This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Document that there are bugs with EBCDIC and regexes
[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 28=over 4
29
54310121 30=item m
d74e8afc 31X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff 32
33Treat string as multiple lines. That is, change "^" and "$" from matching
14218588 34the start or end of the string to matching the start or end of any
7f761169 35line anywhere within the string.
55497cff 36
54310121 37=item s
d74e8afc
ITB
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
55497cff 40
41Treat string as single line. That is, change "." to match any character
19799a22 42whatsoever, even a newline, which normally it would not match.
55497cff 43
34d67d80 44Used together, as C</ms>, they let the "." match any character whatsoever,
fb55449c 45while still allowing "^" and "$" to match, respectively, just after
19799a22 46and just before newlines within the string.
7b8d334a 47
87e95b7f
YO
48=item i
49X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
50X<regular expression, case-insensitive>
51
52Do case-insensitive pattern matching.
53
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 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
7b059540 85the regular expression parser to ignore most whitespace that is neither
55497cff 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 91class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
92escape them (using backslashes or C<\Q...\E>) or encode them using octal,
93hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
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
7651b971
KW
98a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
99whether space interpretation within a single multi-character construct. For
100example in C<\x{...}>, regardless of the C</x> modifier, there can be no
101spaces. Same for a L<quantifier|Quantifiers> such as C<{3}> or
f9e949fd
KW
102C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<?> and C<:>,
103but can between the C<(> and C<?>. Within any delimiters for such a
104construct, allowed spaces are not affected by C</x>, and depend on the
105construct. For example, C<\x{...}> can't have spaces because hexadecimal
106numbers don't have spaces in them. But, Unicode properties can have spaces, so
107in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
108L<perluniprops.pod/Properties accessible through \p{} and \P{}>.
d74e8afc 109X</x>
a0d0e21e
LW
110
111=head2 Regular Expressions
112
04838cea
RGS
113=head3 Metacharacters
114
384f06ae 115The patterns used in Perl pattern matching evolved from those supplied in
14218588 116the Version 8 regex routines. (The routines are derived
19799a22
GS
117(distantly) from Henry Spencer's freely redistributable reimplementation
118of the V8 routines.) See L<Version 8 Regular Expressions> for
119details.
a0d0e21e
LW
120
121In particular the following metacharacters have their standard I<egrep>-ish
122meanings:
d74e8afc
ITB
123X<metacharacter>
124X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
125
a0d0e21e 126
54310121 127 \ Quote the next metacharacter
a0d0e21e
LW
128 ^ Match the beginning of the line
129 . Match any character (except newline)
c07a80fd 130 $ Match the end of the line (or before newline at the end)
a0d0e21e
LW
131 | Alternation
132 () Grouping
133 [] Character class
134
14218588
GS
135By default, the "^" character is guaranteed to match only the
136beginning of the string, the "$" character only the end (or before the
137newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
138assumption that the string contains only one line. Embedded newlines
139will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 140string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
141newline within the string (except if the newline is the last character in
142the string), and "$" will match before any newline. At the
a0d0e21e
LW
143cost of a little more overhead, you can do this by using the /m modifier
144on the pattern match operator. (Older programs did this by setting C<$*>,
f02c194e 145but this practice has been removed in perl 5.9.)
d74e8afc 146X<^> X<$> X</m>
a0d0e21e 147
14218588 148To simplify multi-line substitutions, the "." character never matches a
55497cff 149newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 150the string is a single line--even if it isn't.
d74e8afc 151X<.> X</s>
a0d0e21e 152
04838cea
RGS
153=head3 Quantifiers
154
a0d0e21e 155The following standard quantifiers are recognized:
d74e8afc 156X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e
LW
157
158 * Match 0 or more times
159 + Match 1 or more times
160 ? Match 1 or 0 times
161 {n} Match exactly n times
162 {n,} Match at least n times
163 {n,m} Match at least n but not more than m times
164
165(If a curly bracket occurs in any other context, it is treated
b975c076 166as a regular character. In particular, the lower bound
527e91da
BB
167is not optional.) The "*" quantifier is equivalent to C<{0,}>, the "+"
168quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
9c79236d
GS
169to integral values less than a preset limit defined when perl is built.
170This is usually 32766 on the most common platforms. The actual limit can
171be seen in the error message generated by code such as this:
172
820475bd 173 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 174
54310121 175By default, a quantified subpattern is "greedy", that is, it will match as
176many times as possible (given a particular starting location) while still
177allowing the rest of the pattern to match. If you want it to match the
178minimum number of times possible, follow the quantifier with a "?". Note
179that the meanings don't change, just the "greediness":
0d017f4d 180X<metacharacter> X<greedy> X<greediness>
d74e8afc 181X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 182
0d017f4d
WL
183 *? Match 0 or more times, not greedily
184 +? Match 1 or more times, not greedily
185 ?? Match 0 or 1 time, not greedily
186 {n}? Match exactly n times, not greedily
187 {n,}? Match at least n times, not greedily
188 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 189
b9b4dddf
YO
190By default, when a quantified subpattern does not allow the rest of the
191overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 192sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
193as well.
194
0d017f4d
WL
195 *+ Match 0 or more times and give nothing back
196 ++ Match 1 or more times and give nothing back
197 ?+ Match 0 or 1 time and give nothing back
b9b4dddf 198 {n}+ Match exactly n times and give nothing back (redundant)
04838cea
RGS
199 {n,}+ Match at least n times and give nothing back
200 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
201
202For instance,
203
204 'aaaa' =~ /a++a/
205
206will never match, as the C<a++> will gobble up all the C<a>'s in the
207string and won't leave any for the remaining part of the pattern. This
208feature can be extremely useful to give perl hints about where it
209shouldn't backtrack. For instance, the typical "match a double-quoted
210string" problem can be most efficiently performed when written as:
211
212 /"(?:[^"\\]++|\\.)*+"/
213
0d017f4d 214as we know that if the final quote does not match, backtracking will not
b9b4dddf
YO
215help. See the independent subexpression C<< (?>...) >> for more details;
216possessive quantifiers are just syntactic sugar for that construct. For
217instance the above example could also be written as follows:
218
219 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
220
04838cea
RGS
221=head3 Escape sequences
222
5f05dabc 223Because patterns are processed as double quoted strings, the following
a0d0e21e 224also work:
0d017f4d 225X<\t> X<\n> X<\r> X<\f> X<\e> X<\a> X<\l> X<\u> X<\L> X<\U> X<\E> X<\Q>
d74e8afc 226X<\0> X<\c> X<\N> X<\x>
a0d0e21e 227
0f36ee90 228 \t tab (HT, TAB)
229 \n newline (LF, NL)
230 \r return (CR)
231 \f form feed (FF)
232 \a alarm (bell) (BEL)
233 \e escape (think troff) (ESC)
0d017f4d
WL
234 \033 octal char (example: ESC)
235 \x1B hex char (example: ESC)
196ac2fc 236 \x{263a} long hex char (example: Unicode SMILEY)
0d017f4d 237 \cK control char (example: VT)
196ac2fc 238 \N{name} named Unicode character
e526e8bb 239 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
cb1a09d0
AD
240 \l lowercase next char (think vi)
241 \u uppercase next char (think vi)
242 \L lowercase till \E (think vi)
243 \U uppercase till \E (think vi)
244 \E end case modification (think vi)
5a964f20 245 \Q quote (disable) pattern metacharacters till \E
a0d0e21e 246
a034a98d 247If C<use locale> is in effect, the case map used by C<\l>, C<\L>, C<\u>
423cee85 248and C<\U> is taken from the current locale. See L<perllocale>. For
4a2d328f 249documentation of C<\N{name}>, see L<charnames>.
a034a98d 250
1d2dff63
GS
251You cannot include a literal C<$> or C<@> within a C<\Q> sequence.
252An unescaped C<$> or C<@> interpolates the corresponding variable,
253while escaping will cause the literal string C<\$> to be matched.
254You'll need to write something like C<m/\Quser\E\@\Qhost/>.
255
e1d1eefb 256=head3 Character Classes and other Special Escapes
04838cea 257
a0d0e21e 258In addition, Perl defines the following:
d74e8afc 259X<\w> X<\W> X<\s> X<\S> X<\d> X<\D> X<\X> X<\p> X<\P> X<\C>
f7819f85 260X<\g> X<\k> X<\N> X<\K> X<\v> X<\V> X<\h> X<\H>
0d017f4d 261X<word> X<whitespace> X<character class> X<backreference>
a0d0e21e 262
81714fb9
YO
263 \w Match a "word" character (alphanumeric plus "_")
264 \W Match a non-"word" character
265 \s Match a whitespace character
266 \S Match a non-whitespace character
267 \d Match a digit character
268 \D Match a non-digit character
269 \pP Match P, named property. Use \p{Prop} for longer names.
270 \PP Match non-P
0111a78f 271 \X Match Unicode "eXtended grapheme cluster"
81714fb9
YO
272 \C Match a single C char (octet) even under Unicode.
273 NOTE: breaks up characters into their UTF-8 bytes,
274 so you may end up with malformed pieces of UTF-8.
275 Unsupported in lookbehind.
5d458dd8 276 \1 Backreference to a specific group.
c74340f9 277 '1' may actually be any positive integer.
2bf803e2
YO
278 \g1 Backreference to a specific or previous group,
279 \g{-1} number may be negative indicating a previous buffer and may
280 optionally be wrapped in curly brackets for safer parsing.
