This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add examples to perlre on perils of not using \g{}
[perl5.git] / pod / perlre.pod
CommitLineData
a0d0e21e 1=head1 NAME
d74e8afc 2X<regular expression> X<regex> X<regexp>
a0d0e21e
LW
3
4perlre - Perl regular expressions
5
6=head1 DESCRIPTION
7
5d458dd8 8This page describes the syntax of regular expressions in Perl.
91e0c79e 9
cc46d5f2 10If you haven't used regular expressions before, a quick-start
91e0c79e
MJD
11introduction is available in L<perlrequick>, and a longer tutorial
12introduction is available in L<perlretut>.
13
14For reference on how regular expressions are used in matching
15operations, plus various examples of the same, see discussions of
16C<m//>, C<s///>, C<qr//> and C<??> in L<perlop/"Regexp Quote-Like
17Operators">.
cb1a09d0 18
0d017f4d
WL
19
20=head2 Modifiers
21
19799a22 22Matching operations can have various modifiers. Modifiers
5a964f20 23that relate to the interpretation of the regular expression inside
19799a22 24are listed below. Modifiers that alter the way a regular expression
5d458dd8 25is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
1e66bd83 26L<perlop/"Gory details of parsing quoted constructs">.
a0d0e21e 27
55497cff
PP
28=over 4
29
54310121 30=item m
d74e8afc 31X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff
PP
32
33Treat string as multiple lines. That is, change "^" and "$" from matching
14218588 34the start or end of the string to matching the start or end of any
7f761169 35line anywhere within the string.
55497cff 36
54310121 37=item s
d74e8afc
ITB
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
55497cff
PP
40
41Treat string as single line. That is, change "." to match any character
19799a22 42whatsoever, even a newline, which normally it would not match.
55497cff 43
34d67d80 44Used together, as C</ms>, they let the "." match any character whatsoever,
fb55449c 45while still allowing "^" and "$" to match, respectively, just after
19799a22 46and just before newlines within the string.
7b8d334a 47
87e95b7f
YO
48=item i
49X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
50X<regular expression, case-insensitive>
51
52Do case-insensitive pattern matching.
53
54If C<use locale> is in effect, the case map is taken from the current
55locale. See L<perllocale>.
56
54310121 57=item x
d74e8afc 58X</x>
55497cff
PP
59
60Extend your pattern's legibility by permitting whitespace and comments.
61
87e95b7f
YO
62=item p
63X</p> X<regex, preserve> X<regexp, preserve>
64
632a1772 65Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
87e95b7f
YO
66${^POSTMATCH} are available for use after matching.
67
e2e6bec7
DN
68=item g and c
69X</g> X</c>
70
71Global matching, and keep the Current position after failed matching.
72Unlike i, m, s and x, these two flags affect the way the regex is used
73rather than the regex itself. See
74L<perlretut/"Using regular expressions in Perl"> for further explanation
75of the g and c modifiers.
76
55497cff 77=back
a0d0e21e
LW
78
79These are usually written as "the C</x> modifier", even though the delimiter
14218588 80in question might not really be a slash. Any of these
a0d0e21e 81modifiers may also be embedded within the regular expression itself using
14218588 82the C<(?...)> construct. See below.
a0d0e21e 83
4633a7c4 84The C</x> modifier itself needs a little more explanation. It tells
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
9bb1f947 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
9bb1f947 108L<perluniprops/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
f793d64a
KW
127 \ Quote the next metacharacter
128 ^ Match the beginning of the line
129 . Match any character (except newline)
130 $ Match the end of the line (or before newline at the end)
131 | Alternation
132 () Grouping
133 [] Bracketed Character class
a0d0e21e 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 157
f793d64a
KW
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
a0d0e21e
LW
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
d0b16107 169to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
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
PP
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
f793d64a
KW
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
f793d64a
KW
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
198 {n}+ Match exactly n times and give nothing back (redundant)
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
LW
224also work:
225
f793d64a
KW
226 \t tab (HT, TAB)
227 \n newline (LF, NL)
228 \r return (CR)
229 \f form feed (FF)
230 \a alarm (bell) (BEL)
231 \e escape (think troff) (ESC)
232 \033 octal char (example: ESC)
233 \x1B hex char (example: ESC)
234 \x{263a} long hex char (example: Unicode SMILEY)
235 \cK control char (example: VT)
236 \N{name} named Unicode character
237 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
238 \l lowercase next char (think vi)
239 \u uppercase next char (think vi)
240 \L lowercase till \E (think vi)
241 \U uppercase till \E (think vi)
242 \Q quote (disable) pattern metacharacters till \E
243 \E end either case modification or quoted section, think vi
a0d0e21e 244
9bb1f947 245Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 246
e1d1eefb 247=head3 Character Classes and other Special Escapes
04838cea 248
a0d0e21e 249In addition, Perl defines the following:
d0b16107 250X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 251
f793d64a
KW
252 Sequence Note Description
253 [...] [1] Match a character according to the rules of the
254 bracketed character class defined by the "...".
255 Example: [a-z] matches "a" or "b" or "c" ... or "z"
256 [[:...:]] [2] Match a character according to the rules of the POSIX
257 character class "..." within the outer bracketed
258 character class. Example: [[:upper:]] matches any
259 uppercase character.
260 \w [3] Match a "word" character (alphanumeric plus "_")
261 \W [3] Match a non-"word" character
262 \s [3] Match a whitespace character
263 \S [3] Match a non-whitespace character
264 \d [3] Match a decimal digit character
265 \D [3] Match a non-digit character
266 \pP [3] Match P, named property. Use \p{Prop} for longer names
267 \PP [3] Match non-P
268 \X [4] Match Unicode "eXtended grapheme cluster"
269 \C Match a single C-language char (octet) even if that is
270 part of a larger UTF-8 character. Thus it breaks up
271 characters into their UTF-8 bytes, so you may end up
272 with malformed pieces of UTF-8. Unsupported in
273 lookbehind.
c27a5cfe 274 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
275 '1' may actually be any positive integer.
276 \g1 [5] Backreference to a specific or previous group,
277 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 278 previous group and may optionally be wrapped in
f793d64a
KW
279 curly brackets for safer parsing.
280 \g{name} [5] Named backreference
281 \k<name> [5] Named backreference
282 \K [6] Keep the stuff left of the \K, don't include it in $&
283 \N [7] Any character but \n (experimental). Not affected by
284 /s modifier
285 \v [3] Vertical whitespace
286 \V [3] Not vertical whitespace
287 \h [3] Horizontal whitespace
288 \H [3] Not horizontal whitespace
289 \R [4] Linebreak
e1d1eefb 290
9bb1f947
KW
291=over 4
292
293=item [1]
294
295See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 296
9bb1f947 297=item [2]
b8c5462f 298
9bb1f947 299See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 300
9bb1f947 301=item [3]
5496314a 302
9bb1f947 303See L<perlrecharclass/Backslash sequences> for details.
5496314a 304
9bb1f947 305=item [4]
5496314a 306
9bb1f947 307See L<perlrebackslash/Misc> for details.
d0b16107 308
9bb1f947 309=item [5]
b8c5462f 310
c27a5cfe 311See L</Capture groups> below for details.
93733859 312
9bb1f947 313=item [6]
b8c5462f 314
9bb1f947
KW
315See L</Extended Patterns> below for details.
316
317=item [7]
318
319Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
320character whose name is C<NAME>; and similarly when of the form
321C<\N{U+I<wide hex char>}>, it matches the character whose Unicode ordinal is
322I<wide hex char>. Otherwise it matches any character but C<\n>.
323
324=back
d0b16107 325
04838cea
RGS
326=head3 Assertions
327
a0d0e21e 328Perl defines the following zero-width assertions:
d74e8afc
ITB
329X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
330X<regexp, zero-width assertion>
331X<regular expression, zero-width assertion>
332X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 333
9bb1f947
KW
334 \b Match a word boundary
335 \B Match except at a word boundary
336 \A Match only at beginning of string
337 \Z Match only at end of string, or before newline at the end
338 \z Match only at end of string
339 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 340 of prior m//g)
a0d0e21e 341
14218588 342A word boundary (C<\b>) is a spot between two characters
19799a22
GS
343that has a C<\w> on one side of it and a C<\W> on the other side
344of it (in either order), counting the imaginary characters off the
345beginning and end of the string as matching a C<\W>. (Within
346character classes C<\b> represents backspace rather than a word
347boundary, just as it normally does in any double-quoted string.)
