This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlport: fix 80 col display; broken link
[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
54310121 127 \ Quote the next metacharacter
a0d0e21e
LW
128 ^ Match the beginning of the line
129 . Match any character (except newline)
c07a80fd 130 $ Match the end of the line (or before newline at the end)
a0d0e21e
LW
131 | Alternation
132 () Grouping
9bb1f947 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
LW
157
158 * Match 0 or more times
159 + Match 1 or more times
160 ? Match 1 or 0 times
161 {n} Match exactly n times
162 {n,} Match at least n times
163 {n,m} Match at least n but not more than m times
164
165(If a curly bracket occurs in any other context, it is treated
b975c076 166as a regular character. In particular, the lower bound
527e91da
BB
167is not optional.) The "*" quantifier is equivalent to C<{0,}>, the "+"
168quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
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
0d017f4d
WL
183 *? Match 0 or more times, not greedily
184 +? Match 1 or more times, not greedily
185 ?? Match 0 or 1 time, not greedily
186 {n}? Match exactly n times, not greedily
187 {n,}? Match at least n times, not greedily
188 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 189
b9b4dddf
YO
190By default, when a quantified subpattern does not allow the rest of the
191overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 192sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
193as well.
194
0d017f4d
WL
195 *+ Match 0 or more times and give nothing back
196 ++ Match 1 or more times and give nothing back
197 ?+ Match 0 or 1 time and give nothing back
b9b4dddf 198 {n}+ Match exactly n times and give nothing back (redundant)
04838cea
RGS
199 {n,}+ Match at least n times and give nothing back
200 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
201
202For instance,
203
204 'aaaa' =~ /a++a/
205
206will never match, as the C<a++> will gobble up all the C<a>'s in the
207string and won't leave any for the remaining part of the pattern. This
208feature can be extremely useful to give perl hints about where it
209shouldn't backtrack. For instance, the typical "match a double-quoted
210string" problem can be most efficiently performed when written as:
211
212 /"(?:[^"\\]++|\\.)*+"/
213
0d017f4d 214as we know that if the final quote does not match, backtracking will not
b9b4dddf
YO
215help. See the independent subexpression C<< (?>...) >> for more details;
216possessive quantifiers are just syntactic sugar for that construct. For
217instance the above example could also be written as follows:
218
219 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
220
04838cea
RGS
221=head3 Escape sequences
222
5f05dabc 223Because patterns are processed as double quoted strings, the following
a0d0e21e
LW
224also work:
225
0f36ee90
PP
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)
0d017f4d
WL
232 \033 octal char (example: ESC)
233 \x1B hex char (example: ESC)
196ac2fc 234 \x{263a} long hex char (example: Unicode SMILEY)
0d017f4d 235 \cK control char (example: VT)
196ac2fc 236 \N{name} named Unicode character
e526e8bb 237 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
cb1a09d0
AD
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)
5a964f20 242 \Q quote (disable) pattern metacharacters till \E
9bb1f947 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
9bb1f947
KW
252 Sequence Note Description
253 [...] [1] Match a character according to the rules of the bracketed
254 character class defined by the "...". Example: [a-z]
255 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 character
258 class. Example: [[:upper:]] matches any uppercase
259 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 part
270 of a larger UTF-8 character. Thus it breaks up characters
271 into their UTF-8 bytes, so you may end up with malformed
272 pieces of UTF-8. Unsupported in lookbehind.
273 \1 [5] Backreference to a specific capture buffer or group.
274 '1' may actually be any positive integer.
275 \g1 [5] Backreference to a specific or previous group,
276 \g{-1} [5] The number may be negative indicating a relative previous
277 buffer and may optionally be wrapped in curly brackets for
278 safer parsing.
279 \g{name} [5] Named backreference
280 \k<name> [5] Named backreference
281 \K [6] Keep the stuff left of the \K, don't include it in $&
282 \N [7] Any character but \n (experimental). Not affected by /s
283 modifier
284 \v [3] Vertical whitespace
285 \V [3] Not vertical whitespace
286 \h [3] Horizontal whitespace
287 \H [3] Not horizontal whitespace
288 \R [4] Linebreak
e1d1eefb 289
9bb1f947
KW
290=over 4
291
292=item [1]
293
294See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 295
9bb1f947 296=item [2]
b8c5462f 297
9bb1f947 298See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 299
9bb1f947 300=item [3]
5496314a 301
9bb1f947 302See L<perlrecharclass/Backslash sequences> for details.
