a
492row.
493
494It is worth noting that C<\G> improperly used can result in an infinite
495loop. Take care when using patterns that include C<\G> in an alternation.
496
04838cea
RGS
497=head3 Capture buffers
498
0d017f4d
WL
499The bracketing construct C<( ... )> creates capture buffers. To refer
500to the current contents of a buffer later on, within the same pattern,
501use \1 for the first, \2 for the second, and so on.
502Outside the match use "$" instead of "\". (The
81714fb9 503\<digit> notation works in certain circumstances outside
14218588
GS
504the match. See the warning below about \1 vs $1 for details.)
505Referring back to another part of the match is called a
506I<backreference>.
d74e8afc
ITB
507X<regex, capture buffer> X<regexp, capture buffer>
508X<regular expression, capture buffer> X<backreference>
14218588
GS
509
510There is no limit to the number of captured substrings that you may
511use. However Perl also uses \10, \11, etc. as aliases for \010,
fb55449c
JH
512\011, etc. (Recall that 0 means octal, so \011 is the character at
513number 9 in your coded character set; which would be the 10th character,
81714fb9
YO
514a horizontal tab under ASCII.) Perl resolves this
515ambiguity by interpreting \10 as a backreference only if at least 10
516left parentheses have opened before it. Likewise \11 is a
517backreference only if at least 11 left parentheses have opened
518before it. And so on. \1 through \9 are always interpreted as
5624f11d 519backreferences.
c74340f9 520
1f1031fe 521X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
2bf803e2 522In order to provide a safer and easier way to construct patterns using
99d59c4d
RGS
523backreferences, Perl provides the C<\g{N}> notation (starting with perl
5245.10.0). The curly brackets are optional, however omitting them is less
525safe as the meaning of the pattern can be changed by text (such as digits)
526following it. When N is a positive integer the C<\g{N}> notation is
527exactly equivalent to using normal backreferences. When N is a negative
528integer then it is a relative backreference referring to the previous N'th
529capturing group. When the bracket form is used and N is not an integer, it
530is treated as a reference to a named buffer.
2bf803e2
YO
531
532Thus C<\g{-1}> refers to the last buffer, C<\g{-2}> refers to the
533buffer before that. For example:
5624f11d
YO
534
535 /
536 (Y) # buffer 1
537 ( # buffer 2
538 (X) # buffer 3
2bf803e2
YO
539 \g{-1} # backref to buffer 3
540 \g{-3} # backref to buffer 1
5624f11d
YO
541 )
542 /x
543
2bf803e2 544and would match the same as C</(Y) ( (X) \3 \1 )/x>.
14218588 545
99d59c4d 546Additionally, as of Perl 5.10.0 you may use named capture buffers and named
1f1031fe 547backreferences. The notation is C<< (?<name>...) >> to declare and C<< \k<name> >>
0d017f4d
WL
548to reference. You may also use apostrophes instead of angle brackets to delimit the
549name; and you may use the bracketed C<< \g{name} >> backreference syntax.
550It's possible to refer to a named capture buffer by absolute and relative number as well.
551Outside the pattern, a named capture buffer is available via the C<%+> hash.
552When different buffers within the same pattern have the same name, C<$+{name}>
553and C<< \k<name> >> refer to the leftmost defined group. (Thus it's possible
554to do things with named capture buffers that would otherwise require C<(??{})>
555code to accomplish.)
556X<named capture buffer> X<regular expression, named capture buffer>
64c5a566 557X<%+> X<$+{name}> X<< \k<name> >>
81714fb9 558
14218588 559Examples:
a0d0e21e
LW
560
561 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
562
81714fb9
YO
563 /(.)\1/ # find first doubled char
564 and print "'$1' is the first doubled character\n";
565
566 /(?<char>.)\k<char>/ # ... a different way
567 and print "'$+{char}' is the first doubled character\n";
568
0d017f4d 569 /(?'char'.)\1/ # ... mix and match
81714fb9 570 and print "'$1' is the first doubled character\n";
c47ff5f1 571
14218588 572 if (/Time: (..):(..):(..)/) { # parse out values
a0d0e21e
LW
573 $hours = $1;
574 $minutes = $2;
575 $seconds = $3;
576 }
c47ff5f1 577
14218588
GS
578Several special variables also refer back to portions of the previous
579match. C<$+> returns whatever the last bracket match matched.
580C<$&> returns the entire matched string. (At one point C<$0> did
581also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
582everything before the matched string. C<$'> returns everything
583after the matched string. And C<$^N> contains whatever was matched by
584the most-recently closed group (submatch). C<$^N> can be used in
585extended patterns (see below), for example to assign a submatch to a
81714fb9 586variable.
d74e8afc 587X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 588
665e98b9 589The numbered match variables ($1, $2, $3, etc.) and the related punctuation
77ea4f6d 590set (C<$+>, C<$&>, C<$`>, C<$'>, and C<$^N>) are all dynamically scoped
14218588
GS
591until the end of the enclosing block or until the next successful
592match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
593X<$+> X<$^N> X<$&> X<$`> X<$'>
594X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
595
14218588 596
0d017f4d 597B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 598which makes it easier to write code that tests for a series of more
665e98b9
JH
599specific cases and remembers the best match.
600
14218588
GS
601B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
602C<$'> anywhere in the program, it has to provide them for every
603pattern match. This may substantially slow your program. Perl
604uses the same mechanism to produce $1, $2, etc, so you also pay a
605price for each pattern that contains capturing parentheses. (To
606avoid this cost while retaining the grouping behaviour, use the
607extended regular expression C<(?: ... )> instead.) But if you never
608use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
609parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
610if you can, but if you can't (and some algorithms really appreciate
611them), once you've used them once, use them at will, because you've
612already paid the price. As of 5.005, C<$&> is not so costly as the
613other two.
d74e8afc 614X<$&> X<$`> X<$'>
68dc0745 615
99d59c4d 616As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
617C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
618and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 619successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
620The use of these variables incurs no global performance penalty, unlike
621their punctuation char equivalents, however at the trade-off that you
622have to tell perl when you want to use them.
87e95b7f 623X</p> X<p modifier>
cde0cee5 624
19799a22
GS
625Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
626C<\w>, C<\n>. Unlike some other regular expression languages, there
627are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 628that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
629interpreted as a literal character, not a metacharacter. This was
630once used in a common idiom to disable or quote the special meanings
631of regular expression metacharacters in a string that you want to
36bbe248 632use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
633
634 $pattern =~ s/(\W)/\\$1/g;
635
f1cbbd6e 636(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
637Today it is more common to use the quotemeta() function or the C<\Q>
638metaquoting escape sequence to disable all metacharacters' special
639meanings like this:
a0d0e21e
LW
640
641 /$unquoted\Q$quoted\E$unquoted/
642
9da458fc
IZ
643Beware that if you put literal backslashes (those not inside
644interpolated variables) between C<\Q> and C<\E>, double-quotish
645backslash interpolation may lead to confusing results. If you
646I<need> to use literal backslashes within C<\Q...\E>,
647consult L<perlop/"Gory details of parsing quoted constructs">.
648
19799a22
GS
649=head2 Extended Patterns
650
14218588
GS
651Perl also defines a consistent extension syntax for features not
652found in standard tools like B<awk> and B<lex>. The syntax is a
653pair of parentheses with a question mark as the first thing within
654the parentheses. The character after the question mark indicates
655the extension.
19799a22 656
14218588
GS
657The stability of these extensions varies widely. Some have been
658part of the core language for many years. Others are experimental
659and may change without warning or be completely removed. Check
660the documentation on an individual feature to verify its current
661status.