348The C<\A> and C<\Z> are just like "^" and "$", except that they
349won't match multiple times when the C</m> modifier is used, while
350"^" and "$" will match at every internal line boundary. To match
351the actual end of the string and not ignore an optional trailing
352newline, use C<\z>.
d74e8afc 353X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
354
355The C<\G> assertion can be used to chain global matches (using
356C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
357It is also useful when writing C<lex>-like scanners, when you have
358several patterns that you want to match against consequent substrings
359of your string, see the previous reference. The actual location
360where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d
YO
361an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
362matches is modified somewhat, in that contents to the left of C<\G> is
363not counted when determining the length of the match. Thus the following
364will not match forever:
d74e8afc 365X<\G>
c47ff5f1 366
58e23c8d
YO
367 $str = 'ABC';
368 pos($str) = 1;
369 while (/.\G/g) {
370 print $&;
371 }
372
373It will print 'A' and then terminate, as it considers the match to
374be zero-width, and thus will not match at the same position twice in a
375row.
376
377It is worth noting that C<\G> improperly used can result in an infinite
378loop. Take care when using patterns that include C<\G> in an alternation.
379
c27a5cfe 380=head3 Capture groups
04838cea 381
c27a5cfe
KW
382The bracketing construct C<( ... )> creates capture groups (also referred to as
383capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
384the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
385for the second, and so on.
386This is called a I<backreference>.
d74e8afc 387X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 388X<regex, capture group> X<regexp, capture group>
d74e8afc 389X<regular expression, capture buffer> X<backreference>
c27a5cfe 390X<regular expression, capture group> X<backreference>
1f1031fe 391X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
392X<named capture buffer> X<regular expression, named capture buffer>
393X<named capture group> X<regular expression, named capture group>
394X<%+> X<$+{name}> X<< \k<name> >>
395There is no limit to the number of captured substrings that you may use.
396Groups are numbered with the leftmost open parenthesis being number 1, etc. If
397a group did not match, the associated backreference won't match either. (This
398can happen if the group is optional, or in a different branch of an
399alternation.)
400You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
401this form, described below.
402
403You can also refer to capture groups relatively, by using a negative number, so
404that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
405group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
406example:
5624f11d
YO
407
408 /
c27a5cfe
KW
409 (Y) # group 1
410 ( # group 2
411 (X) # group 3
412 \g{-1} # backref to group 3
413 \g{-3} # backref to group 1
5624f11d
YO
414 )
415 /x
416
d8b950dc
KW
417would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
418interpolate regexes into larger regexes and not have to worry about the
419capture groups being renumbered.
420
421You can dispense with numbers altogether and create named capture groups.
422The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
423reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
424also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
425I<name> must not begin with a number, nor contain hyphens.
426When different groups within the same pattern have the same name, any reference
427to that name assumes the leftmost defined group. Named groups count in
428absolute and relative numbering, and so can also be referred to by those
429numbers.
430(It's possible to do things with named capture groups that would otherwise
431require C<(??{})>.)
432
433Capture group contents are dynamically scoped and available to you outside the
434pattern until the end of the enclosing block or until the next successful
435match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
436You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
437etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
438
439Braces are required in referring to named capture groups, but are optional for
440absolute or relative numbered ones. Braces are safer when creating a regex by
441concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
442contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
443is probably not what you intended.
444
445The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
446there were no named nor relative numbered capture groups. Absolute numbered
447groups were referred to using C<\1>, C<\2>, etc, and this notation is still
448accepted (and likely always will be). But it leads to some ambiguities if
449there are more than 9 capture groups, as C<\10> could mean either the tenth
450capture group, or the character whose ordinal in octal is 010 (a backspace in
451ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
452only if at least 10 left parentheses have opened before it. Likewise C<\11> is
453a backreference only if at least 11 left parentheses have opened before it.
454And so on. C<\1> through C<\9> are always interpreted as backreferences. You
455can minimize the ambiguity by always using C<\g> if you mean capturing groups;
456and always using 3 digits for octal constants, with the first always "0" (which
9d860678
KW
457works if there are 63 (= \077) or fewer capture groups). There are several
458examples below that illustrate these perils.
d8b950dc
KW
459
460The C<\I<digit>> notation also works in certain circumstances outside
461the pattern. See L</Warning on \1 Instead of $1> below for details.)
81714fb9 462
14218588 463Examples:
a0d0e21e
LW
464
465 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
466
d8b950dc 467 /(.)\g1/ # find first doubled char
81714fb9
YO
468 and print "'$1' is the first doubled character\n";
469
470 /(?<char>.)\k<char>/ # ... a different way
471 and print "'$+{char}' is the first doubled character\n";
472
d8b950dc 473 /(?'char'.)\g1/ # ... mix and match
81714fb9 474 and print "'$1' is the first doubled character\n";
c47ff5f1 475
14218588 476 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
477 $hours = $1;
478 $minutes = $2;
479 $seconds = $3;
a0d0e21e 480 }
c47ff5f1 481
9d860678
KW
482 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
483 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
484 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
485 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
486
487 $a = '(.)\1'; # Creates problems when concatenated.
488 $b = '(.)\g{1}'; # Avoids the problems.
489 "aa" =~ /${a}/; # True
490 "aa" =~ /${b}/; # True
491 "aa0" =~ /${a}0/; # False!
492 "aa0" =~ /${b}0/; # True
493 "aa\x8" =~ /${a}0/; # True!
494 "aa\x8" =~ /${b}0/; # False
495
14218588
GS
496Several special variables also refer back to portions of the previous
497match. C<$+> returns whatever the last bracket match matched.
498C<$&> returns the entire matched string. (At one point C<$0> did
499also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
500everything before the matched string. C<$'> returns everything
501after the matched string. And C<$^N> contains whatever was matched by
502the most-recently closed group (submatch). C<$^N> can be used in
503extended patterns (see below), for example to assign a submatch to a
81714fb9 504variable.
d74e8afc 505X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 506
d8b950dc
KW
507These special variables, like the C<%+> hash and the numbered match variables
508(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
509until the end of the enclosing block or until the next successful
510match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
511X<$+> X<$^N> X<$&> X<$`> X<$'>
512X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
513
0d017f4d 514B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 515which makes it easier to write code that tests for a series of more
665e98b9
JH
516specific cases and remembers the best match.
517
14218588
GS
518B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
519C<$'> anywhere in the program, it has to provide them for every
520pattern match. This may substantially slow your program. Perl
d8b950dc 521uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
522price for each pattern that contains capturing parentheses. (To
523avoid this cost while retaining the grouping behaviour, use the
524extended regular expression C<(?: ... )> instead.) But if you never
525use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
526parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
527if you can, but if you can't (and some algorithms really appreciate
528them), once you've used them once, use them at will, because you've
529already paid the price. As of 5.005, C<$&> is not so costly as the
530other two.
d74e8afc 531X<$&> X<$`> X<$'>
68dc0745 532
99d59c4d 533As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
534C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
535and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 536successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
537The use of these variables incurs no global performance penalty, unlike
538their punctuation char equivalents, however at the trade-off that you
539have to tell perl when you want to use them.
87e95b7f 540X</p> X<p modifier>
cde0cee5 541
9d727203
KW
542=head2 Quoting metacharacters
543
19799a22
GS
544Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
545C<\w>, C<\n>. Unlike some other regular expression languages, there
546are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 547that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
548interpreted as a literal character, not a metacharacter. This was
549once used in a common idiom to disable or quote the special meanings
550of regular expression metacharacters in a string that you want to
36bbe248 551use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
552
553 $pattern =~ s/(\W)/\\$1/g;
554
f1cbbd6e 555(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
556Today it is more common to use the quotemeta() function or the C<\Q>
557metaquoting escape sequence to disable all metacharacters' special
558meanings like this:
a0d0e21e
LW
559
560 /$unquoted\Q$quoted\E$unquoted/
561
9da458fc
IZ
562Beware that if you put literal backslashes (those not inside
563interpolated variables) between C<\Q> and C<\E>, double-quotish
564backslash interpolation may lead to confusing results. If you
565I<need> to use literal backslashes within C<\Q...\E>,
566consult L<perlop/"Gory details of parsing quoted constructs">.
567
19799a22
GS
568=head2 Extended Patterns
569
14218588
GS
570Perl also defines a consistent extension syntax for features not
571found in standard tools like B<awk> and B<lex>. The syntax is a
572pair of parentheses with a question mark as the first thing within
573the parentheses. The character after the question mark indicates
574the extension.
19799a22 575
14218588
GS
576The stability of these extensions varies widely. Some have been
577part of the core language for many years. Others are experimental
578and may change without warning or be completely removed. Check
579the documentation on an individual feature to verify its current
580status.