5496314a 303
9bb1f947 304=item [4]
5496314a 305
9bb1f947 306See L<perlrebackslash/Misc> for details.
d0b16107 307
9bb1f947 308=item [5]
b8c5462f 309
9bb1f947 310See L</Capture buffers> below for details.
93733859 311
9bb1f947 312=item [6]
b8c5462f 313
9bb1f947
KW
314See L</Extended Patterns> below for details.
315
316=item [7]
317
318Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
319character whose name is C<NAME>; and similarly when of the form
320C<\N{U+I<wide hex char>}>, it matches the character whose Unicode ordinal is
321I<wide hex char>. Otherwise it matches any character but C<\n>.
322
323=back
d0b16107 324
04838cea
RGS
325=head3 Assertions
326
a0d0e21e 327Perl defines the following zero-width assertions:
d74e8afc
ITB
328X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
329X<regexp, zero-width assertion>
330X<regular expression, zero-width assertion>
331X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 332
9bb1f947
KW
333 \b Match a word boundary
334 \B Match except at a word boundary
335 \A Match only at beginning of string
336 \Z Match only at end of string, or before newline at the end
337 \z Match only at end of string
338 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 339 of prior m//g)
a0d0e21e 340
14218588 341A word boundary (C<\b>) is a spot between two characters
19799a22
GS
342that has a C<\w> on one side of it and a C<\W> on the other side
343of it (in either order), counting the imaginary characters off the
344beginning and end of the string as matching a C<\W>. (Within
345character classes C<\b> represents backspace rather than a word
346boundary, just as it normally does in any double-quoted string.)
347The C<\A> and C<\Z> are just like "^" and "$", except that they
348won't match multiple times when the C</m> modifier is used, while
349"^" and "$" will match at every internal line boundary. To match
350the actual end of the string and not ignore an optional trailing
351newline, use C<\z>.
d74e8afc 352X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
353
354The C<\G> assertion can be used to chain global matches (using
355C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
356It is also useful when writing C<lex>-like scanners, when you have
357several patterns that you want to match against consequent substrings
358of your string, see the previous reference. The actual location
359where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d
YO
360an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
361matches is modified somewhat, in that contents to the left of C<\G> is
362not counted when determining the length of the match. Thus the following
363will not match forever:
d74e8afc 364X<\G>
c47ff5f1 365
58e23c8d
YO
366 $str = 'ABC';
367 pos($str) = 1;
368 while (/.\G/g) {
369 print $&;
370 }
371
372It will print 'A' and then terminate, as it considers the match to
373be zero-width, and thus will not match at the same position twice in a
374row.
375
376It is worth noting that C<\G> improperly used can result in an infinite
377loop. Take care when using patterns that include C<\G> in an alternation.
378
04838cea
RGS
379=head3 Capture buffers
380
0d017f4d
WL
381The bracketing construct C<( ... )> creates capture buffers. To refer
382to the current contents of a buffer later on, within the same pattern,
383use \1 for the first, \2 for the second, and so on.
384Outside the match use "$" instead of "\". (The
81714fb9 385\<digit> notation works in certain circumstances outside
14218588
GS
386the match. See the warning below about \1 vs $1 for details.)
387Referring back to another part of the match is called a
388I<backreference>.
d74e8afc
ITB
389X<regex, capture buffer> X<regexp, capture buffer>
390X<regular expression, capture buffer> X<backreference>
14218588
GS
391
392There is no limit to the number of captured substrings that you may
393use. However Perl also uses \10, \11, etc. as aliases for \010,
fb55449c
JH
394\011, etc. (Recall that 0 means octal, so \011 is the character at
395number 9 in your coded character set; which would be the 10th character,
81714fb9
YO
396a horizontal tab under ASCII.) Perl resolves this
397ambiguity by interpreting \10 as a backreference only if at least 10
398left parentheses have opened before it. Likewise \11 is a
399backreference only if at least 11 left parentheses have opened
400before it. And so on. \1 through \9 are always interpreted as
5624f11d 401backreferences.