19799a22 662
14218588
GS
663A question mark was chosen for this and for the minimal-matching
664construct because 1) question marks are rare in older regular
665expressions, and 2) whenever you see one, you should stop and
666"question" exactly what is going on. That's psychology...
a0d0e21e
LW
667
668=over 10
669
cc6b7395 670=item C<(?#text)>
d74e8afc 671X<(?#)>
a0d0e21e 672
14218588 673A comment. The text is ignored. If the C</x> modifier enables
19799a22 674whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
675the comment as soon as it sees a C<)>, so there is no way to put a literal
676C<)> in the comment.
a0d0e21e 677
f7819f85 678=item C<(?pimsx-imsx)>
d74e8afc 679X<(?)>
19799a22 680
0b6d1084
JH
681One or more embedded pattern-match modifiers, to be turned on (or
682turned off, if preceded by C<->) for the remainder of the pattern or
683the remainder of the enclosing pattern group (if any). This is
684particularly useful for dynamic patterns, such as those read in from a
0d017f4d
WL
685configuration file, taken from an argument, or specified in a table
686somewhere. Consider the case where some patterns want to be case
687sensitive and some do not: The case insensitive ones merely need to
688include C<(?i)> at the front of the pattern. For example:
19799a22
GS
689
690 $pattern = "foobar";
5d458dd8 691 if ( /$pattern/i ) { }
19799a22
GS
692
693 # more flexible:
694
695 $pattern = "(?i)foobar";
5d458dd8 696 if ( /$pattern/ ) { }
19799a22 697
0b6d1084 698These modifiers are restored at the end of the enclosing group. For example,
19799a22
GS
699
700 ( (?i) blah ) \s+ \1
701
0d017f4d
WL
702will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
703repetition of the previous word, assuming the C</x> modifier, and no C</i>
704modifier outside this group.
19799a22 705
8eb5594e
DR
706These modifiers do not carry over into named subpatterns called in the
707enclosing group. In other words, a pattern such as C<((?i)(&NAME))> does not
708change the case-sensitivity of the "NAME" pattern.
709
5530442b 710Note that the C<p> modifier is special in that it can only be enabled,
cde0cee5 711not disabled, and that its presence anywhere in a pattern has a global
5530442b 712effect. Thus C<(?-p)> and C<(?-p:...)> are meaningless and will warn
cde0cee5
YO
713when executed under C<use warnings>.
714
5a964f20 715=item C<(?:pattern)>
d74e8afc 716X<(?:)>
a0d0e21e 717
ca9dfc88
IZ
718=item C<(?imsx-imsx:pattern)>
719
5a964f20
TC
720This is for clustering, not capturing; it groups subexpressions like
721"()", but doesn't make backreferences as "()" does. So
a0d0e21e 722
5a964f20 723 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
724
725is like
726
5a964f20 727 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 728
19799a22
GS
729but doesn't spit out extra fields. It's also cheaper not to capture
730characters if you don't need to.
a0d0e21e 731
19799a22 732Any letters between C<?> and C<:> act as flags modifiers as with
5d458dd8 733C<(?imsx-imsx)>. For example,
ca9dfc88
IZ
734
735 /(?s-i:more.*than).*million/i
736
14218588 737is equivalent to the more verbose
ca9dfc88
IZ
738
739 /(?:(?s-i)more.*than).*million/i
740
594d7033
YO
741=item C<(?|pattern)>
742X<(?|)> X<Branch reset>
743
744This is the "branch reset" pattern, which has the special property
745that the capture buffers are numbered from the same starting point
99d59c4d 746in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 747
693596a8
RGS
748Capture buffers are numbered from left to right, but inside this
749construct the numbering is restarted for each branch.
4deaaa80
PJ
750
751The numbering within each branch will be as normal, and any buffers
752following this construct will be numbered as though the construct
753contained only one branch, that being the one with the most capture
754buffers in it.
755
756This construct will be useful when you want to capture one of a
757number of alternative matches.
758
759Consider the following pattern. The numbers underneath show in
760which buffer the captured content will be stored.
594d7033
YO
761
762
763 # before ---------------branch-reset----------- after
764 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
765 # 1 2 2 3 2 3 4
766
90a18110
RGS
767Note: as of Perl 5.10.0, branch resets interfere with the contents of
768the C<%+> hash, that holds named captures. Consider using C<%-> instead.
769
ee9b8eae
YO
770=item Look-Around Assertions
771X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
772
773Look-around assertions are zero width patterns which match a specific
774pattern without including it in C<$&>. Positive assertions match when
775their subpattern matches, negative assertions match when their subpattern
776fails. Look-behind matches text up to the current match position,
777look-ahead matches text following the current match position.
778
779=over 4
780
5a964f20 781=item C<(?=pattern)>
d74e8afc 782X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 783
19799a22 784A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
785matches a word followed by a tab, without including the tab in C<$&>.
786
5a964f20 787=item C<(?!pattern)>
d74e8afc 788X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 789
19799a22 790A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 791matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
792however that look-ahead and look-behind are NOT the same thing. You cannot
793use this for look-behind.
7b8d334a 794
5a964f20 795If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
796will not do what you want. That's because the C<(?!foo)> is just saying that
797the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
798match. You would have to do something like C</(?!foo)...bar/> for that. We
799say "like" because there's the case of your "bar" not having three characters
800before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
801Sometimes it's still easier just to say:
a0d0e21e 802
a3cb178b 803 if (/bar/ && $` !~ /foo$/)
a0d0e21e 804
19799a22 805For look-behind see below.
c277df42 806
ee9b8eae
YO
807=item C<(?<=pattern)> C<\K>
808X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 809
c47ff5f1 810A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
811matches a word that follows a tab, without including the tab in C<$&>.
812Works only for fixed-width look-behind.
c277df42 813
ee9b8eae
YO
814There is a special form of this construct, called C<\K>, which causes the
815regex engine to "keep" everything it had matched prior to the C<\K> and
816not include it in C<$&>. This effectively provides variable length
817look-behind. The use of C<\K> inside of another look-around assertion
818is allowed, but the behaviour is currently not well defined.
819
c62285ac 820For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
821equivalent C<< (?<=...) >> construct, and it is especially useful in
822situations where you want to efficiently remove something following
823something else in a string. For instance
824
825 s/(foo)bar/$1/g;
826
827can be rewritten as the much more efficient
828
829 s/foo\Kbar//g;
830
5a964f20 831=item C<(?<!pattern)>
d74e8afc 832X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 833
19799a22
GS
834A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
835matches any occurrence of "foo" that does not follow "bar". Works
836only for fixed-width look-behind.
c277df42 837
ee9b8eae
YO
838=back
839
81714fb9
YO
840=item C<(?'NAME'pattern)>
841
842=item C<< (?<NAME>pattern) >>
843X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
844
845A named capture buffer. Identical in every respect to normal capturing
90a18110
RGS
846parentheses C<()> but for the additional fact that C<%+> or C<%-> may be
847used after a successful match to refer to a named buffer. See C<perlvar>
848for more details on the C<%+> and C<%-> hashes.
81714fb9
YO
849
850If multiple distinct capture buffers have the same name then the
851$+{NAME} will refer to the leftmost defined buffer in the match.
852
0d017f4d 853The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
854
855B<NOTE:> While the notation of this construct is the same as the similar
0d017f4d 856function in .NET regexes, the behavior is not. In Perl the buffers are
81714fb9
YO
857numbered sequentially regardless of being named or not. Thus in the
858pattern
859
860 /(x)(?<foo>y)(z)/
861
862$+{foo} will be the same as $2, and $3 will contain 'z' instead of
863the opposite which is what a .NET regex hacker might expect.