19799a22 581
14218588
GS
582A question mark was chosen for this and for the minimal-matching
583construct because 1) question marks are rare in older regular
584expressions, and 2) whenever you see one, you should stop and
585"question" exactly what is going on. That's psychology...
a0d0e21e
LW
586
587=over 10
588
cc6b7395 589=item C<(?#text)>
d74e8afc 590X<(?#)>
a0d0e21e 591
14218588 592A comment. The text is ignored. If the C</x> modifier enables
19799a22 593whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
594the comment as soon as it sees a C<)>, so there is no way to put a literal
595C<)> in the comment.
a0d0e21e 596
f7819f85 597=item C<(?pimsx-imsx)>
d74e8afc 598X<(?)>
19799a22 599
0b6d1084
JH
600One or more embedded pattern-match modifiers, to be turned on (or
601turned off, if preceded by C<->) for the remainder of the pattern or
602the remainder of the enclosing pattern group (if any). This is
603particularly useful for dynamic patterns, such as those read in from a
0d017f4d
WL
604configuration file, taken from an argument, or specified in a table
605somewhere. Consider the case where some patterns want to be case
606sensitive and some do not: The case insensitive ones merely need to
607include C<(?i)> at the front of the pattern. For example:
19799a22
GS
608
609 $pattern = "foobar";
5d458dd8 610 if ( /$pattern/i ) { }
19799a22
GS
611
612 # more flexible:
613
614 $pattern = "(?i)foobar";
5d458dd8 615 if ( /$pattern/ ) { }
19799a22 616
0b6d1084 617These modifiers are restored at the end of the enclosing group. For example,
19799a22 618
d8b950dc 619 ( (?i) blah ) \s+ \g1
19799a22 620
0d017f4d
WL
621will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
622repetition of the previous word, assuming the C</x> modifier, and no C</i>
623modifier outside this group.
19799a22 624
8eb5594e
DR
625These modifiers do not carry over into named subpatterns called in the
626enclosing group. In other words, a pattern such as C<((?i)(&NAME))> does not
627change the case-sensitivity of the "NAME" pattern.
628
5530442b 629Note that the C<p> modifier is special in that it can only be enabled,
cde0cee5 630not disabled, and that its presence anywhere in a pattern has a global
5530442b 631effect. Thus C<(?-p)> and C<(?-p:...)> are meaningless and will warn
cde0cee5
YO
632when executed under C<use warnings>.
633
5a964f20 634=item C<(?:pattern)>
d74e8afc 635X<(?:)>
a0d0e21e 636
ca9dfc88
IZ
637=item C<(?imsx-imsx:pattern)>
638
5a964f20
TC
639This is for clustering, not capturing; it groups subexpressions like
640"()", but doesn't make backreferences as "()" does. So
a0d0e21e 641
5a964f20 642 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
643
644is like
645
5a964f20 646 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 647
19799a22
GS
648but doesn't spit out extra fields. It's also cheaper not to capture
649characters if you don't need to.
a0d0e21e 650
19799a22 651Any letters between C<?> and C<:> act as flags modifiers as with
5d458dd8 652C<(?imsx-imsx)>. For example,
ca9dfc88
IZ
653
654 /(?s-i:more.*than).*million/i
655
14218588 656is equivalent to the more verbose
ca9dfc88
IZ
657
658 /(?:(?s-i)more.*than).*million/i
659
594d7033
YO
660=item C<(?|pattern)>
661X<(?|)> X<Branch reset>
662
663This is the "branch reset" pattern, which has the special property
c27a5cfe 664that the capture groups are numbered from the same starting point
99d59c4d 665in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 666
c27a5cfe 667Capture groups are numbered from left to right, but inside this
693596a8 668construct the numbering is restarted for each branch.
4deaaa80 669
c27a5cfe 670The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
671following this construct will be numbered as though the construct
672contained only one branch, that being the one with the most capture
c27a5cfe 673groups in it.
4deaaa80
PJ
674
675This construct will be useful when you want to capture one of a
676number of alternative matches.
677
678Consider the following pattern. The numbers underneath show in
c27a5cfe 679which group the captured content will be stored.
594d7033
YO
680
681
682 # before ---------------branch-reset----------- after
683 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
684 # 1 2 2 3 2 3 4
685
ab106183
A
686Be careful when using the branch reset pattern in combination with
687named captures. Named captures are implemented as being aliases to
c27a5cfe 688numbered groups holding the captures, and that interferes with the
ab106183
A
689implementation of the branch reset pattern. If you are using named
690captures in a branch reset pattern, it's best to use the same names,
691in the same order, in each of the alternations:
692
693 /(?| (?<a> x ) (?<b> y )
694 | (?<a> z ) (?<b> w )) /x
695
696Not doing so may lead to surprises:
697
698 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
699 say $+ {a}; # Prints '12'
700 say $+ {b}; # *Also* prints '12'.
701
c27a5cfe
KW
702The problem here is that both the group named C<< a >> and the group
703named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 704
ee9b8eae
YO
705=item Look-Around Assertions
706X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
707
708Look-around assertions are zero width patterns which match a specific
709pattern without including it in C<$&>. Positive assertions match when
710their subpattern matches, negative assertions match when their subpattern
711fails. Look-behind matches text up to the current match position,
712look-ahead matches text following the current match position.
713
714=over 4
715
5a964f20 716=item C<(?=pattern)>
d74e8afc 717X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 718
19799a22 719A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
720matches a word followed by a tab, without including the tab in C<$&>.
721
5a964f20 722=item C<(?!pattern)>
d74e8afc 723X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 724
19799a22 725A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 726matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
727however that look-ahead and look-behind are NOT the same thing. You cannot
728use this for look-behind.
7b8d334a 729
5a964f20 730If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
731will not do what you want. That's because the C<(?!foo)> is just saying that
732the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
733match. You would have to do something like C</(?!foo)...bar/> for that. We
734say "like" because there's the case of your "bar" not having three characters
735before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
736Sometimes it's still easier just to say:
a0d0e21e 737
a3cb178b 738 if (/bar/ && $` !~ /foo$/)
a0d0e21e 739
19799a22 740For look-behind see below.
c277df42 741
ee9b8eae
YO
742=item C<(?<=pattern)> C<\K>
743X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 744
c47ff5f1 745A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
746matches a word that follows a tab, without including the tab in C<$&>.
747Works only for fixed-width look-behind.
c277df42 748
ee9b8eae
YO
749There is a special form of this construct, called C<\K>, which causes the
750regex engine to "keep" everything it had matched prior to the C<\K> and
751not include it in C<$&>. This effectively provides variable length
752look-behind. The use of C<\K> inside of another look-around assertion
753is allowed, but the behaviour is currently not well defined.
754
c62285ac 755For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
756equivalent C<< (?<=...) >> construct, and it is especially useful in
757situations where you want to efficiently remove something following
758something else in a string. For instance
759
760 s/(foo)bar/$1/g;
761
762can be rewritten as the much more efficient
763
764 s/foo\Kbar//g;
765
5a964f20 766=item C<(?<!pattern)>
d74e8afc 767X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 768
19799a22
GS
769A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
770matches any occurrence of "foo" that does not follow "bar". Works
771only for fixed-width look-behind.
c277df42 772
ee9b8eae
YO
773=back
774
81714fb9
YO
775=item C<(?'NAME'pattern)>
776
777=item C<< (?<NAME>pattern) >>
778X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
779
c27a5cfe 780A named capture group. Identical in every respect to normal capturing
90a18110 781parentheses C<()> but for the additional fact that C<%+> or C<%-> may be
c27a5cfe 782used after a successful match to refer to a named group. See C<perlvar>
90a18110 783for more details on the C<%+> and C<%-> hashes.
81714fb9 784
c27a5cfe
KW
785If multiple distinct capture groups have the same name then the
786$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 787
0d017f4d 788The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
789
790B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 791function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
792numbered sequentially regardless of being named or not. Thus in the
793pattern
794
795 /(x)(?<foo>y)(z)/
796
797$+{foo} will be the same as $2, and $3 will contain 'z' instead of
798the opposite which is what a .NET regex hacker might expect.
799
1f1031fe
YO
800Currently NAME is restricted to simple identifiers only.
801In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
802its Unicode extension (see L<utf8>),
803though it isn't extended by the locale (see L<perllocale>).
81714fb9 804
1f1031fe 805B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 806with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 807may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 808support the use of single quotes as a delimiter for the name.
81714fb9 809
1f1031fe
YO
810=item C<< \k<NAME> >>
811
812=item C<< \k'NAME' >>
81714fb9
YO
813
814Named backreference. Similar to numeric backreferences, except that
815the group is designated by name and not number. If multiple groups
816have the same name then it refers to the leftmost defined group in
817the current match.
818
0d017f4d 819It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
820earlier in the pattern.
821
822Both forms are equivalent.
823
1f1031fe 824B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 825with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 826may be used instead of C<< \k<NAME> >>.
1f1031fe 827
cc6b7395 828=item C<(?{ code })>
d74e8afc 829X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 830
19799a22 831B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
832experimental, and may be changed without notice. Code executed that
833has side effects may not perform identically from version to version
834due to the effect of future optimisations in the regex engine.
c277df42 835
cc46d5f2 836This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
837always succeeds, and its C<code> is not interpolated. Currently,
838the rules to determine where the C<code> ends are somewhat convoluted.