40863337
ML
402If the bracketing group did not match, the associated backreference won't
403match either. (This can happen if the bracketing group is optional, or
404in a different branch of an alternation.)
405
1f1031fe 406X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
2bf803e2 407In order to provide a safer and easier way to construct patterns using
99d59c4d
RGS
408backreferences, Perl provides the C<\g{N}> notation (starting with perl
4095.10.0). The curly brackets are optional, however omitting them is less
410safe as the meaning of the pattern can be changed by text (such as digits)
411following it. When N is a positive integer the C<\g{N}> notation is
412exactly equivalent to using normal backreferences. When N is a negative
413integer then it is a relative backreference referring to the previous N'th
414capturing group. When the bracket form is used and N is not an integer, it
415is treated as a reference to a named buffer.
2bf803e2
YO
416
417Thus C<\g{-1}> refers to the last buffer, C<\g{-2}> refers to the
418buffer before that. For example:
5624f11d
YO
419
420 /
421 (Y) # buffer 1
422 ( # buffer 2
423 (X) # buffer 3
2bf803e2
YO
424 \g{-1} # backref to buffer 3
425 \g{-3} # backref to buffer 1
5624f11d
YO
426 )
427 /x
428
2bf803e2 429and would match the same as C</(Y) ( (X) \3 \1 )/x>.
14218588 430
99d59c4d 431Additionally, as of Perl 5.10.0 you may use named capture buffers and named
1f1031fe 432backreferences. The notation is C<< (?<name>...) >> to declare and C<< \k<name> >>
0d017f4d
WL
433to reference. You may also use apostrophes instead of angle brackets to delimit the
434name; and you may use the bracketed C<< \g{name} >> backreference syntax.
435It's possible to refer to a named capture buffer by absolute and relative number as well.
436Outside the pattern, a named capture buffer is available via the C<%+> hash.
437When different buffers within the same pattern have the same name, C<$+{name}>
438and C<< \k<name> >> refer to the leftmost defined group. (Thus it's possible
439to do things with named capture buffers that would otherwise require C<(??{})>
440code to accomplish.)
441X<named capture buffer> X<regular expression, named capture buffer>
64c5a566 442X<%+> X<$+{name}> X<< \k<name> >>
81714fb9 443
14218588 444Examples:
a0d0e21e
LW
445
446 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
447
81714fb9
YO
448 /(.)\1/ # find first doubled char
449 and print "'$1' is the first doubled character\n";
450
451 /(?<char>.)\k<char>/ # ... a different way
452 and print "'$+{char}' is the first doubled character\n";
453
0d017f4d 454 /(?'char'.)\1/ # ... mix and match
81714fb9 455 and print "'$1' is the first doubled character\n";
c47ff5f1 456
14218588 457 if (/Time: (..):(..):(..)/) { # parse out values
a0d0e21e
LW
458 $hours = $1;
459 $minutes = $2;
460 $seconds = $3;
461 }
c47ff5f1 462
14218588
GS
463Several special variables also refer back to portions of the previous
464match. C<$+> returns whatever the last bracket match matched.
465C<$&> returns the entire matched string. (At one point C<$0> did
466also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
467everything before the matched string. C<$'> returns everything
468after the matched string. And C<$^N> contains whatever was matched by
469the most-recently closed group (submatch). C<$^N> can be used in
470extended patterns (see below), for example to assign a submatch to a
81714fb9 471variable.
d74e8afc 472X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 473
665e98b9 474The numbered match variables ($1, $2, $3, etc.) and the related punctuation
77ea4f6d 475set (C<$+>, C<$&>, C<$`>, C<$'>, and C<$^N>) are all dynamically scoped
14218588
GS
476until the end of the enclosing block or until the next successful
477match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
478X<$+> X<$^N> X<$&> X<$`> X<$'>
479X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
480
14218588 481
0d017f4d 482B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 483which makes it easier to write code that tests for a series of more
665e98b9
JH
484specific cases and remembers the best match.