864
1f1031fe
YO
865Currently NAME is restricted to simple identifiers only.
866In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
867its Unicode extension (see L<utf8>),
868though it isn't extended by the locale (see L<perllocale>).
81714fb9 869
1f1031fe 870B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 871with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 872may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 873support the use of single quotes as a delimiter for the name.
81714fb9 874
1f1031fe
YO
875=item C<< \k<NAME> >>
876
877=item C<< \k'NAME' >>
81714fb9
YO
878
879Named backreference. Similar to numeric backreferences, except that
880the group is designated by name and not number. If multiple groups
881have the same name then it refers to the leftmost defined group in
882the current match.
883
0d017f4d 884It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
885earlier in the pattern.
886
887Both forms are equivalent.
888
1f1031fe 889B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 890with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 891may be used instead of C<< \k<NAME> >>.
1f1031fe 892
cc6b7395 893=item C<(?{ code })>
d74e8afc 894X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 895
19799a22 896B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
897experimental, and may be changed without notice. Code executed that
898has side effects may not perform identically from version to version
899due to the effect of future optimisations in the regex engine.
c277df42 900
cc46d5f2 901This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
902always succeeds, and its C<code> is not interpolated. Currently,
903the rules to determine where the C<code> ends are somewhat convoluted.
904
77ea4f6d
JV
905This feature can be used together with the special variable C<$^N> to
906capture the results of submatches in variables without having to keep
907track of the number of nested parentheses. For example:
908
909 $_ = "The brown fox jumps over the lazy dog";
910 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
911 print "color = $color, animal = $animal\n";
912
754091cb
RGS
913Inside the C<(?{...})> block, C<$_> refers to the string the regular
914expression is matching against. You can also use C<pos()> to know what is
fa11829f 915the current position of matching within this string.
754091cb 916
19799a22
GS
917The C<code> is properly scoped in the following sense: If the assertion
918is backtracked (compare L<"Backtracking">), all changes introduced after
919C<local>ization are undone, so that
b9ac3b5b
GS
920
921 $_ = 'a' x 8;
5d458dd8 922 m<
b9ac3b5b
GS
923 (?{ $cnt = 0 }) # Initialize $cnt.
924 (
5d458dd8 925 a
b9ac3b5b
GS
926 (?{
927 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
928 })
5d458dd8 929 )*
b9ac3b5b
GS
930 aaaa
931 (?{ $res = $cnt }) # On success copy to non-localized
932 # location.
933 >x;
934
0d017f4d 935will set C<$res = 4>. Note that after the match, C<$cnt> returns to the globally
14218588 936introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
937are unwound.
938
19799a22
GS
939This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
940switch. If I<not> used in this way, the result of evaluation of
941C<code> is put into the special variable C<$^R>. This happens
942immediately, so C<$^R> can be used from other C<(?{ code })> assertions
943inside the same regular expression.
b9ac3b5b 944
19799a22
GS
945The assignment to C<$^R> above is properly localized, so the old
946value of C<$^R> is restored if the assertion is backtracked; compare
947L<"Backtracking">.
b9ac3b5b 948
19799a22
GS
949For reasons of security, this construct is forbidden if the regular
950expression involves run-time interpolation of variables, unless the
951perilous C<use re 'eval'> pragma has been used (see L<re>), or the
952variables contain results of C<qr//> operator (see
5d458dd8 953L<perlop/"qr/STRING/imosx">).
871b0233 954
0d017f4d 955This restriction is due to the wide-spread and remarkably convenient
19799a22 956custom of using run-time determined strings as patterns. For example:
871b0233
IZ
957
958 $re = <>;
959 chomp $re;
960 $string =~ /$re/;
961
14218588
GS
962Before Perl knew how to execute interpolated code within a pattern,
963this operation was completely safe from a security point of view,
964although it could raise an exception from an illegal pattern. If
965you turn on the C<use re 'eval'>, though, it is no longer secure,
966so you should only do so if you are also using taint checking.
967Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 968compartment. See L<perlsec> for details about both these mechanisms.
871b0233 969
e95d7314
GG
970B<WARNING>: Use of lexical (C<my>) variables in these blocks is
971broken. The result is unpredictable and will make perl unstable. The
972workaround is to use global (C<our>) variables.
973
974B<WARNING>: Because Perl's regex engine is currently not re-entrant,
975interpolated code may not invoke the regex engine either directly with
976C<m//> or C<s///>), or indirectly with functions such as
977C<split>. Invoking the regex engine in these blocks will make perl
978unstable.
8988a1bb 979
14455d6c 980=item C<(??{ code })>
d74e8afc
ITB
981X<(??{})>
982X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 983
19799a22 984B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
985experimental, and may be changed without notice. Code executed that
986has side effects may not perform identically from version to version
987due to the effect of future optimisations in the regex engine.
0f5d15d6 988
19799a22
GS
989This is a "postponed" regular subexpression. The C<code> is evaluated
990at run time, at the moment this subexpression may match. The result
991of evaluation is considered as a regular expression and matched as
61528107 992if it were inserted instead of this construct. Note that this means
6bda09f9
YO
993that the contents of capture buffers defined inside an eval'ed pattern
994are not available outside of the pattern, and vice versa, there is no
995way for the inner pattern to refer to a capture buffer defined outside.
996Thus,
997
998 ('a' x 100)=~/(??{'(.)' x 100})/
999
81714fb9 1000B<will> match, it will B<not> set $1.
0f5d15d6 1001
428594d9 1002The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
1003where the C<code> ends are currently somewhat convoluted.
1004
1005The following pattern matches a parenthesized group:
0f5d15d6
IZ
1006
1007 $re = qr{
1008 \(
1009 (?:
1010 (?> [^()]+ ) # Non-parens without backtracking
1011 |
14455d6c 1012 (??{ $re }) # Group with matching parens
0f5d15d6
IZ
1013 )*
1014 \)
1015 }x;
1016
6bda09f9
YO
1017See also C<(?PARNO)> for a different, more efficient way to accomplish
1018the same task.
1019
5d458dd8 1020Because perl's regex engine is not currently re-entrant, delayed
8988a1bb
DD
1021code may not invoke the regex engine either directly with C<m//> or C<s///>),
1022or indirectly with functions such as C<split>.
1023
5d458dd8
YO
1024Recursing deeper than 50 times without consuming any input string will
1025result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
1026changing it requires a custom build.
1027
542fa716
YO
1028=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1029X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1030X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1031X<regex, relative recursion>
6bda09f9 1032
81714fb9
YO
1033Similar to C<(??{ code })> except it does not involve compiling any code,
1034instead it treats the contents of a capture buffer as an independent
61528107 1035pattern that must match at the current position. Capture buffers
81714fb9 1036contained by the pattern will have the value as determined by the
6bda09f9
YO
1037outermost recursion.
1038
894be9b7
YO
1039PARNO is a sequence of digits (not starting with 0) whose value reflects
1040the paren-number of the capture buffer to recurse to. C<(?R)> recurses to
1041the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716
YO
1042C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
1043to be relative, with negative numbers indicating preceding capture buffers
1044and positive ones following. Thus C<(?-1)> refers to the most recently
1045declared buffer, and C<(?+1)> indicates the next buffer to be declared.
c74340f9
YO
1046Note that the counting for relative recursion differs from that of
1047relative backreferences, in that with recursion unclosed buffers B<are>
1048included.
6bda09f9 1049
81714fb9 1050The following pattern matches a function foo() which may contain
f145b7e9 1051balanced parentheses as the argument.