1f1031fe 281 \g{name} Named backreference
81714fb9 282 \k<name> Named backreference
ee9b8eae 283 \K Keep the stuff left of the \K, don't include it in $&
b3b85878 284 \N Any character but \n (experimental)
e1d1eefb
YO
285 \v Vertical whitespace
286 \V Not vertical whitespace
287 \h Horizontal whitespace
288 \H Not horizontal whitespace
2ddf2931 289 \R Linebreak
a0d0e21e 290
08ce8fc6
JH
291A C<\w> matches a single alphanumeric character (an alphabetic
292character, or a decimal digit) or C<_>, not a whole word. Use C<\w+>
293to match a string of Perl-identifier characters (which isn't the same
294as matching an English word). If C<use locale> is in effect, the list
295of alphabetic characters generated by C<\w> is taken from the current
296locale. See L<perllocale>. You may use C<\w>, C<\W>, C<\s>, C<\S>,
0d017f4d
WL
297C<\d>, and C<\D> within character classes, but they aren't usable
298as either end of a range. If any of them precedes or follows a "-",
299the "-" is understood literally. If Unicode is in effect, C<\s> matches
c62285ac 300also "\x{85}", "\x{2028}", and "\x{2029}". See L<perlunicode> for more
0d017f4d
WL
301details about C<\pP>, C<\PP>, C<\X> and the possibility of defining
302your own C<\p> and C<\P> properties, and L<perluniintro> about Unicode
303in general.
d74e8afc 304X<\w> X<\W> X<word>
a0d0e21e 305
e1d1eefb 306C<\R> will atomically match a linebreak, including the network line-ending
e2cb52ee 307"\x0D\x0A". Specifically, X<\R> is exactly equivalent to
e1d1eefb
YO
308
309 (?>\x0D\x0A?|[\x0A-\x0C\x85\x{2028}\x{2029}])
310
311B<Note:> C<\R> has no special meaning inside of a character class;
312use C<\v> instead (vertical whitespace).
313X<\R>
314
df225385 315Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
e526e8bb
KW
316character whose name is C<NAME>; and similarly when of the form
317C<\N{U+I<wide hex char>}>, it matches the character whose Unicode ordinal is
318I<wide hex char>. Otherwise it matches any character but C<\n>.
df225385 319
b8c5462f 320The POSIX character class syntax
d74e8afc 321X<character class>
b8c5462f 322
820475bd 323 [:class:]
b8c5462f 324
0d017f4d 325is also available. Note that the C<[> and C<]> brackets are I<literal>;
5496314a
SP
326they must always be used within a character class expression.
327
328 # this is correct:
329 $string =~ /[[:alpha:]]/;
330
331 # this is not, and will generate a warning:
332 $string =~ /[:alpha:]/;
333
6fa80ea2
YO
334The following table shows the mapping of POSIX character class
335names, common escapes, literal escape sequences and their equivalent
336Unicode style property names.
337X<character class> X<\p> X<\p{}>
d74e8afc
ITB
338X<alpha> X<alnum> X<ascii> X<blank> X<cntrl> X<digit> X<graph>
339X<lower> X<print> X<punct> X<space> X<upper> X<word> X<xdigit>
b8c5462f 340
6fa80ea2
YO
341B<Note:> up to Perl 5.10 the property names used were shared with
342standard Unicode properties, this was changed in Perl 5.11, see
343L<perl5110delta> for details.
344
345 POSIX Esc Class Property Note
346 --------------------------------------------------------
347 alnum [0-9A-Za-z] IsPosixAlnum
348 alpha [A-Za-z] IsPosixAlpha
349 ascii [\000-\177] IsASCII
350 blank [\011 ] IsPosixBlank [1]
351 cntrl [\0-\37\177] IsPosixCntrl
352 digit \d [0-9] IsPosixDigit
353 graph [!-~] IsPosixGraph
354 lower [a-z] IsPosixLower
355 print [ -~] IsPosixPrint
356 punct [!-/:-@[-`{-~] IsPosixPunct
357 space [\11-\15 ] IsPosixSpace [2]
358 \s [\11\12\14\15 ] IsPerlSpace [2]
359 upper [A-Z] IsPosixUpper
360 word \w [0-9A-Z_a-z] IsPerlWord [3]
361 xdigit [0-9A-Fa-f] IsXDigit
b8c5462f 362
07698885
RGS
363=over
364
365=item [1]
366
b432a672 367A GNU extension equivalent to C<[ \t]>, "all horizontal whitespace".
07698885
RGS
368
369=item [2]
370
6fa80ea2
YO
371Note that C<\s> and C<[[:space:]]> are B<not> equivalent as C<[[:space:]]>
372includes also the (very rare) "vertical tabulator", "\cK" or chr(11) in
373ASCII.
07698885
RGS
374
375=item [3]
376
08ce8fc6 377A Perl extension, see above.
07698885
RGS
378
379=back
aaa51d5e 380
26b44a0a 381For example use C<[:upper:]> to match all the uppercase characters.
aaa51d5e
JF
382Note that the C<[]> are part of the C<[::]> construct, not part of the
383whole character class. For example:
b8c5462f 384
820475bd 385 [01[:alpha:]%]
b8c5462f 386
0d017f4d 387matches zero, one, any alphabetic character, and the percent sign.
b8c5462f 388
353c6505 389The other named classes are:
b8c5462f
JH
390
391=over 4
392
393=item cntrl
d74e8afc 394X<cntrl>
b8c5462f 395
820475bd
GS
396Any control character. Usually characters that don't produce output as
397such but instead control the terminal somehow: for example newline and
398backspace are control characters. All characters with ord() less than
0d017f4d 39932 are usually classified as control characters (assuming ASCII,
7be5a6cf
JF
400the ISO Latin character sets, and Unicode), as is the character with
401the ord() value of 127 (C<DEL>).
b8c5462f
JH
402
403=item graph
d74e8afc 404X<graph>
b8c5462f 405
f1cbbd6e 406Any alphanumeric or punctuation (special) character.
b8c5462f
JH
407
408=item print
d74e8afc 409X<print>
b8c5462f 410
f79b3095 411Any alphanumeric or punctuation (special) character or the space character.
b8c5462f
JH
412
413=item punct
d74e8afc 414X<punct>
b8c5462f 415
f1cbbd6e 416Any punctuation (special) character.
b8c5462f
JH
417
418=item xdigit
d74e8afc 419X<xdigit>
b8c5462f 420
593df60c 421Any hexadecimal digit. Though this may feel silly ([0-9A-Fa-f] would
820475bd 422work just fine) it is included for completeness.
b8c5462f 423
b8c5462f
JH
424=back
425
426You can negate the [::] character classes by prefixing the class name
427with a '^'. This is a Perl extension. For example:
d74e8afc 428X<character class, negation>
b8c5462f 429
5496314a 430 POSIX traditional Unicode
93733859 431
6fa80ea2
YO
432 [[:^digit:]] \D \P{IsPosixDigit}
433 [[:^space:]] \S \P{IsPosixSpace}
434 [[:^word:]] \W \P{IsPerlWord}
b8c5462f 435
54c18d04
MK
436Perl respects the POSIX standard in that POSIX character classes are
437only supported within a character class. The POSIX character classes
438[.cc.] and [=cc=] are recognized but B<not> supported and trying to
439use them will cause an error.
b8c5462f 440
04838cea
RGS
441=head3 Assertions
442
a0d0e21e 443Perl defines the following zero-width assertions:
d74e8afc
ITB
444X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
445X<regexp, zero-width assertion>
446X<regular expression, zero-width assertion>
447X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e
LW
448
449 \b Match a word boundary
0d017f4d 450 \B Match except at a word boundary
b85d18e9
IZ
451 \A Match only at beginning of string
452 \Z Match only at end of string, or before newline at the end
453 \z Match only at end of string
9da458fc
IZ
454 \G Match only at pos() (e.g. at the end-of-match position
455 of prior m//g)
a0d0e21e 456
14218588 457A word boundary (C<\b>) is a spot between two characters
19799a22
GS
458that has a C<\w> on one side of it and a C<\W> on the other side
459of it (in either order), counting the imaginary characters off the
460beginning and end of the string as matching a C<\W>. (Within
461character classes C<\b> represents backspace rather than a word
462boundary, just as it normally does in any double-quoted string.)
463The C<\A> and C<\Z> are just like "^" and "$", except that they
464won't match multiple times when the C</m> modifier is used, while
465"^" and "$" will match at every internal line boundary. To match
466the actual end of the string and not ignore an optional trailing
467newline, use C<\z>.
d74e8afc 468X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
469
470The C<\G> assertion can be used to chain global matches (using
471C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
472It is also useful when writing C<lex>-like scanners, when you have
473several patterns that you want to match against consequent substrings
474of your string, see the previous reference. The actual location
475where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d
YO
476an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
477matches is modified somewhat, in that contents to the left of C<\G> is
478not counted when determining the length of the match. Thus the following
479will not match forever:
d74e8afc 480X<\G>
c47ff5f1 481
58e23c8d
YO
482 $str = 'ABC';
483 pos($str) = 1;
484 while (/.\G/g) {
485 print $&;
486 }
487
488It will print 'A' and then terminate, as it considers the match to
489be zero-width, and thus will not match at the same position twice in a
490row.
491
492It is worth noting that C<\G> improperly used can result in an infinite
493loop. Take care when using patterns that include C<\G> in an alternation.
494
04838cea
RGS
495=head3 Capture buffers
496
0d017f4d
WL
497The bracketing construct C<( ... )> creates capture buffers. To refer
498to the current contents of a buffer later on, within the same pattern,
499use \1 for the first, \2 for the second, and so on.
500Outside the match use "$" instead of "\". (The
81714fb9 501\<digit> notation works in certain circumstances outside
14218588
GS
502the match. See the warning below about \1 vs $1 for details.)
503Referring back to another part of the match is called a
504I<backreference>.
d74e8afc
ITB
505X<regex, capture buffer> X<regexp, capture buffer>
506X<regular expression, capture buffer> X<backreference>
14218588
GS
507
508There is no limit to the number of captured substrings that you may
509use. However Perl also uses \10, \11, etc. as aliases for \010,
fb55449c
JH
510\011, etc. (Recall that 0 means octal, so \011 is the character at
511number 9 in your coded character set; which would be the 10th character,
81714fb9
YO
512a horizontal tab under ASCII.) Perl resolves this
513ambiguity by interpreting \10 as a backreference only if at least 10
514left parentheses have opened before it. Likewise \11 is a
515backreference only if at least 11 left parentheses have opened
516before it. And so on. \1 through \9 are always interpreted as
5624f11d 517backreferences.
40863337
ML
518If the bracketing group did not match, the associated backreference won't
519match either. (This can happen if the bracketing group is optional, or
520in a different branch of an alternation.)