839
77ea4f6d
JV
840This feature can be used together with the special variable C<$^N> to
841capture the results of submatches in variables without having to keep
842track of the number of nested parentheses. For example:
843
844 $_ = "The brown fox jumps over the lazy dog";
845 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
846 print "color = $color, animal = $animal\n";
847
754091cb
RGS
848Inside the C<(?{...})> block, C<$_> refers to the string the regular
849expression is matching against. You can also use C<pos()> to know what is
fa11829f 850the current position of matching within this string.
754091cb 851
19799a22
GS
852The C<code> is properly scoped in the following sense: If the assertion
853is backtracked (compare L<"Backtracking">), all changes introduced after
854C<local>ization are undone, so that
b9ac3b5b
GS
855
856 $_ = 'a' x 8;
5d458dd8 857 m<
f793d64a 858 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 859 (
5d458dd8 860 a
b9ac3b5b 861 (?{
f793d64a 862 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
b9ac3b5b 863 })
5d458dd8 864 )*
b9ac3b5b 865 aaaa
f793d64a
KW
866 (?{ $res = $cnt }) # On success copy to
867 # non-localized location.
b9ac3b5b
GS
868 >x;
869
0d017f4d 870will set C<$res = 4>. Note that after the match, C<$cnt> returns to the globally
14218588 871introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
872are unwound.
873
19799a22
GS
874This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
875switch. If I<not> used in this way, the result of evaluation of
876C<code> is put into the special variable C<$^R>. This happens
877immediately, so C<$^R> can be used from other C<(?{ code })> assertions
878inside the same regular expression.
b9ac3b5b 879
19799a22
GS
880The assignment to C<$^R> above is properly localized, so the old
881value of C<$^R> is restored if the assertion is backtracked; compare
882L<"Backtracking">.
b9ac3b5b 883
19799a22
GS
884For reasons of security, this construct is forbidden if the regular
885expression involves run-time interpolation of variables, unless the
886perilous C<use re 'eval'> pragma has been used (see L<re>), or the
887variables contain results of C<qr//> operator (see
9bb1f947 888L<perlop/"qr/STRINGE<sol>msixpo">).
871b0233 889
0d017f4d 890This restriction is due to the wide-spread and remarkably convenient
19799a22 891custom of using run-time determined strings as patterns. For example:
871b0233
IZ
892
893 $re = <>;
894 chomp $re;
895 $string =~ /$re/;
896
14218588
GS
897Before Perl knew how to execute interpolated code within a pattern,
898this operation was completely safe from a security point of view,
899although it could raise an exception from an illegal pattern. If
900you turn on the C<use re 'eval'>, though, it is no longer secure,
901so you should only do so if you are also using taint checking.
902Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 903compartment. See L<perlsec> for details about both these mechanisms.
871b0233 904
e95d7314
GG
905B<WARNING>: Use of lexical (C<my>) variables in these blocks is
906broken. The result is unpredictable and will make perl unstable. The
907workaround is to use global (C<our>) variables.
908
909B<WARNING>: Because Perl's regex engine is currently not re-entrant,
910interpolated code may not invoke the regex engine either directly with
911C<m//> or C<s///>), or indirectly with functions such as
912C<split>. Invoking the regex engine in these blocks will make perl
913unstable.
8988a1bb 914
14455d6c 915=item C<(??{ code })>
d74e8afc
ITB
916X<(??{})>
917X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 918
19799a22 919B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
920experimental, and may be changed without notice. Code executed that
921has side effects may not perform identically from version to version
922due to the effect of future optimisations in the regex engine.
0f5d15d6 923
19799a22
GS
924This is a "postponed" regular subexpression. The C<code> is evaluated
925at run time, at the moment this subexpression may match. The result
926of evaluation is considered as a regular expression and matched as
61528107 927if it were inserted instead of this construct. Note that this means
c27a5cfe 928that the contents of capture groups defined inside an eval'ed pattern
6bda09f9 929are not available outside of the pattern, and vice versa, there is no
c27a5cfe 930way for the inner pattern to refer to a capture group defined outside.
6bda09f9
YO
931Thus,
932
933 ('a' x 100)=~/(??{'(.)' x 100})/
934
81714fb9 935B<will> match, it will B<not> set $1.
0f5d15d6 936
428594d9 937The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
938where the C<code> ends are currently somewhat convoluted.
939
940The following pattern matches a parenthesized group:
0f5d15d6
IZ
941
942 $re = qr{
f793d64a
KW
943 \(
944 (?:
945 (?> [^()]+ ) # Non-parens without backtracking
946 |
947 (??{ $re }) # Group with matching parens
948 )*
949 \)
950 }x;
0f5d15d6 951
6bda09f9
YO
952See also C<(?PARNO)> for a different, more efficient way to accomplish
953the same task.
954
0b370c0a
A
955For reasons of security, this construct is forbidden if the regular
956expression involves run-time interpolation of variables, unless the
957perilous C<use re 'eval'> pragma has been used (see L<re>), or the
958variables contain results of C<qr//> operator (see
9bb1f947 959L<perlop/"qrE<sol>STRINGE<sol>msixpo">).
0b370c0a 960
5d458dd8 961Because perl's regex engine is not currently re-entrant, delayed
8988a1bb
DD
962code may not invoke the regex engine either directly with C<m//> or C<s///>),
963or indirectly with functions such as C<split>.
964
5d458dd8
YO
965Recursing deeper than 50 times without consuming any input string will
966result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
967changing it requires a custom build.
968
542fa716
YO
969=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
970X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 971X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 972X<regex, relative recursion>
6bda09f9 973
81714fb9 974Similar to C<(??{ code })> except it does not involve compiling any code,
c27a5cfe
KW
975instead it treats the contents of a capture group as an independent
976pattern that must match at the current position. Capture groups
81714fb9 977contained by the pattern will have the value as determined by the
6bda09f9
YO
978outermost recursion.
979
894be9b7 980PARNO is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 981the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 982the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716 983C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
c27a5cfe 984to be relative, with negative numbers indicating preceding capture groups
542fa716 985and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 986declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 987Note that the counting for relative recursion differs from that of
c27a5cfe 988relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 989included.
6bda09f9 990
81714fb9 991The following pattern matches a function foo() which may contain
f145b7e9 992balanced parentheses as the argument.
6bda09f9
YO
993
994 $re = qr{ ( # paren group 1 (full function)
81714fb9 995 foo
6bda09f9
YO
996 ( # paren group 2 (parens)
997 \(
998 ( # paren group 3 (contents of parens)
999 (?:
1000 (?> [^()]+ ) # Non-parens without backtracking
1001 |
1002 (?2) # Recurse to start of paren group 2
1003 )*
1004 )
1005 \)
1006 )
1007 )
1008 }x;
1009
1010If the pattern was used as follows
1011
1012 'foo(bar(baz)+baz(bop))'=~/$re/
1013 and print "\$1 = $1\n",
1014 "\$2 = $2\n",
1015 "\$3 = $3\n";
1016
1017the output produced should be the following:
1018
1019 $1 = foo(bar(baz)+baz(bop))
1020 $2 = (bar(baz)+baz(bop))
81714fb9 1021 $3 = bar(baz)+baz(bop)
6bda09f9 1022
c27a5cfe 1023If there is no corresponding capture group defined, then it is a
61528107 1024fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1025string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1026into perl, so changing it requires a custom build.
1027
542fa716
YO
1028The following shows how using negative indexing can make it
1029easier to embed recursive patterns inside of a C<qr//> construct
1030for later use:
1031
1032 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1033 if (/foo $parens \s+ + \s+ bar $parens/x) {
1034 # do something here...
1035 }
1036
81714fb9 1037B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1038PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1039a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1040as atomic. Also, modifiers are resolved at compile time, so constructs
1041like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1042be processed.
6bda09f9 1043
894be9b7
YO
1044=item C<(?&NAME)>
1045X<(?&NAME)>
1046
0d017f4d
WL
1047Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1048parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1049the same name, then it recurses to the leftmost.
1050
1051It is an error to refer to a name that is not declared somewhere in the
1052pattern.
1053
1f1031fe
YO
1054B<NOTE:> In order to make things easier for programmers with experience
1055with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1056may be used instead of C<< (?&NAME) >>.
1f1031fe 1057
e2e6a0f1
YO
1058=item C<(?(condition)yes-pattern|no-pattern)>
1059X<(?()>
286f584a 1060
e2e6a0f1 1061=item C<(?(condition)yes-pattern)>
286f584a 1062
e2e6a0f1
YO
1063Conditional expression. C<(condition)> should be either an integer in
1064parentheses (which is valid if the corresponding pair of parentheses
1065matched), a look-ahead/look-behind/evaluate zero-width assertion, a
c27a5cfe 1066name in angle brackets or single quotes (which is valid if a group
e2e6a0f1
YO
1067with the given name matched), or the special symbol (R) (true when
1068evaluated inside of recursion or eval). Additionally the R may be
1069followed by a number, (which will be true when evaluated when recursing
1070inside of the appropriate group), or by C<&NAME>, in which case it will
1071be true only when evaluated during recursion in the named group.