6bda09f9
YO
1052
1053 $re = qr{ ( # paren group 1 (full function)
81714fb9 1054 foo
6bda09f9
YO
1055 ( # paren group 2 (parens)
1056 \(
1057 ( # paren group 3 (contents of parens)
1058 (?:
1059 (?> [^()]+ ) # Non-parens without backtracking
1060 |
1061 (?2) # Recurse to start of paren group 2
1062 )*
1063 )
1064 \)
1065 )
1066 )
1067 }x;
1068
1069If the pattern was used as follows
1070
1071 'foo(bar(baz)+baz(bop))'=~/$re/
1072 and print "\$1 = $1\n",
1073 "\$2 = $2\n",
1074 "\$3 = $3\n";
1075
1076the output produced should be the following:
1077
1078 $1 = foo(bar(baz)+baz(bop))
1079 $2 = (bar(baz)+baz(bop))
81714fb9 1080 $3 = bar(baz)+baz(bop)
6bda09f9 1081
81714fb9 1082If there is no corresponding capture buffer defined, then it is a
61528107 1083fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1084string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1085into perl, so changing it requires a custom build.
1086
542fa716
YO
1087The following shows how using negative indexing can make it
1088easier to embed recursive patterns inside of a C<qr//> construct
1089for later use:
1090
1091 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1092 if (/foo $parens \s+ + \s+ bar $parens/x) {
1093 # do something here...
1094 }
1095
81714fb9 1096B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1097PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1098a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1099as atomic. Also, modifiers are resolved at compile time, so constructs
1100like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1101be processed.
6bda09f9 1102
894be9b7
YO
1103=item C<(?&NAME)>
1104X<(?&NAME)>
1105
0d017f4d
WL
1106Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1107parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1108the same name, then it recurses to the leftmost.
1109
1110It is an error to refer to a name that is not declared somewhere in the
1111pattern.
1112
1f1031fe
YO
1113B<NOTE:> In order to make things easier for programmers with experience
1114with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1115may be used instead of C<< (?&NAME) >>.
1f1031fe 1116
e2e6a0f1
YO
1117=item C<(?(condition)yes-pattern|no-pattern)>
1118X<(?()>
286f584a 1119
e2e6a0f1 1120=item C<(?(condition)yes-pattern)>
286f584a 1121
e2e6a0f1
YO
1122Conditional expression. C<(condition)> should be either an integer in
1123parentheses (which is valid if the corresponding pair of parentheses
1124matched), a look-ahead/look-behind/evaluate zero-width assertion, a
1125name in angle brackets or single quotes (which is valid if a buffer
1126with the given name matched), or the special symbol (R) (true when
1127evaluated inside of recursion or eval). Additionally the R may be
1128followed by a number, (which will be true when evaluated when recursing
1129inside of the appropriate group), or by C<&NAME>, in which case it will
1130be true only when evaluated during recursion in the named group.
1131
1132Here's a summary of the possible predicates:
1133
1134=over 4
1135
1136=item (1) (2) ...
1137
1138Checks if the numbered capturing buffer has matched something.
1139
1140=item (<NAME>) ('NAME')
1141
1142Checks if a buffer with the given name has matched something.
1143
1144=item (?{ CODE })
1145
1146Treats the code block as the condition.
1147
1148=item (R)
1149
1150Checks if the expression has been evaluated inside of recursion.
1151
1152=item (R1) (R2) ...
1153
1154Checks if the expression has been evaluated while executing directly
1155inside of the n-th capture group. This check is the regex equivalent of
1156
1157 if ((caller(0))[3] eq 'subname') { ... }
1158
1159In other words, it does not check the full recursion stack.
1160
1161=item (R&NAME)
1162
1163Similar to C<(R1)>, this predicate checks to see if we're executing
1164directly inside of the leftmost group with a given name (this is the same
1165logic used by C<(?&NAME)> to disambiguate). It does not check the full
1166stack, but only the name of the innermost active recursion.
1167
1168=item (DEFINE)
1169
1170In this case, the yes-pattern is never directly executed, and no
1171no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1172See below for details.
1173
1174=back
1175
1176For example:
1177
1178 m{ ( \( )?
1179 [^()]+
1180 (?(1) \) )
1181 }x
1182
1183matches a chunk of non-parentheses, possibly included in parentheses
1184themselves.
1185
1186A special form is the C<(DEFINE)> predicate, which never executes directly
1187its yes-pattern, and does not allow a no-pattern. This allows to define
1188subpatterns which will be executed only by using the recursion mechanism.
1189This way, you can define a set of regular expression rules that can be
1190bundled into any pattern you choose.
1191
1192It is recommended that for this usage you put the DEFINE block at the
1193end of the pattern, and that you name any subpatterns defined within it.
1194
1195Also, it's worth noting that patterns defined this way probably will
1196not be as efficient, as the optimiser is not very clever about
1197handling them.
1198
1199An example of how this might be used is as follows:
1200
2bf803e2 1201 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1202 (?(DEFINE)
2bf803e2
YO
1203 (?<NAME_PAT>....)
1204 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1205 )/x
1206
1207Note that capture buffers matched inside of recursion are not accessible
0d017f4d 1208after the recursion returns, so the extra layer of capturing buffers is
e2e6a0f1
YO
1209necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1210C<$+{NAME}> would be.
286f584a 1211
c47ff5f1 1212=item C<< (?>pattern) >>
6bda09f9 1213X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1214
19799a22
GS
1215An "independent" subexpression, one which matches the substring
1216that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1217position, and it matches I<nothing other than this substring>. This
19799a22
GS
1218construct is useful for optimizations of what would otherwise be
1219"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1220It may also be useful in places where the "grab all you can, and do not
1221give anything back" semantic is desirable.
19799a22 1222
c47ff5f1 1223For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1224(anchored at the beginning of string, as above) will match I<all>
1225characters C<a> at the beginning of string, leaving no C<a> for
1226C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1227since the match of the subgroup C<a*> is influenced by the following
1228group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1229C<a*ab> will match fewer characters than a standalone C<a*>, since
1230this makes the tail match.
1231
c47ff5f1 1232An effect similar to C<< (?>pattern) >> may be achieved by writing
19799a22
GS
1233C<(?=(pattern))\1>. This matches the same substring as a standalone
1234C<a+>, and the following C<\1> eats the matched string; it therefore
c47ff5f1 1235makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1236(The difference between these two constructs is that the second one
1237uses a capturing group, thus shifting ordinals of backreferences
1238in the rest of a regular expression.)
1239
1240Consider this pattern:
c277df42 1241
871b0233 1242 m{ \(
e2e6a0f1
YO
1243 (
1244 [^()]+ # x+
1245 |
871b0233
IZ
1246 \( [^()]* \)
1247 )+
e2e6a0f1 1248 \)
871b0233 1249 }x
5a964f20 1250
19799a22
GS
1251That will efficiently match a nonempty group with matching parentheses
1252two levels deep or less. However, if there is no such group, it
1253will take virtually forever on a long string. That's because there
1254are so many different ways to split a long string into several
1255substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1256to a subpattern of the above pattern. Consider how the pattern
1257above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1258seconds, but that each extra letter doubles this time. This
1259exponential performance will make it appear that your program has
14218588 1260hung. However, a tiny change to this pattern
5a964f20 1261
e2e6a0f1
YO
1262 m{ \(
1263 (
1264 (?> [^()]+ ) # change x+ above to (?> x+ )
1265 |
871b0233
IZ
1266 \( [^()]* \)
1267 )+
e2e6a0f1 1268 \)
871b0233 1269 }x
c277df42 1270
c47ff5f1 1271which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1272this yourself would be a productive exercise), but finishes in a fourth
1273the time when used on a similar string with 1000000 C<a>s. Be aware,
1274however, that this pattern currently triggers a warning message under
9f1b1f2d 1275the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1276C<"matches null string many times in regex">.
c277df42 1277
c47ff5f1 1278On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1279effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1280This was only 4 times slower on a string with 1000000 C<a>s.