521
1f1031fe 522X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
2bf803e2 523In order to provide a safer and easier way to construct patterns using
99d59c4d
RGS
524backreferences, Perl provides the C<\g{N}> notation (starting with perl
5255.10.0). The curly brackets are optional, however omitting them is less
526safe as the meaning of the pattern can be changed by text (such as digits)
527following it. When N is a positive integer the C<\g{N}> notation is
528exactly equivalent to using normal backreferences. When N is a negative
529integer then it is a relative backreference referring to the previous N'th
530capturing group. When the bracket form is used and N is not an integer, it
531is treated as a reference to a named buffer.
2bf803e2
YO
532
533Thus C<\g{-1}> refers to the last buffer, C<\g{-2}> refers to the
534buffer before that. For example:
5624f11d
YO
535
536 /
537 (Y) # buffer 1
538 ( # buffer 2
539 (X) # buffer 3
2bf803e2
YO
540 \g{-1} # backref to buffer 3
541 \g{-3} # backref to buffer 1
5624f11d
YO
542 )
543 /x
544
2bf803e2 545and would match the same as C</(Y) ( (X) \3 \1 )/x>.
14218588 546
99d59c4d 547Additionally, as of Perl 5.10.0 you may use named capture buffers and named
1f1031fe 548backreferences. The notation is C<< (?<name>...) >> to declare and C<< \k<name> >>
0d017f4d
WL
549to reference. You may also use apostrophes instead of angle brackets to delimit the
550name; and you may use the bracketed C<< \g{name} >> backreference syntax.
551It's possible to refer to a named capture buffer by absolute and relative number as well.
552Outside the pattern, a named capture buffer is available via the C<%+> hash.
553When different buffers within the same pattern have the same name, C<$+{name}>
554and C<< \k<name> >> refer to the leftmost defined group. (Thus it's possible
555to do things with named capture buffers that would otherwise require C<(??{})>
556code to accomplish.)
557X<named capture buffer> X<regular expression, named capture buffer>
64c5a566 558X<%+> X<$+{name}> X<< \k<name> >>
81714fb9 559
14218588 560Examples:
a0d0e21e
LW
561
562 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
563
81714fb9
YO
564 /(.)\1/ # find first doubled char
565 and print "'$1' is the first doubled character\n";
566
567 /(?<char>.)\k<char>/ # ... a different way
568 and print "'$+{char}' is the first doubled character\n";
569
0d017f4d 570 /(?'char'.)\1/ # ... mix and match
81714fb9 571 and print "'$1' is the first doubled character\n";
c47ff5f1 572
14218588 573 if (/Time: (..):(..):(..)/) { # parse out values
a0d0e21e
LW
574 $hours = $1;
575 $minutes = $2;
576 $seconds = $3;
577 }
c47ff5f1 578
14218588
GS
579Several special variables also refer back to portions of the previous
580match. C<$+> returns whatever the last bracket match matched.
581C<$&> returns the entire matched string. (At one point C<$0> did
582also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
583everything before the matched string. C<$'> returns everything
584after the matched string. And C<$^N> contains whatever was matched by
585the most-recently closed group (submatch). C<$^N> can be used in
586extended patterns (see below), for example to assign a submatch to a
81714fb9 587variable.
d74e8afc 588X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 589
665e98b9 590The numbered match variables ($1, $2, $3, etc.) and the related punctuation
77ea4f6d 591set (C<$+>, C<$&>, C<$`>, C<$'>, and C<$^N>) are all dynamically scoped
14218588
GS
592until the end of the enclosing block or until the next successful
593match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
594X<$+> X<$^N> X<$&> X<$`> X<$'>
595X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
596
14218588 597
0d017f4d 598B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 599which makes it easier to write code that tests for a series of more
665e98b9
JH
600specific cases and remembers the best match.
601
14218588
GS
602B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
603C<$'> anywhere in the program, it has to provide them for every
604pattern match. This may substantially slow your program. Perl
605uses the same mechanism to produce $1, $2, etc, so you also pay a
606price for each pattern that contains capturing parentheses. (To
607avoid this cost while retaining the grouping behaviour, use the
608extended regular expression C<(?: ... )> instead.) But if you never
609use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
610parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
611if you can, but if you can't (and some algorithms really appreciate
612them), once you've used them once, use them at will, because you've
613already paid the price. As of 5.005, C<$&> is not so costly as the
614other two.
d74e8afc 615X<$&> X<$`> X<$'>
68dc0745 616
99d59c4d 617As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
618C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
619and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 620successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
621The use of these variables incurs no global performance penalty, unlike
622their punctuation char equivalents, however at the trade-off that you
623have to tell perl when you want to use them.
87e95b7f 624X</p> X<p modifier>
cde0cee5 625
19799a22
GS
626Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
627C<\w>, C<\n>. Unlike some other regular expression languages, there
628are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 629that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
630interpreted as a literal character, not a metacharacter. This was
631once used in a common idiom to disable or quote the special meanings
632of regular expression metacharacters in a string that you want to
36bbe248 633use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
634
635 $pattern =~ s/(\W)/\\$1/g;
636
f1cbbd6e 637(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
638Today it is more common to use the quotemeta() function or the C<\Q>
639metaquoting escape sequence to disable all metacharacters' special
640meanings like this:
a0d0e21e
LW
641
642 /$unquoted\Q$quoted\E$unquoted/
643
9da458fc
IZ
644Beware that if you put literal backslashes (those not inside
645interpolated variables) between C<\Q> and C<\E>, double-quotish
646backslash interpolation may lead to confusing results. If you
647I<need> to use literal backslashes within C<\Q...\E>,
648consult L<perlop/"Gory details of parsing quoted constructs">.
649
19799a22
GS
650=head2 Extended Patterns
651
14218588
GS
652Perl also defines a consistent extension syntax for features not
653found in standard tools like B<awk> and B<lex>. The syntax is a
654pair of parentheses with a question mark as the first thing within
655the parentheses. The character after the question mark indicates
656the extension.
19799a22 657
14218588
GS
658The stability of these extensions varies widely. Some have been
659part of the core language for many years. Others are experimental
660and may change without warning or be completely removed. Check
661the documentation on an individual feature to verify its current
662status.
19799a22 663
14218588
GS
664A question mark was chosen for this and for the minimal-matching
665construct because 1) question marks are rare in older regular
666expressions, and 2) whenever you see one, you should stop and
667"question" exactly what is going on. That's psychology...
a0d0e21e
LW
668
669=over 10
670
cc6b7395 671=item C<(?#text)>
d74e8afc 672X<(?#)>
a0d0e21e 673
14218588 674A comment. The text is ignored. If the C</x> modifier enables
19799a22 675whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
676the comment as soon as it sees a C<)>, so there is no way to put a literal
677C<)> in the comment.
a0d0e21e 678
f7819f85 679=item C<(?pimsx-imsx)>
d74e8afc 680X<(?)>
19799a22 681
0b6d1084
JH
682One or more embedded pattern-match modifiers, to be turned on (or
683turned off, if preceded by C<->) for the remainder of the pattern or
684the remainder of the enclosing pattern group (if any). This is
685particularly useful for dynamic patterns, such as those read in from a
0d017f4d
WL
686configuration file, taken from an argument, or specified in a table
687somewhere. Consider the case where some patterns want to be case
688sensitive and some do not: The case insensitive ones merely need to
689include C<(?i)> at the front of the pattern. For example:
19799a22
GS
690
691 $pattern = "foobar";
5d458dd8 692 if ( /$pattern/i ) { }
19799a22
GS
693
694 # more flexible:
695
696 $pattern = "(?i)foobar";
5d458dd8 697 if ( /$pattern/ ) { }
19799a22 698
0b6d1084 699These modifiers are restored at the end of the enclosing group. For example,
19799a22
GS
700
701 ( (?i) blah ) \s+ \1
702
0d017f4d
WL
703will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
704repetition of the previous word, assuming the C</x> modifier, and no C</i>
705modifier outside this group.
19799a22 706
8eb5594e
DR
707These modifiers do not carry over into named subpatterns called in the
708enclosing group. In other words, a pattern such as C<((?i)(&NAME))> does not
709change the case-sensitivity of the "NAME" pattern.
710
5530442b 711Note that the C<p> modifier is special in that it can only be enabled,
cde0cee5 712not disabled, and that its presence anywhere in a pattern has a global
5530442b 713effect. Thus C<(?-p)> and C<(?-p:...)> are meaningless and will warn
cde0cee5
YO
714when executed under C<use warnings>.
715
5a964f20 716=item C<(?:pattern)>
d74e8afc 717X<(?:)>
a0d0e21e 718
ca9dfc88
IZ
719=item C<(?imsx-imsx:pattern)>
720
5a964f20
TC
721This is for clustering, not capturing; it groups subexpressions like
722"()", but doesn't make backreferences as "()" does. So
a0d0e21e 723
5a964f20 724 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
725
726is like
727
5a964f20 728 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 729
19799a22
GS
730but doesn't spit out extra fields. It's also cheaper not to capture
731characters if you don't need to.
a0d0e21e 732
19799a22 733Any letters between C<?> and C<:> act as flags modifiers as with
5d458dd8 734C<(?imsx-imsx)>. For example,
ca9dfc88
IZ
735
736 /(?s-i:more.*than).*million/i
737
14218588 738is equivalent to the more verbose
ca9dfc88
IZ
739
740 /(?:(?s-i)more.*than).*million/i
741
594d7033
YO
742=item C<(?|pattern)>
743X<(?|)> X<Branch reset>
744
745This is the "branch reset" pattern, which has the special property
746that the capture buffers are numbered from the same starting point
99d59c4d 747in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 748
693596a8
RGS
749Capture buffers are numbered from left to right, but inside this
750construct the numbering is restarted for each branch.
4deaaa80
PJ
751
752The numbering within each branch will be as normal, and any buffers
753following this construct will be numbered as though the construct
754contained only one branch, that being the one with the most capture
755buffers in it.
756
757This construct will be useful when you want to capture one of a
758number of alternative matches.
759
760Consider the following pattern. The numbers underneath show in
761which buffer the captured content will be stored.