485
14218588
GS
486B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
487C<$'> anywhere in the program, it has to provide them for every
488pattern match. This may substantially slow your program. Perl
489uses the same mechanism to produce $1, $2, etc, so you also pay a
490price for each pattern that contains capturing parentheses. (To
491avoid this cost while retaining the grouping behaviour, use the
492extended regular expression C<(?: ... )> instead.) But if you never
493use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
494parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
495if you can, but if you can't (and some algorithms really appreciate
496them), once you've used them once, use them at will, because you've
497already paid the price. As of 5.005, C<$&> is not so costly as the
498other two.
d74e8afc 499X<$&> X<$`> X<$'>
68dc0745 500
99d59c4d 501As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
502C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
503and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 504successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
505The use of these variables incurs no global performance penalty, unlike
506their punctuation char equivalents, however at the trade-off that you
507have to tell perl when you want to use them.
87e95b7f 508X</p> X<p modifier>
cde0cee5 509
19799a22
GS
510Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
511C<\w>, C<\n>. Unlike some other regular expression languages, there
512are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 513that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
514interpreted as a literal character, not a metacharacter. This was
515once used in a common idiom to disable or quote the special meanings
516of regular expression metacharacters in a string that you want to
36bbe248 517use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
518
519 $pattern =~ s/(\W)/\\$1/g;
520
f1cbbd6e 521(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
522Today it is more common to use the quotemeta() function or the C<\Q>
523metaquoting escape sequence to disable all metacharacters' special
524meanings like this:
a0d0e21e
LW
525
526 /$unquoted\Q$quoted\E$unquoted/
527
9da458fc
IZ
528Beware that if you put literal backslashes (those not inside
529interpolated variables) between C<\Q> and C<\E>, double-quotish
530backslash interpolation may lead to confusing results. If you
531I<need> to use literal backslashes within C<\Q...\E>,
532consult L<perlop/"Gory details of parsing quoted constructs">.
533
19799a22
GS
534=head2 Extended Patterns
535
14218588
GS
536Perl also defines a consistent extension syntax for features not
537found in standard tools like B<awk> and B<lex>. The syntax is a
538pair of parentheses with a question mark as the first thing within
539the parentheses. The character after the question mark indicates
540the extension.
19799a22 541
14218588
GS
542The stability of these extensions varies widely. Some have been
543part of the core language for many years. Others are experimental
544and may change without warning or be completely removed. Check
545the documentation on an individual feature to verify its current
546status.
19799a22 547
14218588
GS
548A question mark was chosen for this and for the minimal-matching
549construct because 1) question marks are rare in older regular
550expressions, and 2) whenever you see one, you should stop and
551"question" exactly what is going on. That's psychology...
a0d0e21e
LW
552
553=over 10
554
cc6b7395 555=item C<(?#text)>
d74e8afc 556X<(?#)>
a0d0e21e 557
14218588 558A comment. The text is ignored. If the C</x> modifier enables
19799a22 559whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
560the comment as soon as it sees a C<)>, so there is no way to put a literal
561C<)> in the comment.
a0d0e21e 562
f7819f85 563=item C<(?pimsx-imsx)>
d74e8afc 564X<(?)>
19799a22 565
0b6d1084
JH
566One or more embedded pattern-match modifiers, to be turned on (or
567turned off, if preceded by C<->) for the remainder of the pattern or
568the remainder of the enclosing pattern group (if any). This is
569particularly useful for dynamic patterns, such as those read in from a
0d017f4d
WL
570configuration file, taken from an argument, or specified in a table
571somewhere. Consider the case where some patterns want to be case
572sensitive and some do not: The case insensitive ones merely need to
573include C<(?i)> at the front of the pattern. For example:
19799a22
GS
574
575 $pattern = "foobar";
5d458dd8 576 if ( /$pattern/i ) { }
19799a22
GS
577
578 # more flexible:
579
580 $pattern = "(?i)foobar";
5d458dd8 581 if ( /$pattern/ ) { }
19799a22 582
0b6d1084 583These modifiers are restored at the end of the enclosing group. For example,
19799a22
GS
584
585 ( (?i) blah ) \s+ \1
586
0d017f4d
WL
587will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
588repetition of the previous word, assuming the C</x> modifier, and no C</i>
589modifier outside this group.