1072
1073Here's a summary of the possible predicates:
1074
1075=over 4
1076
1077=item (1) (2) ...
1078
c27a5cfe 1079Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1080
1081=item (<NAME>) ('NAME')
1082
c27a5cfe 1083Checks if a group with the given name has matched something.
e2e6a0f1
YO
1084
1085=item (?{ CODE })
1086
1087Treats the code block as the condition.
1088
1089=item (R)
1090
1091Checks if the expression has been evaluated inside of recursion.
1092
1093=item (R1) (R2) ...
1094
1095Checks if the expression has been evaluated while executing directly
1096inside of the n-th capture group. This check is the regex equivalent of
1097
1098 if ((caller(0))[3] eq 'subname') { ... }
1099
1100In other words, it does not check the full recursion stack.
1101
1102=item (R&NAME)
1103
1104Similar to C<(R1)>, this predicate checks to see if we're executing
1105directly inside of the leftmost group with a given name (this is the same
1106logic used by C<(?&NAME)> to disambiguate). It does not check the full
1107stack, but only the name of the innermost active recursion.
1108
1109=item (DEFINE)
1110
1111In this case, the yes-pattern is never directly executed, and no
1112no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1113See below for details.
1114
1115=back
1116
1117For example:
1118
1119 m{ ( \( )?
1120 [^()]+
1121 (?(1) \) )
1122 }x
1123
1124matches a chunk of non-parentheses, possibly included in parentheses
1125themselves.
1126
1127A special form is the C<(DEFINE)> predicate, which never executes directly
1128its yes-pattern, and does not allow a no-pattern. This allows to define
1129subpatterns which will be executed only by using the recursion mechanism.
1130This way, you can define a set of regular expression rules that can be
1131bundled into any pattern you choose.
1132
1133It is recommended that for this usage you put the DEFINE block at the
1134end of the pattern, and that you name any subpatterns defined within it.
1135
1136Also, it's worth noting that patterns defined this way probably will
1137not be as efficient, as the optimiser is not very clever about
1138handling them.
1139
1140An example of how this might be used is as follows:
1141
2bf803e2 1142 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1143 (?(DEFINE)
2bf803e2
YO
1144 (?<NAME_PAT>....)
1145 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1146 )/x
1147
c27a5cfe
KW
1148Note that capture groups matched inside of recursion are not accessible
1149after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1150necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1151C<$+{NAME}> would be.
286f584a 1152
c47ff5f1 1153=item C<< (?>pattern) >>
6bda09f9 1154X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1155
19799a22
GS
1156An "independent" subexpression, one which matches the substring
1157that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1158position, and it matches I<nothing other than this substring>. This
19799a22
GS
1159construct is useful for optimizations of what would otherwise be
1160"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1161It may also be useful in places where the "grab all you can, and do not
1162give anything back" semantic is desirable.
19799a22 1163
c47ff5f1 1164For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1165(anchored at the beginning of string, as above) will match I<all>
1166characters C<a> at the beginning of string, leaving no C<a> for
1167C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1168since the match of the subgroup C<a*> is influenced by the following
1169group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1170C<a*ab> will match fewer characters than a standalone C<a*>, since
1171this makes the tail match.
1172
c47ff5f1 1173An effect similar to C<< (?>pattern) >> may be achieved by writing
d8b950dc
KW
1174C<(?=(pattern))\g1>. This matches the same substring as a standalone
1175C<a+>, and the following C<\g1> eats the matched string; it therefore
c47ff5f1 1176makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1177(The difference between these two constructs is that the second one
1178uses a capturing group, thus shifting ordinals of backreferences
1179in the rest of a regular expression.)
1180
1181Consider this pattern:
c277df42 1182
871b0233 1183 m{ \(
e2e6a0f1 1184 (
f793d64a 1185 [^()]+ # x+
e2e6a0f1 1186 |
871b0233
IZ
1187 \( [^()]* \)
1188 )+
e2e6a0f1 1189 \)
871b0233 1190 }x
5a964f20 1191
19799a22
GS
1192That will efficiently match a nonempty group with matching parentheses
1193two levels deep or less. However, if there is no such group, it
1194will take virtually forever on a long string. That's because there
1195are so many different ways to split a long string into several
1196substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1197to a subpattern of the above pattern. Consider how the pattern
1198above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1199seconds, but that each extra letter doubles this time. This
1200exponential performance will make it appear that your program has
14218588 1201hung. However, a tiny change to this pattern
5a964f20 1202
e2e6a0f1
YO
1203 m{ \(
1204 (
f793d64a 1205 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1206 |
871b0233
IZ
1207 \( [^()]* \)
1208 )+
e2e6a0f1 1209 \)
871b0233 1210 }x
c277df42 1211
c47ff5f1 1212which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1213this yourself would be a productive exercise), but finishes in a fourth
1214the time when used on a similar string with 1000000 C<a>s. Be aware,
1215however, that this pattern currently triggers a warning message under
9f1b1f2d 1216the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1217C<"matches null string many times in regex">.
c277df42 1218
c47ff5f1 1219On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1220effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1221This was only 4 times slower on a string with 1000000 C<a>s.
1222
9da458fc
IZ
1223The "grab all you can, and do not give anything back" semantic is desirable
1224in many situations where on the first sight a simple C<()*> looks like
1225the correct solution. Suppose we parse text with comments being delimited
1226by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1227its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1228the comment delimiter, because it may "give up" some whitespace if
1229the remainder of the pattern can be made to match that way. The correct
1230answer is either one of these:
1231
1232 (?>#[ \t]*)
1233 #[ \t]*(?![ \t])
1234
1235For example, to grab non-empty comments into $1, one should use either
1236one of these:
1237
1238 / (?> \# [ \t]* ) ( .+ ) /x;
1239 / \# [ \t]* ( [^ \t] .* ) /x;
1240
1241Which one you pick depends on which of these expressions better reflects
1242the above specification of comments.
1243
6bda09f9
YO
1244In some literature this construct is called "atomic matching" or
1245"possessive matching".
1246
b9b4dddf
YO
1247Possessive quantifiers are equivalent to putting the item they are applied
1248to inside of one of these constructs. The following equivalences apply:
1249
1250 Quantifier Form Bracketing Form
1251 --------------- ---------------
1252 PAT*+ (?>PAT*)
1253 PAT++ (?>PAT+)
1254 PAT?+ (?>PAT?)
1255 PAT{min,max}+ (?>PAT{min,max})
1256
e2e6a0f1
YO
1257=back
1258
1259=head2 Special Backtracking Control Verbs
1260
1261B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1262removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1263be noted to avoid problems during upgrades.
1264
1265These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1266otherwise stated the ARG argument is optional; in some cases, it is
1267forbidden.
1268
1269Any pattern containing a special backtracking verb that allows an argument
e1020413 1270has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1271C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1272rules apply:
e2e6a0f1 1273
5d458dd8
YO
1274On failure, the C<$REGERROR> variable will be set to the ARG value of the
1275verb pattern, if the verb was involved in the failure of the match. If the
1276ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1277name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1278none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1279
5d458dd8
YO
1280On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1281the C<$REGMARK> variable will be set to the name of the last
1282C<(*MARK:NAME)> pattern executed. See the explanation for the
1283C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1284
5d458dd8
YO
1285B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1286and most other regex related variables. They are not local to a scope, nor
1287readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1288Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1289
1290If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1291argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1
YO
1292
1293=over 4
1294
1295=item Verbs that take an argument
1296
1297=over 4
1298
5d458dd8 1299=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1300X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1301
5d458dd8
YO
1302This zero-width pattern prunes the backtracking tree at the current point
1303when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1304where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1305A may backtrack as necessary to match. Once it is reached, matching
1306continues in B, which may also backtrack as necessary; however, should B
1307not match, then no further backtracking will take place, and the pattern
1308will fail outright at the current starting position.
54612592
YO
1309
1310The following example counts all the possible matching strings in a
1311pattern (without actually matching any of them).
1312
e2e6a0f1 1313 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1314 print "Count=$count\n";
1315
1316which produces:
1317
1318 aaab
1319 aaa
1320 aa
1321 a
1322 aab
1323 aa
1324 a
1325 ab
1326 a
1327 Count=9
1328
5d458dd8 1329If we add a C<(*PRUNE)> before the count like the following
54612592 1330
5d458dd8 1331 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1332 print "Count=$count\n";
1333
1334we prevent backtracking and find the count of the longest matching
353c6505 1335at each matching starting point like so:
54612592
YO
1336
1337 aaab
1338 aab
1339 ab
1340 Count=3
1341
5d458dd8 1342Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1343
5d458dd8
YO
1344See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1345control backtracking. In some cases, the use of C<(*PRUNE)> can be
1346replaced with a C<< (?>pattern) >> with no functional difference; however,
1347C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1348C<< (?>pattern) >> alone.