594d7033
YO
762
763
764 # before ---------------branch-reset----------- after
765 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
766 # 1 2 2 3 2 3 4
767
ab106183
A
768Be careful when using the branch reset pattern in combination with
769named captures. Named captures are implemented as being aliases to
770numbered buffers holding the captures, and that interferes with the
771implementation of the branch reset pattern. If you are using named
772captures in a branch reset pattern, it's best to use the same names,
773in the same order, in each of the alternations:
774
775 /(?| (?<a> x ) (?<b> y )
776 | (?<a> z ) (?<b> w )) /x
777
778Not doing so may lead to surprises:
779
780 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
781 say $+ {a}; # Prints '12'
782 say $+ {b}; # *Also* prints '12'.
783
784The problem here is that both the buffer named C<< a >> and the buffer
785named C<< b >> are aliases for the buffer belonging to C<< $1 >>.
90a18110 786
ee9b8eae
YO
787=item Look-Around Assertions
788X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
789
790Look-around assertions are zero width patterns which match a specific
791pattern without including it in C<$&>. Positive assertions match when
792their subpattern matches, negative assertions match when their subpattern
793fails. Look-behind matches text up to the current match position,
794look-ahead matches text following the current match position.
795
796=over 4
797
5a964f20 798=item C<(?=pattern)>
d74e8afc 799X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 800
19799a22 801A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
802matches a word followed by a tab, without including the tab in C<$&>.
803
5a964f20 804=item C<(?!pattern)>
d74e8afc 805X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 806
19799a22 807A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 808matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
809however that look-ahead and look-behind are NOT the same thing. You cannot
810use this for look-behind.
7b8d334a 811
5a964f20 812If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
813will not do what you want. That's because the C<(?!foo)> is just saying that
814the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
815match. You would have to do something like C</(?!foo)...bar/> for that. We
816say "like" because there's the case of your "bar" not having three characters
817before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
818Sometimes it's still easier just to say:
a0d0e21e 819
a3cb178b 820 if (/bar/ && $` !~ /foo$/)
a0d0e21e 821
19799a22 822For look-behind see below.
c277df42 823
ee9b8eae
YO
824=item C<(?<=pattern)> C<\K>
825X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 826
c47ff5f1 827A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
828matches a word that follows a tab, without including the tab in C<$&>.
829Works only for fixed-width look-behind.
c277df42 830
ee9b8eae
YO
831There is a special form of this construct, called C<\K>, which causes the
832regex engine to "keep" everything it had matched prior to the C<\K> and
833not include it in C<$&>. This effectively provides variable length
834look-behind. The use of C<\K> inside of another look-around assertion
835is allowed, but the behaviour is currently not well defined.
836
c62285ac 837For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
838equivalent C<< (?<=...) >> construct, and it is especially useful in
839situations where you want to efficiently remove something following
840something else in a string. For instance
841
842 s/(foo)bar/$1/g;
843
844can be rewritten as the much more efficient
845
846 s/foo\Kbar//g;
847
5a964f20 848=item C<(?<!pattern)>
d74e8afc 849X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 850
19799a22
GS
851A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
852matches any occurrence of "foo" that does not follow "bar". Works
853only for fixed-width look-behind.
c277df42 854
ee9b8eae
YO
855=back
856
81714fb9
YO
857=item C<(?'NAME'pattern)>
858
859=item C<< (?<NAME>pattern) >>
860X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
861
862A named capture buffer. Identical in every respect to normal capturing
90a18110
RGS
863parentheses C<()> but for the additional fact that C<%+> or C<%-> may be
864used after a successful match to refer to a named buffer. See C<perlvar>
865for more details on the C<%+> and C<%-> hashes.
81714fb9
YO
866
867If multiple distinct capture buffers have the same name then the
868$+{NAME} will refer to the leftmost defined buffer in the match.
869
0d017f4d 870The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
871
872B<NOTE:> While the notation of this construct is the same as the similar
0d017f4d 873function in .NET regexes, the behavior is not. In Perl the buffers are
81714fb9
YO
874numbered sequentially regardless of being named or not. Thus in the
875pattern
876
877 /(x)(?<foo>y)(z)/
878
879$+{foo} will be the same as $2, and $3 will contain 'z' instead of
880the opposite which is what a .NET regex hacker might expect.
881
1f1031fe
YO
882Currently NAME is restricted to simple identifiers only.
883In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
884its Unicode extension (see L<utf8>),
885though it isn't extended by the locale (see L<perllocale>).
81714fb9 886
1f1031fe 887B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 888with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 889may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 890support the use of single quotes as a delimiter for the name.
81714fb9 891
1f1031fe
YO
892=item C<< \k<NAME> >>
893
894=item C<< \k'NAME' >>
81714fb9
YO
895
896Named backreference. Similar to numeric backreferences, except that
897the group is designated by name and not number. If multiple groups
898have the same name then it refers to the leftmost defined group in
899the current match.
900
0d017f4d 901It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
902earlier in the pattern.
903
904Both forms are equivalent.
905
1f1031fe 906B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 907with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 908may be used instead of C<< \k<NAME> >>.
1f1031fe 909
cc6b7395 910=item C<(?{ code })>
d74e8afc 911X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 912
19799a22 913B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
914experimental, and may be changed without notice. Code executed that
915has side effects may not perform identically from version to version
916due to the effect of future optimisations in the regex engine.
c277df42 917
cc46d5f2 918This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
919always succeeds, and its C<code> is not interpolated. Currently,
920the rules to determine where the C<code> ends are somewhat convoluted.
921
77ea4f6d
JV
922This feature can be used together with the special variable C<$^N> to
923capture the results of submatches in variables without having to keep
924track of the number of nested parentheses. For example:
925
926 $_ = "The brown fox jumps over the lazy dog";
927 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
928 print "color = $color, animal = $animal\n";
929
754091cb
RGS
930Inside the C<(?{...})> block, C<$_> refers to the string the regular
931expression is matching against. You can also use C<pos()> to know what is
fa11829f 932the current position of matching within this string.
754091cb 933
19799a22
GS
934The C<code> is properly scoped in the following sense: If the assertion
935is backtracked (compare L<"Backtracking">), all changes introduced after
936C<local>ization are undone, so that
b9ac3b5b
GS
937
938 $_ = 'a' x 8;
5d458dd8 939 m<
b9ac3b5b
GS
940 (?{ $cnt = 0 }) # Initialize $cnt.
941 (
5d458dd8 942 a
b9ac3b5b
GS
943 (?{
944 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
945 })
5d458dd8 946 )*
b9ac3b5b
GS
947 aaaa
948 (?{ $res = $cnt }) # On success copy to non-localized
949 # location.
950 >x;
951
0d017f4d 952will set C<$res = 4>. Note that after the match, C<$cnt> returns to the globally
14218588 953introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
954are unwound.
955
19799a22
GS
956This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
957switch. If I<not> used in this way, the result of evaluation of
958C<code> is put into the special variable C<$^R>. This happens
959immediately, so C<$^R> can be used from other C<(?{ code })> assertions
960inside the same regular expression.
b9ac3b5b 961
19799a22
GS
962The assignment to C<$^R> above is properly localized, so the old
963value of C<$^R> is restored if the assertion is backtracked; compare
964L<"Backtracking">.
b9ac3b5b 965
19799a22
GS
966For reasons of security, this construct is forbidden if the regular
967expression involves run-time interpolation of variables, unless the
968perilous C<use re 'eval'> pragma has been used (see L<re>), or the
969variables contain results of C<qr//> operator (see
5d458dd8 970L<perlop/"qr/STRING/imosx">).
871b0233 971
0d017f4d 972This restriction is due to the wide-spread and remarkably convenient
19799a22 973custom of using run-time determined strings as patterns. For example:
871b0233
IZ
974
975 $re = <>;
976 chomp $re;
977 $string =~ /$re/;
978
14218588
GS
979Before Perl knew how to execute interpolated code within a pattern,
980this operation was completely safe from a security point of view,
981although it could raise an exception from an illegal pattern. If
982you turn on the C<use re 'eval'>, though, it is no longer secure,
983so you should only do so if you are also using taint checking.
984Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 985compartment. See L<perlsec> for details about both these mechanisms.
871b0233 986
e95d7314
GG
987B<WARNING>: Use of lexical (C<my>) variables in these blocks is
988broken. The result is unpredictable and will make perl unstable. The
989workaround is to use global (C<our>) variables.
990
991B<WARNING>: Because Perl's regex engine is currently not re-entrant,
992interpolated code may not invoke the regex engine either directly with
993C<m//> or C<s///>), or indirectly with functions such as
994C<split>. Invoking the regex engine in these blocks will make perl
995unstable.
8988a1bb 996
14455d6c 997=item C<(??{ code })>
d74e8afc
ITB
998X<(??{})>
999X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1000
19799a22 1001B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1002experimental, and may be changed without notice. Code executed that
1003has side effects may not perform identically from version to version
1004due to the effect of future optimisations in the regex engine.
0f5d15d6 1005
19799a22
GS
1006This is a "postponed" regular subexpression. The C<code> is evaluated
1007at run time, at the moment this subexpression may match. The result
1008of evaluation is considered as a regular expression and matched as
61528107 1009if it were inserted instead of this construct. Note that this means
6bda09f9
YO
1010that the contents of capture buffers defined inside an eval'ed pattern
1011are not available outside of the pattern, and vice versa, there is no
1012way for the inner pattern to refer to a capture buffer defined outside.
1013Thus,
1014
1015 ('a' x 100)=~/(??{'(.)' x 100})/
1016
81714fb9 1017B<will> match, it will B<not> set $1.
0f5d15d6 1018
428594d9 1019The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
1020where the C<code> ends are currently somewhat convoluted.
1021
1022The following pattern matches a parenthesized group:
0f5d15d6
IZ
1023
1024 $re = qr{
1025 \(
1026 (?:
1027 (?> [^()]+ ) # Non-parens without backtracking
1028 |
14455d6c 1029 (??{ $re }) # Group with matching parens
0f5d15d6
IZ
1030 )*
1031 \)
1032 }x;
1033
6bda09f9
YO
1034See also C<(?PARNO)> for a different, more efficient way to accomplish
1035the same task.