1281
9da458fc
IZ
1282The "grab all you can, and do not give anything back" semantic is desirable
1283in many situations where on the first sight a simple C<()*> looks like
1284the correct solution. Suppose we parse text with comments being delimited
1285by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1286its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1287the comment delimiter, because it may "give up" some whitespace if
1288the remainder of the pattern can be made to match that way. The correct
1289answer is either one of these:
1290
1291 (?>#[ \t]*)
1292 #[ \t]*(?![ \t])
1293
1294For example, to grab non-empty comments into $1, one should use either
1295one of these:
1296
1297 / (?> \# [ \t]* ) ( .+ ) /x;
1298 / \# [ \t]* ( [^ \t] .* ) /x;
1299
1300Which one you pick depends on which of these expressions better reflects
1301the above specification of comments.
1302
6bda09f9
YO
1303In some literature this construct is called "atomic matching" or
1304"possessive matching".
1305
b9b4dddf
YO
1306Possessive quantifiers are equivalent to putting the item they are applied
1307to inside of one of these constructs. The following equivalences apply:
1308
1309 Quantifier Form Bracketing Form
1310 --------------- ---------------
1311 PAT*+ (?>PAT*)
1312 PAT++ (?>PAT+)
1313 PAT?+ (?>PAT?)
1314 PAT{min,max}+ (?>PAT{min,max})
1315
e2e6a0f1
YO
1316=back
1317
1318=head2 Special Backtracking Control Verbs
1319
1320B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1321removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1322be noted to avoid problems during upgrades.
1323
1324These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1325otherwise stated the ARG argument is optional; in some cases, it is
1326forbidden.
1327
1328Any pattern containing a special backtracking verb that allows an argument
1329has the special behaviour that when executed it sets the current packages'
5d458dd8
YO
1330C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1331rules apply:
e2e6a0f1 1332
5d458dd8
YO
1333On failure, the C<$REGERROR> variable will be set to the ARG value of the
1334verb pattern, if the verb was involved in the failure of the match. If the
1335ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1336name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1337none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1338
5d458dd8
YO
1339On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1340the C<$REGMARK> variable will be set to the name of the last
1341C<(*MARK:NAME)> pattern executed. See the explanation for the
1342C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1343
5d458dd8
YO
1344B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1345and most other regex related variables. They are not local to a scope, nor
1346readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1347Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1348
1349If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1350argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1
YO
1351
1352=over 4
1353
1354=item Verbs that take an argument
1355
1356=over 4
1357
5d458dd8 1358=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1359X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1360
5d458dd8
YO
1361This zero-width pattern prunes the backtracking tree at the current point
1362when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1363where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1364A may backtrack as necessary to match. Once it is reached, matching
1365continues in B, which may also backtrack as necessary; however, should B
1366not match, then no further backtracking will take place, and the pattern
1367will fail outright at the current starting position.
54612592
YO
1368
1369The following example counts all the possible matching strings in a
1370pattern (without actually matching any of them).
1371
e2e6a0f1 1372 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1373 print "Count=$count\n";
1374
1375which produces:
1376
1377 aaab
1378 aaa
1379 aa
1380 a
1381 aab
1382 aa
1383 a
1384 ab
1385 a
1386 Count=9
1387
5d458dd8 1388If we add a C<(*PRUNE)> before the count like the following
54612592 1389
5d458dd8 1390 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1391 print "Count=$count\n";
1392
1393we prevent backtracking and find the count of the longest matching
353c6505 1394at each matching starting point like so:
54612592
YO
1395
1396 aaab
1397 aab
1398 ab
1399 Count=3
1400
5d458dd8 1401Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1402
5d458dd8
YO
1403See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1404control backtracking. In some cases, the use of C<(*PRUNE)> can be
1405replaced with a C<< (?>pattern) >> with no functional difference; however,
1406C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1407C<< (?>pattern) >> alone.
54612592 1408
e2e6a0f1 1409
5d458dd8
YO
1410=item C<(*SKIP)> C<(*SKIP:NAME)>
1411X<(*SKIP)>
e2e6a0f1 1412
5d458dd8 1413This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1414failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1415to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1416of this pattern. This effectively means that the regex engine "skips" forward
1417to this position on failure and tries to match again, (assuming that
1418there is sufficient room to match).
1419
1420The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1421C<(*MARK:NAME)> was encountered while matching, then it is that position
1422which is used as the "skip point". If no C<(*MARK)> of that name was
1423encountered, then the C<(*SKIP)> operator has no effect. When used
1424without a name the "skip point" is where the match point was when
1425executing the (*SKIP) pattern.
1426
1427Compare the following to the examples in C<(*PRUNE)>, note the string
24b23f37
YO
1428is twice as long:
1429
5d458dd8 1430 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1431 print "Count=$count\n";
1432
1433outputs
1434
1435 aaab
1436 aaab
1437 Count=2
1438
5d458dd8 1439Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1440executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1441C<(*SKIP)> was executed.
1442
5d458dd8
YO
1443=item C<(*MARK:NAME)> C<(*:NAME)>
1444X<(*MARK)> C<(*MARK:NAME)> C<(*:NAME)>
1445
1446This zero-width pattern can be used to mark the point reached in a string
1447when a certain part of the pattern has been successfully matched. This
1448mark may be given a name. A later C<(*SKIP)> pattern will then skip
1449forward to that point if backtracked into on failure. Any number of
1450C<(*MARK)> patterns are allowed, and the NAME portion is optional and may
1451be duplicated.
1452
1453In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1454can be used to "label" a pattern branch, so that after matching, the
1455program can determine which branches of the pattern were involved in the
1456match.
1457
1458When a match is successful, the C<$REGMARK> variable will be set to the
1459name of the most recently executed C<(*MARK:NAME)> that was involved
1460in the match.
1461
1462This can be used to determine which branch of a pattern was matched
c62285ac 1463without using a separate capture buffer for each branch, which in turn
5d458dd8
YO
1464can result in a performance improvement, as perl cannot optimize
1465C</(?:(x)|(y)|(z))/> as efficiently as something like
1466C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1467
1468When a match has failed, and unless another verb has been involved in
1469failing the match and has provided its own name to use, the C<$REGERROR>
1470variable will be set to the name of the most recently executed
1471C<(*MARK:NAME)>.
1472
1473See C<(*SKIP)> for more details.
1474
b62d2d15
YO
1475As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1476
5d458dd8
YO
1477=item C<(*THEN)> C<(*THEN:NAME)>
1478
241e7389 1479This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1480C<(*PRUNE)>, this verb always matches, and when backtracked into on
1481failure, it causes the regex engine to try the next alternation in the
1482innermost enclosing group (capturing or otherwise).
1483
1484Its name comes from the observation that this operation combined with the
1485alternation operator (C<|>) can be used to create what is essentially a
1486pattern-based if/then/else block:
1487
1488 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1489
1490Note that if this operator is used and NOT inside of an alternation then
1491it acts exactly like the C<(*PRUNE)> operator.
1492
1493 / A (*PRUNE) B /
1494
1495is the same as
1496
1497 / A (*THEN) B /
1498
1499but
1500
1501 / ( A (*THEN) B | C (*THEN) D ) /
1502
1503is not the same as
1504
1505 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1506
1507as after matching the A but failing on the B the C<(*THEN)> verb will
1508backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1509
e2e6a0f1
YO
1510=item C<(*COMMIT)>
1511X<(*COMMIT)>
24b23f37 1512
241e7389 1513This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1514zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1515into on failure it causes the match to fail outright. No further attempts
1516to find a valid match by advancing the start pointer will occur again.