54612592 1349
e2e6a0f1 1350
5d458dd8
YO
1351=item C<(*SKIP)> C<(*SKIP:NAME)>
1352X<(*SKIP)>
e2e6a0f1 1353
5d458dd8 1354This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1355failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1356to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1357of this pattern. This effectively means that the regex engine "skips" forward
1358to this position on failure and tries to match again, (assuming that
1359there is sufficient room to match).
1360
1361The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1362C<(*MARK:NAME)> was encountered while matching, then it is that position
1363which is used as the "skip point". If no C<(*MARK)> of that name was
1364encountered, then the C<(*SKIP)> operator has no effect. When used
1365without a name the "skip point" is where the match point was when
1366executing the (*SKIP) pattern.
1367
1368Compare the following to the examples in C<(*PRUNE)>, note the string
24b23f37
YO
1369is twice as long:
1370
5d458dd8 1371 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1372 print "Count=$count\n";
1373
1374outputs
1375
1376 aaab
1377 aaab
1378 Count=2
1379
5d458dd8 1380Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1381executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1382C<(*SKIP)> was executed.
1383
5d458dd8
YO
1384=item C<(*MARK:NAME)> C<(*:NAME)>
1385X<(*MARK)> C<(*MARK:NAME)> C<(*:NAME)>
1386
1387This zero-width pattern can be used to mark the point reached in a string
1388when a certain part of the pattern has been successfully matched. This
1389mark may be given a name. A later C<(*SKIP)> pattern will then skip
1390forward to that point if backtracked into on failure. Any number of
b4222fa9 1391C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1392
1393In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1394can be used to "label" a pattern branch, so that after matching, the
1395program can determine which branches of the pattern were involved in the
1396match.
1397
1398When a match is successful, the C<$REGMARK> variable will be set to the
1399name of the most recently executed C<(*MARK:NAME)> that was involved
1400in the match.
1401
1402This can be used to determine which branch of a pattern was matched
c27a5cfe 1403without using a separate capture group for each branch, which in turn
5d458dd8
YO
1404can result in a performance improvement, as perl cannot optimize
1405C</(?:(x)|(y)|(z))/> as efficiently as something like
1406C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1407
1408When a match has failed, and unless another verb has been involved in
1409failing the match and has provided its own name to use, the C<$REGERROR>
1410variable will be set to the name of the most recently executed
1411C<(*MARK:NAME)>.
1412
1413See C<(*SKIP)> for more details.
1414
b62d2d15
YO
1415As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1416
5d458dd8
YO
1417=item C<(*THEN)> C<(*THEN:NAME)>
1418
241e7389 1419This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1420C<(*PRUNE)>, this verb always matches, and when backtracked into on
1421failure, it causes the regex engine to try the next alternation in the
1422innermost enclosing group (capturing or otherwise).
1423
1424Its name comes from the observation that this operation combined with the
1425alternation operator (C<|>) can be used to create what is essentially a
1426pattern-based if/then/else block:
1427
1428 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1429
1430Note that if this operator is used and NOT inside of an alternation then
1431it acts exactly like the C<(*PRUNE)> operator.
1432
1433 / A (*PRUNE) B /
1434
1435is the same as
1436
1437 / A (*THEN) B /
1438
1439but
1440
1441 / ( A (*THEN) B | C (*THEN) D ) /
1442
1443is not the same as
1444
1445 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1446
1447as after matching the A but failing on the B the C<(*THEN)> verb will
1448backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1449
e2e6a0f1
YO
1450=item C<(*COMMIT)>
1451X<(*COMMIT)>
24b23f37 1452
241e7389 1453This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1454zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1455into on failure it causes the match to fail outright. No further attempts
1456to find a valid match by advancing the start pointer will occur again.
1457For example,
24b23f37 1458
e2e6a0f1 1459 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1460 print "Count=$count\n";
1461
1462outputs
1463
1464 aaab
1465 Count=1
1466
e2e6a0f1
YO
1467In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1468does not match, the regex engine will not try any further matching on the
1469rest of the string.
c277df42 1470
e2e6a0f1 1471=back
9af228c6 1472
e2e6a0f1 1473=item Verbs without an argument
9af228c6
YO
1474
1475=over 4
1476
e2e6a0f1
YO
1477=item C<(*FAIL)> C<(*F)>
1478X<(*FAIL)> X<(*F)>
9af228c6 1479
e2e6a0f1
YO
1480This pattern matches nothing and always fails. It can be used to force the
1481engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1482fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1483
e2e6a0f1 1484It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1485
e2e6a0f1
YO
1486=item C<(*ACCEPT)>
1487X<(*ACCEPT)>
9af228c6 1488
e2e6a0f1
YO
1489B<WARNING:> This feature is highly experimental. It is not recommended
1490for production code.
9af228c6 1491
e2e6a0f1
YO
1492This pattern matches nothing and causes the end of successful matching at
1493the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1494whether there is actually more to match in the string. When inside of a
0d017f4d 1495nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1496via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1497
c27a5cfe 1498If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1499marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1500For instance:
9af228c6 1501
e2e6a0f1 1502 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1503
e2e6a0f1 1504will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0d017f4d 1505be set. If another branch in the inner parentheses were matched, such as in the
e2e6a0f1 1506string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1507
1508=back
c277df42 1509
a0d0e21e
LW
1510=back
1511
c07a80fd 1512=head2 Backtracking
d74e8afc 1513X<backtrack> X<backtracking>
c07a80fd 1514
35a734be
IZ
1515NOTE: This section presents an abstract approximation of regular
1516expression behavior. For a more rigorous (and complicated) view of
1517the rules involved in selecting a match among possible alternatives,
0d017f4d 1518see L<Combining RE Pieces>.
35a734be 1519
c277df42 1520A fundamental feature of regular expression matching involves the
5a964f20 1521notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1522by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1523C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1524internally, but the general principle outlined here is valid.
c07a80fd
PP
1525
1526For a regular expression to match, the I<entire> regular expression must
1527match, not just part of it. So if the beginning of a pattern containing a
1528quantifier succeeds in a way that causes later parts in the pattern to
1529fail, the matching engine backs up and recalculates the beginning
1530part--that's why it's called backtracking.
1531
1532Here is an example of backtracking: Let's say you want to find the
1533word following "foo" in the string "Food is on the foo table.":
1534
1535 $_ = "Food is on the foo table.";
1536 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 1537 print "$2 follows $1.\n";
c07a80fd
PP
1538 }
1539
1540When the match runs, the first part of the regular expression (C<\b(foo)>)
1541finds a possible match right at the beginning of the string, and loads up
1542$1 with "Foo". However, as soon as the matching engine sees that there's
1543no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1544mistake and starts over again one character after where it had the
c07a80fd
PP
1545tentative match. This time it goes all the way until the next occurrence
1546of "foo". The complete regular expression matches this time, and you get
1547the expected output of "table follows foo."
1548
1549Sometimes minimal matching can help a lot. Imagine you'd like to match
1550everything between "foo" and "bar". Initially, you write something
1551like this:
1552
1553 $_ = "The food is under the bar in the barn.";
1554 if ( /foo(.*)bar/ ) {
f793d64a 1555 print "got <$1>\n";
c07a80fd
PP
1556 }
1557
1558Which perhaps unexpectedly yields:
1559
1560 got <d is under the bar in the >
1561
1562That's because C<.*> was greedy, so you get everything between the
14218588 1563I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1564to use minimal matching to make sure you get the text between a "foo"
1565and the first "bar" thereafter.
1566
1567 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1568 got <d is under the >
1569
0d017f4d 1570Here's another example. Let's say you'd like to match a number at the end
b6e13d97 1571of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
1572So you write this:
1573
1574 $_ = "I have 2 numbers: 53147";
f793d64a
KW
1575 if ( /(.*)(\d*)/ ) { # Wrong!
1576 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
1577 }
1578
1579That won't work at all, because C<.*> was greedy and gobbled up the
1580whole string. As C<\d*> can match on an empty string the complete
1581regular expression matched successfully.
1582
8e1088bc 1583 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
1584
1585Here are some variants, most of which don't work:
1586
1587 $_ = "I have 2 numbers: 53147";
1588 @pats = qw{
f793d64a
KW
1589 (.*)(\d*)
1590 (.*)(\d+)
1591 (.*?)(\d*)
1592 (.*?)(\d+)
1593 (.*)(\d+)$
1594 (.*?)(\d+)$
1595 (.*)\b(\d+)$
1596 (.*\D)(\d+)$
c07a80fd
PP
1597 };
1598
1599 for $pat (@pats) {
f793d64a
KW
1600 printf "%-12s ", $pat;
1601 if ( /$pat/ ) {
1602 print "<$1> <$2>\n";
1603 } else {
1604 print "FAIL\n";
1605 }
c07a80fd
PP
1606 }
1607
1608That will print out:
1609
1610 (.*)(\d*) <I have 2 numbers: 53147> <>
1611 (.*)(\d+) <I have 2 numbers: 5314> <7>
1612 (.*?)(\d*) <> <>
1613 (.*?)(\d+) <I have > <2>
1614 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1615 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1616 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1617 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1618
1619As you see, this can be a bit tricky. It's important to realize that a
1620regular expression is merely a set of assertions that gives a definition
1621of success. There may be 0, 1, or several different ways that the
1622definition might succeed against a particular string. And if there are
5a964f20
TC
1623multiple ways it might succeed, you need to understand backtracking to
1624know which variety of success you will achieve.
c07a80fd 1625
19799a22 1626When using look-ahead assertions and negations, this can all get even
8b19b778 1627trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
1628followed by "123". You might try to write that as
1629
871b0233 1630 $_ = "ABC123";
f793d64a
KW
1631 if ( /^\D*(?!123)/ ) { # Wrong!