1036
0b370c0a
A
1037For reasons of security, this construct is forbidden if the regular
1038expression involves run-time interpolation of variables, unless the
1039perilous C<use re 'eval'> pragma has been used (see L<re>), or the
1040variables contain results of C<qr//> operator (see
1041L<perlop/"qr/STRING/imosx">).
1042
5d458dd8 1043Because perl's regex engine is not currently re-entrant, delayed
8988a1bb
DD
1044code may not invoke the regex engine either directly with C<m//> or C<s///>),
1045or indirectly with functions such as C<split>.
1046
5d458dd8
YO
1047Recursing deeper than 50 times without consuming any input string will
1048result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
1049changing it requires a custom build.
1050
542fa716
YO
1051=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1052X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1053X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1054X<regex, relative recursion>
6bda09f9 1055
81714fb9
YO
1056Similar to C<(??{ code })> except it does not involve compiling any code,
1057instead it treats the contents of a capture buffer as an independent
61528107 1058pattern that must match at the current position. Capture buffers
81714fb9 1059contained by the pattern will have the value as determined by the
6bda09f9
YO
1060outermost recursion.
1061
894be9b7
YO
1062PARNO is a sequence of digits (not starting with 0) whose value reflects
1063the paren-number of the capture buffer to recurse to. C<(?R)> recurses to
1064the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716
YO
1065C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
1066to be relative, with negative numbers indicating preceding capture buffers
1067and positive ones following. Thus C<(?-1)> refers to the most recently
1068declared buffer, and C<(?+1)> indicates the next buffer to be declared.
c74340f9
YO
1069Note that the counting for relative recursion differs from that of
1070relative backreferences, in that with recursion unclosed buffers B<are>
1071included.
6bda09f9 1072
81714fb9 1073The following pattern matches a function foo() which may contain
f145b7e9 1074balanced parentheses as the argument.
6bda09f9
YO
1075
1076 $re = qr{ ( # paren group 1 (full function)
81714fb9 1077 foo
6bda09f9
YO
1078 ( # paren group 2 (parens)
1079 \(
1080 ( # paren group 3 (contents of parens)
1081 (?:
1082 (?> [^()]+ ) # Non-parens without backtracking
1083 |
1084 (?2) # Recurse to start of paren group 2
1085 )*
1086 )
1087 \)
1088 )
1089 )
1090 }x;
1091
1092If the pattern was used as follows
1093
1094 'foo(bar(baz)+baz(bop))'=~/$re/
1095 and print "\$1 = $1\n",
1096 "\$2 = $2\n",
1097 "\$3 = $3\n";
1098
1099the output produced should be the following:
1100
1101 $1 = foo(bar(baz)+baz(bop))
1102 $2 = (bar(baz)+baz(bop))
81714fb9 1103 $3 = bar(baz)+baz(bop)
6bda09f9 1104
81714fb9 1105If there is no corresponding capture buffer defined, then it is a
61528107 1106fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1107string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1108into perl, so changing it requires a custom build.
1109
542fa716
YO
1110The following shows how using negative indexing can make it
1111easier to embed recursive patterns inside of a C<qr//> construct
1112for later use:
1113
1114 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1115 if (/foo $parens \s+ + \s+ bar $parens/x) {
1116 # do something here...
1117 }
1118
81714fb9 1119B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1120PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1121a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1122as atomic. Also, modifiers are resolved at compile time, so constructs
1123like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1124be processed.
6bda09f9 1125
894be9b7
YO
1126=item C<(?&NAME)>
1127X<(?&NAME)>
1128
0d017f4d
WL
1129Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1130parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1131the same name, then it recurses to the leftmost.
1132
1133It is an error to refer to a name that is not declared somewhere in the
1134pattern.
1135
1f1031fe
YO
1136B<NOTE:> In order to make things easier for programmers with experience
1137with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1138may be used instead of C<< (?&NAME) >>.
1f1031fe 1139
e2e6a0f1
YO
1140=item C<(?(condition)yes-pattern|no-pattern)>
1141X<(?()>
286f584a 1142
e2e6a0f1 1143=item C<(?(condition)yes-pattern)>
286f584a 1144
e2e6a0f1
YO
1145Conditional expression. C<(condition)> should be either an integer in
1146parentheses (which is valid if the corresponding pair of parentheses
1147matched), a look-ahead/look-behind/evaluate zero-width assertion, a
1148name in angle brackets or single quotes (which is valid if a buffer
1149with the given name matched), or the special symbol (R) (true when
1150evaluated inside of recursion or eval). Additionally the R may be
1151followed by a number, (which will be true when evaluated when recursing
1152inside of the appropriate group), or by C<&NAME>, in which case it will
1153be true only when evaluated during recursion in the named group.
1154
1155Here's a summary of the possible predicates:
1156
1157=over 4
1158
1159=item (1) (2) ...
1160
1161Checks if the numbered capturing buffer has matched something.
1162
1163=item (<NAME>) ('NAME')
1164
1165Checks if a buffer with the given name has matched something.
1166
1167=item (?{ CODE })
1168
1169Treats the code block as the condition.
1170
1171=item (R)
1172
1173Checks if the expression has been evaluated inside of recursion.
1174
1175=item (R1) (R2) ...
1176
1177Checks if the expression has been evaluated while executing directly
1178inside of the n-th capture group. This check is the regex equivalent of
1179
1180 if ((caller(0))[3] eq 'subname') { ... }
1181
1182In other words, it does not check the full recursion stack.
1183
1184=item (R&NAME)
1185
1186Similar to C<(R1)>, this predicate checks to see if we're executing
1187directly inside of the leftmost group with a given name (this is the same
1188logic used by C<(?&NAME)> to disambiguate). It does not check the full
1189stack, but only the name of the innermost active recursion.
1190
1191=item (DEFINE)
1192
1193In this case, the yes-pattern is never directly executed, and no
1194no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1195See below for details.
1196
1197=back
1198
1199For example:
1200
1201 m{ ( \( )?
1202 [^()]+
1203 (?(1) \) )
1204 }x
1205
1206matches a chunk of non-parentheses, possibly included in parentheses
1207themselves.
1208
1209A special form is the C<(DEFINE)> predicate, which never executes directly
1210its yes-pattern, and does not allow a no-pattern. This allows to define
1211subpatterns which will be executed only by using the recursion mechanism.
1212This way, you can define a set of regular expression rules that can be
1213bundled into any pattern you choose.
1214
1215It is recommended that for this usage you put the DEFINE block at the
1216end of the pattern, and that you name any subpatterns defined within it.
1217
1218Also, it's worth noting that patterns defined this way probably will
1219not be as efficient, as the optimiser is not very clever about
1220handling them.
1221
1222An example of how this might be used is as follows:
1223
2bf803e2 1224 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1225 (?(DEFINE)
2bf803e2
YO
1226 (?<NAME_PAT>....)
1227 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1228 )/x
1229
1230Note that capture buffers matched inside of recursion are not accessible
0d017f4d 1231after the recursion returns, so the extra layer of capturing buffers is
e2e6a0f1
YO
1232necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1233C<$+{NAME}> would be.
286f584a 1234
c47ff5f1 1235=item C<< (?>pattern) >>
6bda09f9 1236X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1237
19799a22
GS
1238An "independent" subexpression, one which matches the substring
1239that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1240position, and it matches I<nothing other than this substring>. This
19799a22
GS
1241construct is useful for optimizations of what would otherwise be
1242"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1243It may also be useful in places where the "grab all you can, and do not
1244give anything back" semantic is desirable.
19799a22 1245
c47ff5f1 1246For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1247(anchored at the beginning of string, as above) will match I<all>
1248characters C<a> at the beginning of string, leaving no C<a> for
1249C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1250since the match of the subgroup C<a*> is influenced by the following
1251group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1252C<a*ab> will match fewer characters than a standalone C<a*>, since
1253this makes the tail match.
1254
c47ff5f1 1255An effect similar to C<< (?>pattern) >> may be achieved by writing
19799a22
GS
1256C<(?=(pattern))\1>. This matches the same substring as a standalone
1257C<a+>, and the following C<\1> eats the matched string; it therefore
c47ff5f1 1258makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1259(The difference between these two constructs is that the second one
1260uses a capturing group, thus shifting ordinals of backreferences
1261in the rest of a regular expression.)
1262
1263Consider this pattern:
c277df42 1264
871b0233 1265 m{ \(
e2e6a0f1
YO
1266 (
1267 [^()]+ # x+
1268 |
871b0233
IZ
1269 \( [^()]* \)
1270 )+
e2e6a0f1 1271 \)
871b0233 1272 }x
5a964f20 1273
19799a22
GS
1274That will efficiently match a nonempty group with matching parentheses
1275two levels deep or less. However, if there is no such group, it
1276will take virtually forever on a long string. That's because there
1277are so many different ways to split a long string into several
1278substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1279to a subpattern of the above pattern. Consider how the pattern
1280above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1281seconds, but that each extra letter doubles this time. This
1282exponential performance will make it appear that your program has
14218588 1283hung. However, a tiny change to this pattern
5a964f20 1284
e2e6a0f1
YO
1285 m{ \(
1286 (
1287 (?> [^()]+ ) # change x+ above to (?> x+ )
1288 |
871b0233
IZ
1289 \( [^()]* \)
1290 )+
e2e6a0f1 1291 \)
871b0233 1292 }x
c277df42 1293
c47ff5f1 1294which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1295this yourself would be a productive exercise), but finishes in a fourth
1296the time when used on a similar string with 1000000 C<a>s. Be aware,
1297however, that this pattern currently triggers a warning message under
9f1b1f2d 1298the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1299C<"matches null string many times in regex">.
c277df42 1300
c47ff5f1 1301On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1302effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1303This was only 4 times slower on a string with 1000000 C<a>s.