1517For example,
24b23f37 1518
e2e6a0f1 1519 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1520 print "Count=$count\n";
1521
1522outputs
1523
1524 aaab
1525 Count=1
1526
e2e6a0f1
YO
1527In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1528does not match, the regex engine will not try any further matching on the
1529rest of the string.
c277df42 1530
e2e6a0f1 1531=back
9af228c6 1532
e2e6a0f1 1533=item Verbs without an argument
9af228c6
YO
1534
1535=over 4
1536
e2e6a0f1
YO
1537=item C<(*FAIL)> C<(*F)>
1538X<(*FAIL)> X<(*F)>
9af228c6 1539
e2e6a0f1
YO
1540This pattern matches nothing and always fails. It can be used to force the
1541engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1542fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1543
e2e6a0f1 1544It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1545
e2e6a0f1
YO
1546=item C<(*ACCEPT)>
1547X<(*ACCEPT)>
9af228c6 1548
e2e6a0f1
YO
1549B<WARNING:> This feature is highly experimental. It is not recommended
1550for production code.
9af228c6 1551
e2e6a0f1
YO
1552This pattern matches nothing and causes the end of successful matching at
1553the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1554whether there is actually more to match in the string. When inside of a
0d017f4d 1555nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1556via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1557
e2e6a0f1
YO
1558If the C<(*ACCEPT)> is inside of capturing buffers then the buffers are
1559marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1560For instance:
9af228c6 1561
e2e6a0f1 1562 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1563
e2e6a0f1 1564will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0d017f4d 1565be set. If another branch in the inner parentheses were matched, such as in the
e2e6a0f1 1566string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1567
1568=back
c277df42 1569
a0d0e21e
LW
1570=back
1571
c07a80fd 1572=head2 Backtracking
d74e8afc 1573X<backtrack> X<backtracking>
c07a80fd 1574
35a734be
IZ
1575NOTE: This section presents an abstract approximation of regular
1576expression behavior. For a more rigorous (and complicated) view of
1577the rules involved in selecting a match among possible alternatives,
0d017f4d 1578see L<Combining RE Pieces>.
35a734be 1579
c277df42 1580A fundamental feature of regular expression matching involves the
5a964f20 1581notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1582by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1583C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1584internally, but the general principle outlined here is valid.
c07a80fd
PP
1585
1586For a regular expression to match, the I<entire> regular expression must
1587match, not just part of it. So if the beginning of a pattern containing a
1588quantifier succeeds in a way that causes later parts in the pattern to
1589fail, the matching engine backs up and recalculates the beginning
1590part--that's why it's called backtracking.
1591
1592Here is an example of backtracking: Let's say you want to find the
1593word following "foo" in the string "Food is on the foo table.":
1594
1595 $_ = "Food is on the foo table.";
1596 if ( /\b(foo)\s+(\w+)/i ) {
1597 print "$2 follows $1.\n";
1598 }
1599
1600When the match runs, the first part of the regular expression (C<\b(foo)>)
1601finds a possible match right at the beginning of the string, and loads up
1602$1 with "Foo". However, as soon as the matching engine sees that there's
1603no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1604mistake and starts over again one character after where it had the
c07a80fd
PP
1605tentative match. This time it goes all the way until the next occurrence
1606of "foo". The complete regular expression matches this time, and you get
1607the expected output of "table follows foo."
1608
1609Sometimes minimal matching can help a lot. Imagine you'd like to match
1610everything between "foo" and "bar". Initially, you write something
1611like this:
1612
1613 $_ = "The food is under the bar in the barn.";
1614 if ( /foo(.*)bar/ ) {
1615 print "got <$1>\n";
1616 }
1617
1618Which perhaps unexpectedly yields:
1619
1620 got <d is under the bar in the >
1621
1622That's because C<.*> was greedy, so you get everything between the
14218588 1623I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1624to use minimal matching to make sure you get the text between a "foo"
1625and the first "bar" thereafter.
1626
1627 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1628 got <d is under the >
1629
0d017f4d 1630Here's another example. Let's say you'd like to match a number at the end
b6e13d97 1631of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
1632So you write this:
1633
1634 $_ = "I have 2 numbers: 53147";
1635 if ( /(.*)(\d*)/ ) { # Wrong!
1636 print "Beginning is <$1>, number is <$2>.\n";
1637 }
1638
1639That won't work at all, because C<.*> was greedy and gobbled up the
1640whole string. As C<\d*> can match on an empty string the complete
1641regular expression matched successfully.
1642
8e1088bc 1643 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
1644
1645Here are some variants, most of which don't work:
1646
1647 $_ = "I have 2 numbers: 53147";
1648 @pats = qw{
1649 (.*)(\d*)
1650 (.*)(\d+)
1651 (.*?)(\d*)
1652 (.*?)(\d+)
1653 (.*)(\d+)$
1654 (.*?)(\d+)$
1655 (.*)\b(\d+)$
1656 (.*\D)(\d+)$
1657 };
1658
1659 for $pat (@pats) {
1660 printf "%-12s ", $pat;
1661 if ( /$pat/ ) {
1662 print "<$1> <$2>\n";
1663 } else {
1664 print "FAIL\n";
1665 }
1666 }
1667
1668That will print out:
1669
1670 (.*)(\d*) <I have 2 numbers: 53147> <>
1671 (.*)(\d+) <I have 2 numbers: 5314> <7>
1672 (.*?)(\d*) <> <>
1673 (.*?)(\d+) <I have > <2>
1674 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1675 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1676 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1677 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1678
1679As you see, this can be a bit tricky. It's important to realize that a
1680regular expression is merely a set of assertions that gives a definition
1681of success. There may be 0, 1, or several different ways that the
1682definition might succeed against a particular string. And if there are
5a964f20
TC
1683multiple ways it might succeed, you need to understand backtracking to
1684know which variety of success you will achieve.
c07a80fd 1685
19799a22 1686When using look-ahead assertions and negations, this can all get even
8b19b778 1687trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
1688followed by "123". You might try to write that as
1689
871b0233
IZ
1690 $_ = "ABC123";
1691 if ( /^\D*(?!123)/ ) { # Wrong!
1692 print "Yup, no 123 in $_\n";
1693 }
c07a80fd
PP
1694
1695But that isn't going to match; at least, not the way you're hoping. It
1696claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1697why that pattern matches, contrary to popular expectations:
c07a80fd 1698
4358a253
SS
1699 $x = 'ABC123';
1700 $y = 'ABC445';
c07a80fd 1701
4358a253
SS
1702 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1703 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1704
4358a253
SS
1705 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1706 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
1707
1708This prints
1709
1710 2: got ABC
1711 3: got AB
1712 4: got ABC
1713
5f05dabc 1714You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
1715general purpose version of test 1. The important difference between
1716them is that test 3 contains a quantifier (C<\D*>) and so can use
1717backtracking, whereas test 1 will not. What's happening is
1718that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1719non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1720let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1721fail.
14218588 1722
c07a80fd 1723The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1724try to match C<(?!123> with "123", which fails. But because
c07a80fd
PP
1725a quantifier (C<\D*>) has been used in the regular expression, the
1726search engine can backtrack and retry the match differently
54310121 1727in the hope of matching the complete regular expression.
c07a80fd 1728
5a964f20
TC
1729The pattern really, I<really> wants to succeed, so it uses the
1730standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1731time. Now there's indeed something following "AB" that is not
14218588 1732"123". It's "C123", which suffices.
c07a80fd 1733
14218588
GS
1734We can deal with this by using both an assertion and a negation.