1632 print "Yup, no 123 in $_\n";
871b0233 1633 }
c07a80fd
PP
1634
1635But that isn't going to match; at least, not the way you're hoping. It
1636claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1637why that pattern matches, contrary to popular expectations:
c07a80fd 1638
4358a253
SS
1639 $x = 'ABC123';
1640 $y = 'ABC445';
c07a80fd 1641
4358a253
SS
1642 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1643 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1644
4358a253
SS
1645 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1646 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
1647
1648This prints
1649
1650 2: got ABC
1651 3: got AB
1652 4: got ABC
1653
5f05dabc 1654You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
1655general purpose version of test 1. The important difference between
1656them is that test 3 contains a quantifier (C<\D*>) and so can use
1657backtracking, whereas test 1 will not. What's happening is
1658that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1659non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1660let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1661fail.
14218588 1662
c07a80fd 1663The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1664try to match C<(?!123> with "123", which fails. But because
c07a80fd
PP
1665a quantifier (C<\D*>) has been used in the regular expression, the
1666search engine can backtrack and retry the match differently
54310121 1667in the hope of matching the complete regular expression.
c07a80fd 1668
5a964f20
TC
1669The pattern really, I<really> wants to succeed, so it uses the
1670standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1671time. Now there's indeed something following "AB" that is not
14218588 1672"123". It's "C123", which suffices.
c07a80fd 1673
14218588
GS
1674We can deal with this by using both an assertion and a negation.
1675We'll say that the first part in $1 must be followed both by a digit
1676and by something that's not "123". Remember that the look-aheads
1677are zero-width expressions--they only look, but don't consume any
1678of the string in their match. So rewriting this way produces what
c07a80fd
PP
1679you'd expect; that is, case 5 will fail, but case 6 succeeds:
1680
4358a253
SS
1681 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1682 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
1683
1684 6: got ABC
1685
5a964f20 1686In other words, the two zero-width assertions next to each other work as though
19799a22 1687they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
1688matches only if you're at the beginning of the line AND the end of the
1689line simultaneously. The deeper underlying truth is that juxtaposition in
1690regular expressions always means AND, except when you write an explicit OR
1691using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1692although the attempted matches are made at different positions because "a"
1693is not a zero-width assertion, but a one-width assertion.
1694
0d017f4d 1695B<WARNING>: Particularly complicated regular expressions can take
14218588 1696exponential time to solve because of the immense number of possible
0d017f4d 1697ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
1698internal optimizations done by the regular expression engine, this will
1699take a painfully long time to run:
c07a80fd 1700
e1901655
IZ
1701 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1702
1703And if you used C<*>'s in the internal groups instead of limiting them
1704to 0 through 5 matches, then it would take forever--or until you ran
1705out of stack space. Moreover, these internal optimizations are not
1706always applicable. For example, if you put C<{0,5}> instead of C<*>
1707on the external group, no current optimization is applicable, and the
1708match takes a long time to finish.
c07a80fd 1709
9da458fc
IZ
1710A powerful tool for optimizing such beasts is what is known as an
1711"independent group",
c47ff5f1 1712which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1713zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 1714the tail match, since they are in "logical" context: only
14218588 1715whether they match is considered relevant. For an example
9da458fc 1716where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1717following match, see L<C<< (?>pattern) >>>.
c277df42 1718
a0d0e21e 1719=head2 Version 8 Regular Expressions
d74e8afc 1720X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1721
5a964f20 1722In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1723routines, here are the pattern-matching rules not described above.
1724
54310121 1725Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1726with a special meaning described here or above. You can cause
5a964f20 1727characters that normally function as metacharacters to be interpreted
5f05dabc 1728literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
1729character; "\\" matches a "\"). This escape mechanism is also required
1730for the character used as the pattern delimiter.
1731
1732A series of characters matches that series of characters in the target
1733string, so the pattern C<blurfl> would match "blurfl" in the target
1734string.
a0d0e21e
LW
1735
1736You can specify a character class, by enclosing a list of characters
5d458dd8 1737in C<[]>, which will match any character from the list. If the
a0d0e21e 1738first character after the "[" is "^", the class matches any character not
14218588 1739in the list. Within a list, the "-" character specifies a
5a964f20 1740range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1741inclusive. If you want either "-" or "]" itself to be a member of a
1742class, put it at the start of the list (possibly after a "^"), or
1743escape it with a backslash. "-" is also taken literally when it is
1744at the end of the list, just before the closing "]". (The
84850974
DD
1745following all specify the same class of three characters: C<[-az]>,
1746C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
1747specifies a class containing twenty-six characters, even on EBCDIC-based
1748character sets.) Also, if you try to use the character
1749classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1750a range, the "-" is understood literally.
a0d0e21e 1751
8ada0baa
JH
1752Note also that the whole range idea is rather unportable between
1753character sets--and even within character sets they may cause results
1754you probably didn't expect. A sound principle is to use only ranges
0d017f4d 1755that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
1756[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1757spell out the character sets in full.
1758
54310121 1759Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1760used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1761"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
5d458dd8
YO
1762of octal digits, matches the character whose coded character set value
1763is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
1764matches the character whose numeric value is I<nn>. The expression \cI<x>
1765matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 1766matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1767
1768You can specify a series of alternatives for a pattern using "|" to
1769separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1770or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1771first alternative includes everything from the last pattern delimiter
1772("(", "[", or the beginning of the pattern) up to the first "|", and
1773the last alternative contains everything from the last "|" to the next
14218588
GS
1774pattern delimiter. That's why it's common practice to include
1775alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1776start and end.
1777
5a964f20 1778Alternatives are tried from left to right, so the first
a3cb178b
GS
1779alternative found for which the entire expression matches, is the one that
1780is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1781example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1782part will match, as that is the first alternative tried, and it successfully
1783matches the target string. (This might not seem important, but it is
1784important when you are capturing matched text using parentheses.)
1785
5a964f20 1786Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1787so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1788
14218588
GS
1789Within a pattern, you may designate subpatterns for later reference
1790by enclosing them in parentheses, and you may refer back to the
1791I<n>th subpattern later in the pattern using the metacharacter
1792\I<n>. Subpatterns are numbered based on the left to right order
1793of their opening parenthesis. A backreference matches whatever
1794actually matched the subpattern in the string being examined, not
d8b950dc 1795the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
1796match "0x1234 0x4321", but not "0x1234 01234", because subpattern
17971 matched "0x", even though the rule C<0|0x> could potentially match
1798the leading 0 in the second number.
cb1a09d0 1799
0d017f4d 1800=head2 Warning on \1 Instead of $1
cb1a09d0 1801
5a964f20 1802Some people get too used to writing things like:
cb1a09d0
AD
1803
1804 $pattern =~ s/(\W)/\\\1/g;
1805
3ff1c45a
KW
1806This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
1807shocking the
cb1a09d0 1808B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1809PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1810the usual double-quoted string means a control-A. The customary Unix
1811meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1812of doing that, you get yourself into trouble if you then add an C</e>
1813modifier.
1814
f793d64a 1815 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1816
1817Or if you try to do
1818
1819 s/(\d+)/\1000/;
1820
1821You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1822C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1823with the operation of matching a backreference. Certainly they mean two
1824different things on the I<left> side of the C<s///>.
9fa51da4 1825
0d017f4d 1826=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 1827
19799a22 1828B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1829
1830Regular expressions provide a terse and powerful programming language. As
1831with most other power tools, power comes together with the ability
1832to wreak havoc.
1833
1834A common abuse of this power stems from the ability to make infinite
628afcb5 1835loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1836
1837 'foo' =~ m{ ( o? )* }x;
1838
0d017f4d 1839The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 1840in the string is not moved by the match, C<o?> would match again and again
527e91da 1841because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
1842is with the looping modifier C<//g>:
1843
1844 @matches = ( 'foo' =~ m{ o? }xg );
1845
1846or
1847
1848 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1849
1850or the loop implied by split().