1304
9da458fc
IZ
1305The "grab all you can, and do not give anything back" semantic is desirable
1306in many situations where on the first sight a simple C<()*> looks like
1307the correct solution. Suppose we parse text with comments being delimited
1308by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1309its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1310the comment delimiter, because it may "give up" some whitespace if
1311the remainder of the pattern can be made to match that way. The correct
1312answer is either one of these:
1313
1314 (?>#[ \t]*)
1315 #[ \t]*(?![ \t])
1316
1317For example, to grab non-empty comments into $1, one should use either
1318one of these:
1319
1320 / (?> \# [ \t]* ) ( .+ ) /x;
1321 / \# [ \t]* ( [^ \t] .* ) /x;
1322
1323Which one you pick depends on which of these expressions better reflects
1324the above specification of comments.
1325
6bda09f9
YO
1326In some literature this construct is called "atomic matching" or
1327"possessive matching".
1328
b9b4dddf
YO
1329Possessive quantifiers are equivalent to putting the item they are applied
1330to inside of one of these constructs. The following equivalences apply:
1331
1332 Quantifier Form Bracketing Form
1333 --------------- ---------------
1334 PAT*+ (?>PAT*)
1335 PAT++ (?>PAT+)
1336 PAT?+ (?>PAT?)
1337 PAT{min,max}+ (?>PAT{min,max})
1338
e2e6a0f1
YO
1339=back
1340
1341=head2 Special Backtracking Control Verbs
1342
1343B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1344removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1345be noted to avoid problems during upgrades.
1346
1347These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1348otherwise stated the ARG argument is optional; in some cases, it is
1349forbidden.
1350
1351Any pattern containing a special backtracking verb that allows an argument
e1020413 1352has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1353C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1354rules apply:
e2e6a0f1 1355
5d458dd8
YO
1356On failure, the C<$REGERROR> variable will be set to the ARG value of the
1357verb pattern, if the verb was involved in the failure of the match. If the
1358ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1359name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1360none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1361
5d458dd8
YO
1362On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1363the C<$REGMARK> variable will be set to the name of the last
1364C<(*MARK:NAME)> pattern executed. See the explanation for the
1365C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1366
5d458dd8
YO
1367B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1368and most other regex related variables. They are not local to a scope, nor
1369readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1370Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1371
1372If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1373argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1
YO
1374
1375=over 4
1376
1377=item Verbs that take an argument
1378
1379=over 4
1380
5d458dd8 1381=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1382X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1383
5d458dd8
YO
1384This zero-width pattern prunes the backtracking tree at the current point
1385when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1386where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1387A may backtrack as necessary to match. Once it is reached, matching
1388continues in B, which may also backtrack as necessary; however, should B
1389not match, then no further backtracking will take place, and the pattern
1390will fail outright at the current starting position.
54612592
YO
1391
1392The following example counts all the possible matching strings in a
1393pattern (without actually matching any of them).
1394
e2e6a0f1 1395 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1396 print "Count=$count\n";
1397
1398which produces:
1399
1400 aaab
1401 aaa
1402 aa
1403 a
1404 aab
1405 aa
1406 a
1407 ab
1408 a
1409 Count=9
1410
5d458dd8 1411If we add a C<(*PRUNE)> before the count like the following
54612592 1412
5d458dd8 1413 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1414 print "Count=$count\n";
1415
1416we prevent backtracking and find the count of the longest matching
353c6505 1417at each matching starting point like so:
54612592
YO
1418
1419 aaab
1420 aab
1421 ab
1422 Count=3
1423
5d458dd8 1424Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1425
5d458dd8
YO
1426See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1427control backtracking. In some cases, the use of C<(*PRUNE)> can be
1428replaced with a C<< (?>pattern) >> with no functional difference; however,
1429C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1430C<< (?>pattern) >> alone.
54612592 1431
e2e6a0f1 1432
5d458dd8
YO
1433=item C<(*SKIP)> C<(*SKIP:NAME)>
1434X<(*SKIP)>
e2e6a0f1 1435
5d458dd8 1436This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1437failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1438to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1439of this pattern. This effectively means that the regex engine "skips" forward
1440to this position on failure and tries to match again, (assuming that
1441there is sufficient room to match).
1442
1443The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1444C<(*MARK:NAME)> was encountered while matching, then it is that position
1445which is used as the "skip point". If no C<(*MARK)> of that name was
1446encountered, then the C<(*SKIP)> operator has no effect. When used
1447without a name the "skip point" is where the match point was when
1448executing the (*SKIP) pattern.
1449
1450Compare the following to the examples in C<(*PRUNE)>, note the string
24b23f37
YO
1451is twice as long:
1452
5d458dd8 1453 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1454 print "Count=$count\n";
1455
1456outputs
1457
1458 aaab
1459 aaab
1460 Count=2
1461
5d458dd8 1462Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1463executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1464C<(*SKIP)> was executed.
1465
5d458dd8
YO
1466=item C<(*MARK:NAME)> C<(*:NAME)>
1467X<(*MARK)> C<(*MARK:NAME)> C<(*:NAME)>
1468
1469This zero-width pattern can be used to mark the point reached in a string
1470when a certain part of the pattern has been successfully matched. This
1471mark may be given a name. A later C<(*SKIP)> pattern will then skip
1472forward to that point if backtracked into on failure. Any number of
b4222fa9 1473C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1474
1475In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1476can be used to "label" a pattern branch, so that after matching, the
1477program can determine which branches of the pattern were involved in the
1478match.
1479
1480When a match is successful, the C<$REGMARK> variable will be set to the
1481name of the most recently executed C<(*MARK:NAME)> that was involved
1482in the match.
1483
1484This can be used to determine which branch of a pattern was matched
c62285ac 1485without using a separate capture buffer for each branch, which in turn
5d458dd8
YO
1486can result in a performance improvement, as perl cannot optimize
1487C</(?:(x)|(y)|(z))/> as efficiently as something like
1488C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1489
1490When a match has failed, and unless another verb has been involved in
1491failing the match and has provided its own name to use, the C<$REGERROR>
1492variable will be set to the name of the most recently executed
1493C<(*MARK:NAME)>.
1494
1495See C<(*SKIP)> for more details.
1496
b62d2d15
YO
1497As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1498
5d458dd8
YO
1499=item C<(*THEN)> C<(*THEN:NAME)>
1500
241e7389 1501This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1502C<(*PRUNE)>, this verb always matches, and when backtracked into on
1503failure, it causes the regex engine to try the next alternation in the
1504innermost enclosing group (capturing or otherwise).
1505
1506Its name comes from the observation that this operation combined with the
1507alternation operator (C<|>) can be used to create what is essentially a
1508pattern-based if/then/else block:
1509
1510 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1511
1512Note that if this operator is used and NOT inside of an alternation then
1513it acts exactly like the C<(*PRUNE)> operator.
1514
1515 / A (*PRUNE) B /
1516
1517is the same as
1518
1519 / A (*THEN) B /
1520
1521but
1522
1523 / ( A (*THEN) B | C (*THEN) D ) /
1524
1525is not the same as
1526
1527 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1528
1529as after matching the A but failing on the B the C<(*THEN)> verb will
1530backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1531
e2e6a0f1
YO
1532=item C<(*COMMIT)>
1533X<(*COMMIT)>
24b23f37 1534
241e7389 1535This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1536zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1537into on failure it causes the match to fail outright. No further attempts
1538to find a valid match by advancing the start pointer will occur again.
1539For example,
24b23f37 1540
e2e6a0f1 1541 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1542 print "Count=$count\n";
1543
1544outputs
1545
1546 aaab
1547 Count=1
1548
e2e6a0f1
YO
1549In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1550does not match, the regex engine will not try any further matching on the
1551rest of the string.
c277df42 1552
e2e6a0f1 1553=back
9af228c6 1554
e2e6a0f1 1555=item Verbs without an argument
9af228c6
YO
1556
1557=over 4
1558
e2e6a0f1
YO
1559=item C<(*FAIL)> C<(*F)>
1560X<(*FAIL)> X<(*F)>
9af228c6 1561
e2e6a0f1
YO
1562This pattern matches nothing and always fails. It can be used to force the
1563engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1564fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1565
e2e6a0f1 1566It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1567
e2e6a0f1
YO
1568=item C<(*ACCEPT)>
1569X<(*ACCEPT)>
9af228c6 1570
e2e6a0f1
YO
1571B<WARNING:> This feature is highly experimental. It is not recommended
1572for production code.
9af228c6 1573
e2e6a0f1
YO
1574This pattern matches nothing and causes the end of successful matching at
1575the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1576whether there is actually more to match in the string. When inside of a
0d017f4d 1577nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1578via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1579
e2e6a0f1
YO
1580If the C<(*ACCEPT)> is inside of capturing buffers then the buffers are
1581marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1582For instance:
9af228c6 1583
e2e6a0f1 1584 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1585
e2e6a0f1 1586will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0d017f4d 1587be set. If another branch in the inner parentheses were matched, such as in the
e2e6a0f1 1588string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1589
1590=back
c277df42 1591
a0d0e21e
LW
1592=back
1593
c07a80fd 1594=head2 Backtracking
d74e8afc 1595X<backtrack> X<backtracking>
c07a80fd 1596
35a734be
IZ
1597NOTE: This section presents an abstract approximation of regular
1598expression behavior. For a more rigorous (and complicated) view of
1599the rules involved in selecting a match among possible alternatives,
0d017f4d 1600see L<Combining RE Pieces>.
35a734be 1601
c277df42 1602A fundamental feature of regular expression matching involves the
5a964f20 1603notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1604by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1605C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1606internally, but the general principle outlined here is valid.
c07a80fd 1607
1608For a regular expression to match, the I<entire> regular expression must
1609match, not just part of it. So if the beginning of a pattern containing a
1610quantifier succeeds in a way that causes later parts in the pattern to
1611fail, the matching engine backs up and recalculates the beginning
1612part--that's why it's called backtracking.
1613
1614Here is an example of backtracking: Let's say you want to find the
1615word following "foo" in the string "Food is on the foo table.":
1616
1617 $_ = "Food is on the foo table.";
1618 if ( /\b(foo)\s+(\w+)/i ) {
1619 print "$2 follows $1.\n";
1620 }
1621
1622When the match runs, the first part of the regular expression (C<\b(foo)>)
1623finds a possible match right at the beginning of the string, and loads up
1624$1 with "Foo". However, as soon as the matching engine sees that there's
1625no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1626mistake and starts over again one character after where it had the
c07a80fd 1627tentative match. This time it goes all the way until the next occurrence
1628of "foo". The complete regular expression matches this time, and you get
1629the expected output of "table follows foo."