1735We'll say that the first part in $1 must be followed both by a digit
1736and by something that's not "123". Remember that the look-aheads
1737are zero-width expressions--they only look, but don't consume any
1738of the string in their match. So rewriting this way produces what
c07a80fd
PP
1739you'd expect; that is, case 5 will fail, but case 6 succeeds:
1740
4358a253
SS
1741 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1742 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
1743
1744 6: got ABC
1745
5a964f20 1746In other words, the two zero-width assertions next to each other work as though
19799a22 1747they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
1748matches only if you're at the beginning of the line AND the end of the
1749line simultaneously. The deeper underlying truth is that juxtaposition in
1750regular expressions always means AND, except when you write an explicit OR
1751using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1752although the attempted matches are made at different positions because "a"
1753is not a zero-width assertion, but a one-width assertion.
1754
0d017f4d 1755B<WARNING>: Particularly complicated regular expressions can take
14218588 1756exponential time to solve because of the immense number of possible
0d017f4d 1757ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
1758internal optimizations done by the regular expression engine, this will
1759take a painfully long time to run:
c07a80fd 1760
e1901655
IZ
1761 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1762
1763And if you used C<*>'s in the internal groups instead of limiting them
1764to 0 through 5 matches, then it would take forever--or until you ran
1765out of stack space. Moreover, these internal optimizations are not
1766always applicable. For example, if you put C<{0,5}> instead of C<*>
1767on the external group, no current optimization is applicable, and the
1768match takes a long time to finish.
c07a80fd 1769
9da458fc
IZ
1770A powerful tool for optimizing such beasts is what is known as an
1771"independent group",
c47ff5f1 1772which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1773zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 1774the tail match, since they are in "logical" context: only
14218588 1775whether they match is considered relevant. For an example
9da458fc 1776where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1777following match, see L<C<< (?>pattern) >>>.
c277df42 1778
a0d0e21e 1779=head2 Version 8 Regular Expressions
d74e8afc 1780X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1781
5a964f20 1782In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1783routines, here are the pattern-matching rules not described above.
1784
54310121 1785Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1786with a special meaning described here or above. You can cause
5a964f20 1787characters that normally function as metacharacters to be interpreted
5f05dabc 1788literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
1789character; "\\" matches a "\"). This escape mechanism is also required
1790for the character used as the pattern delimiter.
1791
1792A series of characters matches that series of characters in the target
1793string, so the pattern C<blurfl> would match "blurfl" in the target
1794string.
a0d0e21e
LW
1795
1796You can specify a character class, by enclosing a list of characters
5d458dd8 1797in C<[]>, which will match any character from the list. If the
a0d0e21e 1798first character after the "[" is "^", the class matches any character not
14218588 1799in the list. Within a list, the "-" character specifies a
5a964f20 1800range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1801inclusive. If you want either "-" or "]" itself to be a member of a
1802class, put it at the start of the list (possibly after a "^"), or
1803escape it with a backslash. "-" is also taken literally when it is
1804at the end of the list, just before the closing "]". (The
84850974
DD
1805following all specify the same class of three characters: C<[-az]>,
1806C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
1807specifies a class containing twenty-six characters, even on EBCDIC-based
1808character sets.) Also, if you try to use the character
1809classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1810a range, the "-" is understood literally.
a0d0e21e 1811
8ada0baa
JH
1812Note also that the whole range idea is rather unportable between
1813character sets--and even within character sets they may cause results
1814you probably didn't expect. A sound principle is to use only ranges
0d017f4d 1815that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
1816[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1817spell out the character sets in full.
1818
54310121 1819Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1820used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1821"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
5d458dd8
YO
1822of octal digits, matches the character whose coded character set value
1823is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
1824matches the character whose numeric value is I<nn>. The expression \cI<x>
1825matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 1826matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1827
1828You can specify a series of alternatives for a pattern using "|" to
1829separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1830or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1831first alternative includes everything from the last pattern delimiter
1832("(", "[", or the beginning of the pattern) up to the first "|", and
1833the last alternative contains everything from the last "|" to the next
14218588
GS
1834pattern delimiter. That's why it's common practice to include
1835alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1836start and end.
1837
5a964f20 1838Alternatives are tried from left to right, so the first
a3cb178b
GS
1839alternative found for which the entire expression matches, is the one that
1840is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1841example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1842part will match, as that is the first alternative tried, and it successfully
1843matches the target string. (This might not seem important, but it is
1844important when you are capturing matched text using parentheses.)
1845
5a964f20 1846Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1847so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1848
14218588
GS
1849Within a pattern, you may designate subpatterns for later reference
1850by enclosing them in parentheses, and you may refer back to the
1851I<n>th subpattern later in the pattern using the metacharacter
1852\I<n>. Subpatterns are numbered based on the left to right order
1853of their opening parenthesis. A backreference matches whatever
1854actually matched the subpattern in the string being examined, not
1855the rules for that subpattern. Therefore, C<(0|0x)\d*\s\1\d*> will
1856match "0x1234 0x4321", but not "0x1234 01234", because subpattern
18571 matched "0x", even though the rule C<0|0x> could potentially match
1858the leading 0 in the second number.
cb1a09d0 1859
0d017f4d 1860=head2 Warning on \1 Instead of $1
cb1a09d0 1861
5a964f20 1862Some people get too used to writing things like:
cb1a09d0
AD
1863
1864 $pattern =~ s/(\W)/\\\1/g;
1865
1866This is grandfathered for the RHS of a substitute to avoid shocking the
1867B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1868PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1869the usual double-quoted string means a control-A. The customary Unix
1870meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1871of doing that, you get yourself into trouble if you then add an C</e>
1872modifier.
1873
5a964f20 1874 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1875
1876Or if you try to do
1877
1878 s/(\d+)/\1000/;
1879
1880You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1881C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1882with the operation of matching a backreference. Certainly they mean two
1883different things on the I<left> side of the C<s///>.
9fa51da4 1884
0d017f4d 1885=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 1886
19799a22 1887B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1888
1889Regular expressions provide a terse and powerful programming language. As
1890with most other power tools, power comes together with the ability
1891to wreak havoc.
1892
1893A common abuse of this power stems from the ability to make infinite
628afcb5 1894loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1895
1896 'foo' =~ m{ ( o? )* }x;
1897
0d017f4d 1898The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 1899in the string is not moved by the match, C<o?> would match again and again
527e91da 1900because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
1901is with the looping modifier C<//g>:
1902
1903 @matches = ( 'foo' =~ m{ o? }xg );
1904
1905or
1906
1907 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1908
1909or the loop implied by split().
1910
1911However, long experience has shown that many programming tasks may
14218588
GS
1912be significantly simplified by using repeated subexpressions that
1913may match zero-length substrings. Here's a simple example being:
c84d73f1
IZ
1914
1915 @chars = split //, $string; # // is not magic in split
1916 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1917
9da458fc 1918Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 1919the infinite loop>. The rules for this are different for lower-level
527e91da 1920loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
1921ones like the C</g> modifier or split() operator.
1922
19799a22
GS
1923The lower-level loops are I<interrupted> (that is, the loop is
1924broken) when Perl detects that a repeated expression matched a
1925zero-length substring. Thus
c84d73f1
IZ
1926
1927 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1928
5d458dd8 1929is made equivalent to
c84d73f1 1930
5d458dd8
YO
1931 m{ (?: NON_ZERO_LENGTH )*
1932 |
1933 (?: ZERO_LENGTH )?
c84d73f1
IZ
1934 }x;
1935
1936The higher level-loops preserve an additional state between iterations:
5d458dd8 1937whether the last match was zero-length. To break the loop, the following
c84d73f1 1938match after a zero-length match is prohibited to have a length of zero.