1851
1852However, long experience has shown that many programming tasks may
14218588
GS
1853be significantly simplified by using repeated subexpressions that
1854may match zero-length substrings. Here's a simple example being:
c84d73f1 1855
f793d64a 1856 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
1857 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1858
9da458fc 1859Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 1860the infinite loop>. The rules for this are different for lower-level
527e91da 1861loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
1862ones like the C</g> modifier or split() operator.
1863
19799a22
GS
1864The lower-level loops are I<interrupted> (that is, the loop is
1865broken) when Perl detects that a repeated expression matched a
1866zero-length substring. Thus
c84d73f1
IZ
1867
1868 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1869
5d458dd8 1870is made equivalent to
c84d73f1 1871
5d458dd8
YO
1872 m{ (?: NON_ZERO_LENGTH )*
1873 |
1874 (?: ZERO_LENGTH )?
c84d73f1
IZ
1875 }x;
1876
1877The higher level-loops preserve an additional state between iterations:
5d458dd8 1878whether the last match was zero-length. To break the loop, the following
c84d73f1 1879match after a zero-length match is prohibited to have a length of zero.
5d458dd8 1880This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
1881and so the I<second best> match is chosen if the I<best> match is of
1882zero length.
1883
19799a22 1884For example:
c84d73f1
IZ
1885
1886 $_ = 'bar';
1887 s/\w??/<$&>/g;
1888
20fb949f 1889results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 1890match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
1891best> match is what is matched by C<\w>. Thus zero-length matches
1892alternate with one-character-long matches.
1893
5d458dd8 1894Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
1895position one notch further in the string.
1896
19799a22 1897The additional state of being I<matched with zero-length> is associated with
c84d73f1 1898the matched string, and is reset by each assignment to pos().
9da458fc
IZ
1899Zero-length matches at the end of the previous match are ignored
1900during C<split>.
c84d73f1 1901
0d017f4d 1902=head2 Combining RE Pieces
35a734be
IZ
1903
1904Each of the elementary pieces of regular expressions which were described
1905before (such as C<ab> or C<\Z>) could match at most one substring
1906at the given position of the input string. However, in a typical regular
1907expression these elementary pieces are combined into more complicated
1908patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1909(in these examples C<S> and C<T> are regular subexpressions).
1910
1911Such combinations can include alternatives, leading to a problem of choice:
1912if we match a regular expression C<a|ab> against C<"abc">, will it match
1913substring C<"a"> or C<"ab">? One way to describe which substring is
1914actually matched is the concept of backtracking (see L<"Backtracking">).
1915However, this description is too low-level and makes you think
1916in terms of a particular implementation.
1917
1918Another description starts with notions of "better"/"worse". All the
1919substrings which may be matched by the given regular expression can be
1920sorted from the "best" match to the "worst" match, and it is the "best"
1921match which is chosen. This substitutes the question of "what is chosen?"
1922by the question of "which matches are better, and which are worse?".
1923
1924Again, for elementary pieces there is no such question, since at most
1925one match at a given position is possible. This section describes the
1926notion of better/worse for combining operators. In the description
1927below C<S> and C<T> are regular subexpressions.
1928
13a2d996 1929=over 4
35a734be
IZ
1930
1931=item C<ST>
1932
1933Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
1934substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 1935which can be matched by C<T>.
35a734be
IZ
1936
1937If C<A> is better match for C<S> than C<A'>, C<AB> is a better
1938match than C<A'B'>.
1939
1940If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
1941C<B> is better match for C<T> than C<B'>.
1942
1943=item C<S|T>
1944
1945When C<S> can match, it is a better match than when only C<T> can match.
1946
1947Ordering of two matches for C<S> is the same as for C<S>. Similar for
1948two matches for C<T>.
1949
1950=item C<S{REPEAT_COUNT}>
1951
1952Matches as C<SSS...S> (repeated as many times as necessary).
1953
1954=item C<S{min,max}>
1955
1956Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
1957
1958=item C<S{min,max}?>
1959
1960Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
1961
1962=item C<S?>, C<S*>, C<S+>
1963
1964Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
1965
1966=item C<S??>, C<S*?>, C<S+?>
1967
1968Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
1969
c47ff5f1 1970=item C<< (?>S) >>
35a734be
IZ
1971
1972Matches the best match for C<S> and only that.
1973
1974=item C<(?=S)>, C<(?<=S)>
1975
1976Only the best match for C<S> is considered. (This is important only if
1977C<S> has capturing parentheses, and backreferences are used somewhere
1978else in the whole regular expression.)
1979
1980=item C<(?!S)>, C<(?<!S)>
1981
1982For this grouping operator there is no need to describe the ordering, since
1983only whether or not C<S> can match is important.
1984
6bda09f9 1985=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
1986
1987The ordering is the same as for the regular expression which is
c27a5cfe 1988the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
1989
1990=item C<(?(condition)yes-pattern|no-pattern)>
1991
1992Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
1993already determined. The ordering of the matches is the same as for the
1994chosen subexpression.
1995
1996=back
1997
1998The above recipes describe the ordering of matches I<at a given position>.
1999One more rule is needed to understand how a match is determined for the
2000whole regular expression: a match at an earlier position is always better
2001than a match at a later position.
2002
0d017f4d 2003=head2 Creating Custom RE Engines
c84d73f1
IZ
2004
2005Overloaded constants (see L<overload>) provide a simple way to extend
2006the functionality of the RE engine.
2007
2008Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2009matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2010characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2011at these positions, so we want to have each C<\Y|> in the place of the
2012more complicated version. We can create a module C<customre> to do
2013this:
2014
2015 package customre;
2016 use overload;
2017
2018 sub import {
2019 shift;
2020 die "No argument to customre::import allowed" if @_;
2021 overload::constant 'qr' => \&convert;
2022 }
2023
2024 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2025
580a9fe1
RGS
2026 # We must also take care of not escaping the legitimate \\Y|
2027 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2028 my %rules = ( '\\' => '\\\\',
f793d64a 2029 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2030 sub convert {
2031 my $re = shift;
5d458dd8 2032 $re =~ s{
c84d73f1
IZ
2033 \\ ( \\ | Y . )
2034 }
5d458dd8 2035 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2036 return $re;
2037 }
2038
2039Now C<use customre> enables the new escape in constant regular
2040expressions, i.e., those without any runtime variable interpolations.
2041As documented in L<overload>, this conversion will work only over
2042literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2043part of this regular expression needs to be converted explicitly
2044(but only if the special meaning of C<\Y|> should be enabled inside $re):
2045
2046 use customre;
2047 $re = <>;
2048 chomp $re;
2049 $re = customre::convert $re;
2050 /\Y|$re\Y|/;
2051
1f1031fe
YO
2052=head1 PCRE/Python Support
2053
99d59c4d 2054As of Perl 5.10.0, Perl supports several Python/PCRE specific extensions
1f1031fe 2055to the regex syntax. While Perl programmers are encouraged to use the
99d59c4d 2056Perl specific syntax, the following are also accepted:
1f1031fe
YO
2057
2058=over 4
2059
ae5648b3 2060=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2061
c27a5cfe 2062Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2063
2064=item C<< (?P=NAME) >>
2065
c27a5cfe 2066Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2067
2068=item C<< (?P>NAME) >>
2069
c27a5cfe 2070Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2071
ee9b8eae 2072=back
1f1031fe 2073
19799a22
GS
2074=head1 BUGS
2075
78288b8e
KW
2076There are numerous problems with case insensitive matching of characters
2077outside the ASCII range, especially with those whose folds are multiple
2078characters, such as ligatures like C<LATIN SMALL LIGATURE FF>.
2079
f253210b
KW
2080In a bracketed character class with case insensitive matching, ranges only work
2081for ASCII characters. For example,
2082C<m/[\N{CYRILLIC CAPITAL LETTER A}-\N{CYRILLIC CAPITAL LETTER YA}]/i>
2083doesn't match all the Russian upper and lower case letters.
2084
88c9975e
KW
2085Many regular expression constructs don't work on EBCDIC platforms.
2086
9da458fc
IZ
2087This document varies from difficult to understand to completely
2088and utterly opaque. The wandering prose riddled with jargon is
2089hard to fathom in several places.
2090
2091This document needs a rewrite that separates the tutorial content
2092from the reference content.
19799a22
GS
2093
2094=head1 SEE ALSO
9fa51da4 2095
91e0c79e
MJD
2096L<perlrequick>.
2097
2098L<perlretut>.
2099
9b599b2a
GS
2100L<perlop/"Regexp Quote-Like Operators">.
2101
1e66bd83
PP
2102L<perlop/"Gory details of parsing quoted constructs">.
2103
14218588
GS
2104L<perlfaq6>.
2105
9b599b2a
GS
2106L<perlfunc/pos>.
2107
2108L<perllocale>.
2109
fb55449c
JH
2110L<perlebcdic>.
2111
14218588
GS
2112I<Mastering Regular Expressions> by Jeffrey Friedl, published
2113by O'Reilly and Associates.