1630
1631Sometimes minimal matching can help a lot. Imagine you'd like to match
1632everything between "foo" and "bar". Initially, you write something
1633like this:
1634
1635 $_ = "The food is under the bar in the barn.";
1636 if ( /foo(.*)bar/ ) {
1637 print "got <$1>\n";
1638 }
1639
1640Which perhaps unexpectedly yields:
1641
1642 got <d is under the bar in the >
1643
1644That's because C<.*> was greedy, so you get everything between the
14218588 1645I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd 1646to use minimal matching to make sure you get the text between a "foo"
1647and the first "bar" thereafter.
1648
1649 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1650 got <d is under the >
1651
0d017f4d 1652Here's another example. Let's say you'd like to match a number at the end
b6e13d97 1653of a string, and you also want to keep the preceding part of the match.
c07a80fd 1654So you write this:
1655
1656 $_ = "I have 2 numbers: 53147";
1657 if ( /(.*)(\d*)/ ) { # Wrong!
1658 print "Beginning is <$1>, number is <$2>.\n";
1659 }
1660
1661That won't work at all, because C<.*> was greedy and gobbled up the
1662whole string. As C<\d*> can match on an empty string the complete
1663regular expression matched successfully.
1664
8e1088bc 1665 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd 1666
1667Here are some variants, most of which don't work:
1668
1669 $_ = "I have 2 numbers: 53147";
1670 @pats = qw{
1671 (.*)(\d*)
1672 (.*)(\d+)
1673 (.*?)(\d*)
1674 (.*?)(\d+)
1675 (.*)(\d+)$
1676 (.*?)(\d+)$
1677 (.*)\b(\d+)$
1678 (.*\D)(\d+)$
1679 };
1680
1681 for $pat (@pats) {
1682 printf "%-12s ", $pat;
1683 if ( /$pat/ ) {
1684 print "<$1> <$2>\n";
1685 } else {
1686 print "FAIL\n";
1687 }
1688 }
1689
1690That will print out:
1691
1692 (.*)(\d*) <I have 2 numbers: 53147> <>
1693 (.*)(\d+) <I have 2 numbers: 5314> <7>
1694 (.*?)(\d*) <> <>
1695 (.*?)(\d+) <I have > <2>
1696 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1697 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1698 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1699 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1700
1701As you see, this can be a bit tricky. It's important to realize that a
1702regular expression is merely a set of assertions that gives a definition
1703of success. There may be 0, 1, or several different ways that the
1704definition might succeed against a particular string. And if there are
5a964f20
TC
1705multiple ways it might succeed, you need to understand backtracking to
1706know which variety of success you will achieve.
c07a80fd 1707
19799a22 1708When using look-ahead assertions and negations, this can all get even
8b19b778 1709trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd 1710followed by "123". You might try to write that as
1711
871b0233
IZ
1712 $_ = "ABC123";
1713 if ( /^\D*(?!123)/ ) { # Wrong!
1714 print "Yup, no 123 in $_\n";
1715 }
c07a80fd 1716
1717But that isn't going to match; at least, not the way you're hoping. It
1718claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1719why that pattern matches, contrary to popular expectations:
c07a80fd 1720
4358a253
SS
1721 $x = 'ABC123';
1722 $y = 'ABC445';
c07a80fd 1723
4358a253
SS
1724 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1725 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1726
4358a253
SS
1727 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1728 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd 1729
1730This prints
1731
1732 2: got ABC
1733 3: got AB
1734 4: got ABC
1735
5f05dabc 1736You might have expected test 3 to fail because it seems to a more
c07a80fd 1737general purpose version of test 1. The important difference between
1738them is that test 3 contains a quantifier (C<\D*>) and so can use
1739backtracking, whereas test 1 will not. What's happening is
1740that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1741non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1742let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1743fail.
14218588 1744
c07a80fd 1745The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1746try to match C<(?!123> with "123", which fails. But because
c07a80fd 1747a quantifier (C<\D*>) has been used in the regular expression, the
1748search engine can backtrack and retry the match differently
54310121 1749in the hope of matching the complete regular expression.
c07a80fd 1750
5a964f20
TC
1751The pattern really, I<really> wants to succeed, so it uses the
1752standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1753time. Now there's indeed something following "AB" that is not
14218588 1754"123". It's "C123", which suffices.
c07a80fd 1755
14218588
GS
1756We can deal with this by using both an assertion and a negation.
1757We'll say that the first part in $1 must be followed both by a digit
1758and by something that's not "123". Remember that the look-aheads
1759are zero-width expressions--they only look, but don't consume any
1760of the string in their match. So rewriting this way produces what
c07a80fd 1761you'd expect; that is, case 5 will fail, but case 6 succeeds:
1762
4358a253
SS
1763 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1764 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd 1765
1766 6: got ABC
1767
5a964f20 1768In other words, the two zero-width assertions next to each other work as though
19799a22 1769they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd 1770matches only if you're at the beginning of the line AND the end of the
1771line simultaneously. The deeper underlying truth is that juxtaposition in
1772regular expressions always means AND, except when you write an explicit OR
1773using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1774although the attempted matches are made at different positions because "a"
1775is not a zero-width assertion, but a one-width assertion.
1776
0d017f4d 1777B<WARNING>: Particularly complicated regular expressions can take
14218588 1778exponential time to solve because of the immense number of possible
0d017f4d 1779ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
1780internal optimizations done by the regular expression engine, this will
1781take a painfully long time to run:
c07a80fd 1782
e1901655
IZ
1783 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1784
1785And if you used C<*>'s in the internal groups instead of limiting them
1786to 0 through 5 matches, then it would take forever--or until you ran
1787out of stack space. Moreover, these internal optimizations are not
1788always applicable. For example, if you put C<{0,5}> instead of C<*>
1789on the external group, no current optimization is applicable, and the
1790match takes a long time to finish.
c07a80fd 1791
9da458fc
IZ
1792A powerful tool for optimizing such beasts is what is known as an
1793"independent group",
c47ff5f1 1794which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1795zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 1796the tail match, since they are in "logical" context: only
14218588 1797whether they match is considered relevant. For an example
9da458fc 1798where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1799following match, see L<C<< (?>pattern) >>>.
c277df42 1800
a0d0e21e 1801=head2 Version 8 Regular Expressions
d74e8afc 1802X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1803
5a964f20 1804In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1805routines, here are the pattern-matching rules not described above.
1806
54310121 1807Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1808with a special meaning described here or above. You can cause
5a964f20 1809characters that normally function as metacharacters to be interpreted
5f05dabc 1810literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
1811character; "\\" matches a "\"). This escape mechanism is also required
1812for the character used as the pattern delimiter.
1813
1814A series of characters matches that series of characters in the target
1815string, so the pattern C<blurfl> would match "blurfl" in the target
1816string.
a0d0e21e
LW
1817
1818You can specify a character class, by enclosing a list of characters
5d458dd8 1819in C<[]>, which will match any character from the list. If the
a0d0e21e 1820first character after the "[" is "^", the class matches any character not
14218588 1821in the list. Within a list, the "-" character specifies a
5a964f20 1822range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1823inclusive. If you want either "-" or "]" itself to be a member of a
1824class, put it at the start of the list (possibly after a "^"), or
1825escape it with a backslash. "-" is also taken literally when it is
1826at the end of the list, just before the closing "]". (The
84850974
DD
1827following all specify the same class of three characters: C<[-az]>,
1828C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
1829specifies a class containing twenty-six characters, even on EBCDIC-based
1830character sets.) Also, if you try to use the character
1831classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1832a range, the "-" is understood literally.
a0d0e21e 1833
8ada0baa
JH
1834Note also that the whole range idea is rather unportable between
1835character sets--and even within character sets they may cause results
1836you probably didn't expect. A sound principle is to use only ranges
0d017f4d 1837that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
1838[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1839spell out the character sets in full.
1840
54310121 1841Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1842used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1843"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
5d458dd8
YO
1844of octal digits, matches the character whose coded character set value
1845is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
1846matches the character whose numeric value is I<nn>. The expression \cI<x>
1847matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 1848matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1849
1850You can specify a series of alternatives for a pattern using "|" to
1851separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1852or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1853first alternative includes everything from the last pattern delimiter
1854("(", "[", or the beginning of the pattern) up to the first "|", and
1855the last alternative contains everything from the last "|" to the next
14218588
GS
1856pattern delimiter. That's why it's common practice to include
1857alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1858start and end.
1859
5a964f20 1860Alternatives are tried from left to right, so the first
a3cb178b
GS
1861alternative found for which the entire expression matches, is the one that
1862is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1863example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1864part will match, as that is the first alternative tried, and it successfully
1865matches the target string. (This might not seem important, but it is
1866important when you are capturing matched text using parentheses.)
1867
5a964f20 1868Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1869so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1870
14218588
GS
1871Within a pattern, you may designate subpatterns for later reference
1872by enclosing them in parentheses, and you may refer back to the
1873I<n>th subpattern later in the pattern using the metacharacter
1874\I<n>. Subpatterns are numbered based on the left to right order
1875of their opening parenthesis. A backreference matches whatever
1876actually matched the subpattern in the string being examined, not
1877the rules for that subpattern. Therefore, C<(0|0x)\d*\s\1\d*> will
1878match "0x1234 0x4321", but not "0x1234 01234", because subpattern
18791 matched "0x", even though the rule C<0|0x> could potentially match
1880the leading 0 in the second number.
cb1a09d0 1881
0d017f4d 1882=head2 Warning on \1 Instead of $1
cb1a09d0 1883
5a964f20 1884Some people get too used to writing things like:
cb1a09d0
AD
1885
1886 $pattern =~ s/(\W)/\\\1/g;
1887
1888This is grandfathered for the RHS of a substitute to avoid shocking the
1889B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1890PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1891the usual double-quoted string means a control-A. The customary Unix
1892meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1893of doing that, you get yourself into trouble if you then add an C</e>
1894modifier.