5d458dd8 1939This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
1940and so the I<second best> match is chosen if the I<best> match is of
1941zero length.
1942
19799a22 1943For example:
c84d73f1
IZ
1944
1945 $_ = 'bar';
1946 s/\w??/<$&>/g;
1947
20fb949f 1948results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 1949match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
1950best> match is what is matched by C<\w>. Thus zero-length matches
1951alternate with one-character-long matches.
1952
5d458dd8 1953Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
1954position one notch further in the string.
1955
19799a22 1956The additional state of being I<matched with zero-length> is associated with
c84d73f1 1957the matched string, and is reset by each assignment to pos().
9da458fc
IZ
1958Zero-length matches at the end of the previous match are ignored
1959during C<split>.
c84d73f1 1960
0d017f4d 1961=head2 Combining RE Pieces
35a734be
IZ
1962
1963Each of the elementary pieces of regular expressions which were described
1964before (such as C<ab> or C<\Z>) could match at most one substring
1965at the given position of the input string. However, in a typical regular
1966expression these elementary pieces are combined into more complicated
1967patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1968(in these examples C<S> and C<T> are regular subexpressions).
1969
1970Such combinations can include alternatives, leading to a problem of choice:
1971if we match a regular expression C<a|ab> against C<"abc">, will it match
1972substring C<"a"> or C<"ab">? One way to describe which substring is
1973actually matched is the concept of backtracking (see L<"Backtracking">).
1974However, this description is too low-level and makes you think
1975in terms of a particular implementation.
1976
1977Another description starts with notions of "better"/"worse". All the
1978substrings which may be matched by the given regular expression can be
1979sorted from the "best" match to the "worst" match, and it is the "best"
1980match which is chosen. This substitutes the question of "what is chosen?"
1981by the question of "which matches are better, and which are worse?".
1982
1983Again, for elementary pieces there is no such question, since at most
1984one match at a given position is possible. This section describes the
1985notion of better/worse for combining operators. In the description
1986below C<S> and C<T> are regular subexpressions.
1987
13a2d996 1988=over 4
35a734be
IZ
1989
1990=item C<ST>
1991
1992Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
1993substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 1994which can be matched by C<T>.
35a734be
IZ
1995
1996If C<A> is better match for C<S> than C<A'>, C<AB> is a better
1997match than C<A'B'>.
1998
1999If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
2000C<B> is better match for C<T> than C<B'>.
2001
2002=item C<S|T>
2003
2004When C<S> can match, it is a better match than when only C<T> can match.
2005
2006Ordering of two matches for C<S> is the same as for C<S>. Similar for
2007two matches for C<T>.
2008
2009=item C<S{REPEAT_COUNT}>
2010
2011Matches as C<SSS...S> (repeated as many times as necessary).
2012
2013=item C<S{min,max}>
2014
2015Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2016
2017=item C<S{min,max}?>
2018
2019Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2020
2021=item C<S?>, C<S*>, C<S+>
2022
2023Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2024
2025=item C<S??>, C<S*?>, C<S+?>
2026
2027Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2028
c47ff5f1 2029=item C<< (?>S) >>
35a734be
IZ
2030
2031Matches the best match for C<S> and only that.
2032
2033=item C<(?=S)>, C<(?<=S)>
2034
2035Only the best match for C<S> is considered. (This is important only if
2036C<S> has capturing parentheses, and backreferences are used somewhere
2037else in the whole regular expression.)
2038
2039=item C<(?!S)>, C<(?<!S)>
2040
2041For this grouping operator there is no need to describe the ordering, since
2042only whether or not C<S> can match is important.
2043
6bda09f9 2044=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2045
2046The ordering is the same as for the regular expression which is
6bda09f9 2047the result of EXPR, or the pattern contained by capture buffer PARNO.
35a734be
IZ
2048
2049=item C<(?(condition)yes-pattern|no-pattern)>
2050
2051Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2052already determined. The ordering of the matches is the same as for the
2053chosen subexpression.
2054
2055=back
2056
2057The above recipes describe the ordering of matches I<at a given position>.
2058One more rule is needed to understand how a match is determined for the
2059whole regular expression: a match at an earlier position is always better
2060than a match at a later position.
2061
0d017f4d 2062=head2 Creating Custom RE Engines
c84d73f1
IZ
2063
2064Overloaded constants (see L<overload>) provide a simple way to extend
2065the functionality of the RE engine.
2066
2067Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2068matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2069characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2070at these positions, so we want to have each C<\Y|> in the place of the
2071more complicated version. We can create a module C<customre> to do
2072this:
2073
2074 package customre;
2075 use overload;
2076
2077 sub import {
2078 shift;
2079 die "No argument to customre::import allowed" if @_;
2080 overload::constant 'qr' => \&convert;
2081 }
2082
2083 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2084
580a9fe1
RGS
2085 # We must also take care of not escaping the legitimate \\Y|
2086 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2087 my %rules = ( '\\' => '\\\\',
c84d73f1
IZ
2088 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
2089 sub convert {
2090 my $re = shift;
5d458dd8 2091 $re =~ s{
c84d73f1
IZ
2092 \\ ( \\ | Y . )
2093 }
5d458dd8 2094 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2095 return $re;
2096 }
2097
2098Now C<use customre> enables the new escape in constant regular
2099expressions, i.e., those without any runtime variable interpolations.
2100As documented in L<overload>, this conversion will work only over
2101literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2102part of this regular expression needs to be converted explicitly
2103(but only if the special meaning of C<\Y|> should be enabled inside $re):
2104
2105 use customre;
2106 $re = <>;
2107 chomp $re;
2108 $re = customre::convert $re;
2109 /\Y|$re\Y|/;
2110
1f1031fe
YO
2111=head1 PCRE/Python Support
2112
99d59c4d 2113As of Perl 5.10.0, Perl supports several Python/PCRE specific extensions
1f1031fe 2114to the regex syntax. While Perl programmers are encouraged to use the
99d59c4d 2115Perl specific syntax, the following are also accepted:
1f1031fe
YO
2116
2117=over 4
2118
ae5648b3 2119=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe
YO
2120
2121Define a named capture buffer. Equivalent to C<< (?<NAME>pattern) >>.
2122
2123=item C<< (?P=NAME) >>
2124
2125Backreference to a named capture buffer. Equivalent to C<< \g{NAME} >>.
2126
2127=item C<< (?P>NAME) >>
2128
2129Subroutine call to a named capture buffer. Equivalent to C<< (?&NAME) >>.
2130
ee9b8eae 2131=back
1f1031fe 2132
19799a22
GS
2133=head1 BUGS
2134
9da458fc
IZ
2135This document varies from difficult to understand to completely
2136and utterly opaque. The wandering prose riddled with jargon is
2137hard to fathom in several places.
2138
2139This document needs a rewrite that separates the tutorial content
2140from the reference content.
19799a22
GS
2141
2142=head1 SEE ALSO
9fa51da4 2143
91e0c79e
MJD
2144L<perlrequick>.
2145
2146L<perlretut>.
2147
9b599b2a
GS
2148L<perlop/"Regexp Quote-Like Operators">.
2149
1e66bd83
PP
2150L<perlop/"Gory details of parsing quoted constructs">.
2151
14218588
GS
2152L<perlfaq6>.
2153
9b599b2a
GS
2154L<perlfunc/pos>.
2155
2156L<perllocale>.
2157
fb55449c
JH
2158L<perlebcdic>.
2159
14218588
GS
2160I<Mastering Regular Expressions> by Jeffrey Friedl, published
2161by O'Reilly and Associates.