1895
5a964f20 1896 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1897
1898Or if you try to do
1899
1900 s/(\d+)/\1000/;
1901
1902You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1903C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1904with the operation of matching a backreference. Certainly they mean two
1905different things on the I<left> side of the C<s///>.
9fa51da4 1906
0d017f4d 1907=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 1908
19799a22 1909B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1910
1911Regular expressions provide a terse and powerful programming language. As
1912with most other power tools, power comes together with the ability
1913to wreak havoc.
1914
1915A common abuse of this power stems from the ability to make infinite
628afcb5 1916loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1917
1918 'foo' =~ m{ ( o? )* }x;
1919
0d017f4d 1920The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 1921in the string is not moved by the match, C<o?> would match again and again
527e91da 1922because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
1923is with the looping modifier C<//g>:
1924
1925 @matches = ( 'foo' =~ m{ o? }xg );
1926
1927or
1928
1929 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1930
1931or the loop implied by split().
1932
1933However, long experience has shown that many programming tasks may
14218588
GS
1934be significantly simplified by using repeated subexpressions that
1935may match zero-length substrings. Here's a simple example being:
c84d73f1
IZ
1936
1937 @chars = split //, $string; # // is not magic in split
1938 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1939
9da458fc 1940Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 1941the infinite loop>. The rules for this are different for lower-level
527e91da 1942loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
1943ones like the C</g> modifier or split() operator.
1944
19799a22
GS
1945The lower-level loops are I<interrupted> (that is, the loop is
1946broken) when Perl detects that a repeated expression matched a
1947zero-length substring. Thus
c84d73f1
IZ
1948
1949 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1950
5d458dd8 1951is made equivalent to
c84d73f1 1952
5d458dd8
YO
1953 m{ (?: NON_ZERO_LENGTH )*
1954 |
1955 (?: ZERO_LENGTH )?
c84d73f1
IZ
1956 }x;
1957
1958The higher level-loops preserve an additional state between iterations:
5d458dd8 1959whether the last match was zero-length. To break the loop, the following
c84d73f1 1960match after a zero-length match is prohibited to have a length of zero.
5d458dd8 1961This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
1962and so the I<second best> match is chosen if the I<best> match is of
1963zero length.
1964
19799a22 1965For example:
c84d73f1
IZ
1966
1967 $_ = 'bar';
1968 s/\w??/<$&>/g;
1969
20fb949f 1970results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 1971match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
1972best> match is what is matched by C<\w>. Thus zero-length matches
1973alternate with one-character-long matches.
1974
5d458dd8 1975Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
1976position one notch further in the string.
1977
19799a22 1978The additional state of being I<matched with zero-length> is associated with
c84d73f1 1979the matched string, and is reset by each assignment to pos().
9da458fc
IZ
1980Zero-length matches at the end of the previous match are ignored
1981during C<split>.
c84d73f1 1982
0d017f4d 1983=head2 Combining RE Pieces
35a734be
IZ
1984
1985Each of the elementary pieces of regular expressions which were described
1986before (such as C<ab> or C<\Z>) could match at most one substring
1987at the given position of the input string. However, in a typical regular
1988expression these elementary pieces are combined into more complicated
1989patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1990(in these examples C<S> and C<T> are regular subexpressions).
1991
1992Such combinations can include alternatives, leading to a problem of choice:
1993if we match a regular expression C<a|ab> against C<"abc">, will it match
1994substring C<"a"> or C<"ab">? One way to describe which substring is
1995actually matched is the concept of backtracking (see L<"Backtracking">).
1996However, this description is too low-level and makes you think
1997in terms of a particular implementation.
1998
1999Another description starts with notions of "better"/"worse". All the
2000substrings which may be matched by the given regular expression can be
2001sorted from the "best" match to the "worst" match, and it is the "best"
2002match which is chosen. This substitutes the question of "what is chosen?"
2003by the question of "which matches are better, and which are worse?".
2004
2005Again, for elementary pieces there is no such question, since at most
2006one match at a given position is possible. This section describes the
2007notion of better/worse for combining operators. In the description
2008below C<S> and C<T> are regular subexpressions.
2009
13a2d996 2010=over 4
35a734be
IZ
2011
2012=item C<ST>
2013
2014Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2015substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2016which can be matched by C<T>.
35a734be
IZ
2017
2018If C<A> is better match for C<S> than C<A'>, C<AB> is a better
2019match than C<A'B'>.
2020
2021If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
2022C<B> is better match for C<T> than C<B'>.
2023
2024=item C<S|T>
2025
2026When C<S> can match, it is a better match than when only C<T> can match.
2027
2028Ordering of two matches for C<S> is the same as for C<S>. Similar for
2029two matches for C<T>.
2030
2031=item C<S{REPEAT_COUNT}>
2032
2033Matches as C<SSS...S> (repeated as many times as necessary).
2034
2035=item C<S{min,max}>
2036
2037Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2038
2039=item C<S{min,max}?>
2040
2041Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2042
2043=item C<S?>, C<S*>, C<S+>
2044
2045Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2046
2047=item C<S??>, C<S*?>, C<S+?>
2048
2049Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2050
c47ff5f1 2051=item C<< (?>S) >>
35a734be
IZ
2052
2053Matches the best match for C<S> and only that.
2054
2055=item C<(?=S)>, C<(?<=S)>
2056
2057Only the best match for C<S> is considered. (This is important only if
2058C<S> has capturing parentheses, and backreferences are used somewhere
2059else in the whole regular expression.)
2060
2061=item C<(?!S)>, C<(?<!S)>
2062
2063For this grouping operator there is no need to describe the ordering, since
2064only whether or not C<S> can match is important.
2065
6bda09f9 2066=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2067
2068The ordering is the same as for the regular expression which is
6bda09f9 2069the result of EXPR, or the pattern contained by capture buffer PARNO.
35a734be
IZ
2070
2071=item C<(?(condition)yes-pattern|no-pattern)>
2072
2073Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2074already determined. The ordering of the matches is the same as for the
2075chosen subexpression.
2076
2077=back
2078
2079The above recipes describe the ordering of matches I<at a given position>.
2080One more rule is needed to understand how a match is determined for the
2081whole regular expression: a match at an earlier position is always better
2082than a match at a later position.
2083
0d017f4d 2084=head2 Creating Custom RE Engines
c84d73f1
IZ
2085
2086Overloaded constants (see L<overload>) provide a simple way to extend
2087the functionality of the RE engine.
2088
2089Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2090matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2091characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2092at these positions, so we want to have each C<\Y|> in the place of the
2093more complicated version. We can create a module C<customre> to do
2094this:
2095
2096 package customre;
2097 use overload;
2098
2099 sub import {
2100 shift;
2101 die "No argument to customre::import allowed" if @_;
2102 overload::constant 'qr' => \&convert;
2103 }
2104
2105 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2106
580a9fe1
RGS
2107 # We must also take care of not escaping the legitimate \\Y|
2108 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2109 my %rules = ( '\\' => '\\\\',
c84d73f1
IZ
2110 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
2111 sub convert {
2112 my $re = shift;
5d458dd8 2113 $re =~ s{
c84d73f1
IZ
2114 \\ ( \\ | Y . )
2115 }
5d458dd8 2116 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2117 return $re;
2118 }
2119
2120Now C<use customre> enables the new escape in constant regular
2121expressions, i.e., those without any runtime variable interpolations.
2122As documented in L<overload>, this conversion will work only over
2123literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2124part of this regular expression needs to be converted explicitly
2125(but only if the special meaning of C<\Y|> should be enabled inside $re):
2126
2127 use customre;
2128 $re = <>;
2129 chomp $re;
2130 $re = customre::convert $re;
2131 /\Y|$re\Y|/;
2132
1f1031fe
YO
2133=head1 PCRE/Python Support
2134
99d59c4d 2135As of Perl 5.10.0, Perl supports several Python/PCRE specific extensions
1f1031fe 2136to the regex syntax. While Perl programmers are encouraged to use the
99d59c4d 2137Perl specific syntax, the following are also accepted:
1f1031fe
YO
2138
2139=over 4
2140
ae5648b3 2141=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe
YO
2142
2143Define a named capture buffer. Equivalent to C<< (?<NAME>pattern) >>.
2144
2145=item C<< (?P=NAME) >>
2146
2147Backreference to a named capture buffer. Equivalent to C<< \g{NAME} >>.
2148
2149=item C<< (?P>NAME) >>
2150
2151Subroutine call to a named capture buffer. Equivalent to C<< (?&NAME) >>.
2152
ee9b8eae 2153=back
1f1031fe 2154
19799a22
GS
2155=head1 BUGS
2156
78288b8e
KW
2157There are numerous problems with case insensitive matching of characters
2158outside the ASCII range, especially with those whose folds are multiple
2159characters, such as ligatures like C<LATIN SMALL LIGATURE FF>.
2160
f253210b
KW
2161In a bracketed character class with case insensitive matching, ranges only work
2162for ASCII characters. For example,
2163C<m/[\N{CYRILLIC CAPITAL LETTER A}-\N{CYRILLIC CAPITAL LETTER YA}]/i>
2164doesn't match all the Russian upper and lower case letters.
2165
88c9975e
KW
2166Many regular expression constructs don't work on EBCDIC platforms.
2167
9da458fc
IZ
2168This document varies from difficult to understand to completely
2169and utterly opaque. The wandering prose riddled with jargon is
2170hard to fathom in several places.
2171
2172This document needs a rewrite that separates the tutorial content
2173from the reference content.
19799a22
GS
2174
2175=head1 SEE ALSO
9fa51da4 2176
91e0c79e
MJD
2177L<perlrequick>.
2178
2179L<perlretut>.
2180
9b599b2a
GS
2181L<perlop/"Regexp Quote-Like Operators">.
2182
1e66bd83
PP
2183L<perlop/"Gory details of parsing quoted constructs">.
2184
14218588
GS
2185L<perlfaq6>.
2186
9b599b2a
GS
2187L<perlfunc/pos>.
2188
2189L<perllocale>.
2190
fb55449c
JH
2191L<perlebcdic>.
2192
14218588
GS
2193I<Mastering Regular Expressions> by Jeffrey Friedl, published
2194by O'Reilly and Associates.