19799a22 590
8eb5594e
DR
591These modifiers do not carry over into named subpatterns called in the
592enclosing group. In other words, a pattern such as C<((?i)(&NAME))> does not
593change the case-sensitivity of the "NAME" pattern.
594
5530442b 595Note that the C<p> modifier is special in that it can only be enabled,
cde0cee5 596not disabled, and that its presence anywhere in a pattern has a global
5530442b 597effect. Thus C<(?-p)> and C<(?-p:...)> are meaningless and will warn
cde0cee5
YO
598when executed under C<use warnings>.
599
5a964f20 600=item C<(?:pattern)>
d74e8afc 601X<(?:)>
a0d0e21e 602
ca9dfc88
IZ
603=item C<(?imsx-imsx:pattern)>
604
5a964f20
TC
605This is for clustering, not capturing; it groups subexpressions like
606"()", but doesn't make backreferences as "()" does. So
a0d0e21e 607
5a964f20 608 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
609
610is like
611
5a964f20 612 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 613
19799a22
GS
614but doesn't spit out extra fields. It's also cheaper not to capture
615characters if you don't need to.
a0d0e21e 616
19799a22 617Any letters between C<?> and C<:> act as flags modifiers as with
5d458dd8 618C<(?imsx-imsx)>. For example,
ca9dfc88
IZ
619
620 /(?s-i:more.*than).*million/i
621
14218588 622is equivalent to the more verbose
ca9dfc88
IZ
623
624 /(?:(?s-i)more.*than).*million/i
625
594d7033
YO
626=item C<(?|pattern)>
627X<(?|)> X<Branch reset>
628
629This is the "branch reset" pattern, which has the special property
630that the capture buffers are numbered from the same starting point
99d59c4d 631in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 632
693596a8
RGS
633Capture buffers are numbered from left to right, but inside this
634construct the numbering is restarted for each branch.
4deaaa80
PJ
635
636The numbering within each branch will be as normal, and any buffers
637following this construct will be numbered as though the construct
638contained only one branch, that being the one with the most capture
639buffers in it.
640
641This construct will be useful when you want to capture one of a
642number of alternative matches.
643
644Consider the following pattern. The numbers underneath show in
645which buffer the captured content will be stored.
594d7033
YO
646
647
648 # before ---------------branch-reset----------- after
649 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
650 # 1 2 2 3 2 3 4
651
ab106183
A
652Be careful when using the branch reset pattern in combination with
653named captures. Named captures are implemented as being aliases to
654numbered buffers holding the captures, and that interferes with the
655implementation of the branch reset pattern. If you are using named
656captures in a branch reset pattern, it's best to use the same names,
657in the same order, in each of the alternations:
658
659 /(?| (?<a> x ) (?<b> y )
660 | (?<a> z ) (?<b> w )) /x
661
662Not doing so may lead to surprises:
663
664 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
665 say $+ {a}; # Prints '12'
666 say $+ {b}; # *Also* prints '12'.
667
668The problem here is that both the buffer named C<< a >> and the buffer
669named C<< b >> are aliases for the buffer belonging to C<< $1 >>.
90a18110 670
ee9b8eae
YO
671=item Look-Around Assertions
672X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
673
674Look-around assertions are zero width patterns which match a specific
675pattern without including it in C<$&>. Positive assertions match when
676their subpattern matches, negative assertions match when their subpattern
677fails. Look-behind matches text up to the current match position,
678look-ahead matches text following the current match position.
679
680=over 4
681
5a964f20 682=item C<(?=pattern)>
d74e8afc 683X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 684
19799a22 685A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
686matches a word followed by a tab, without including the tab in C<$&>.
687
5a964f20 688=item C<(?!pattern)>
d74e8afc 689X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 690
19799a22 691A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 692matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
693however that look-ahead and look-behind are NOT the same thing. You cannot
694use this for look-behind.
7b8d334a 695
5a964f20 696If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
697will not do what you want. That's because the C<(?!foo)> is just saying that
698the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
699match. You would have to do something like C</(?!foo)...bar/> for that. We
700say "like" because there's the case of your "bar" not having three characters
701before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
702Sometimes it's still easier just to say:
a0d0e21e 703
a3cb178b 704 if (/bar/ && $` !~ /foo$/)
a0d0e21e 705
19799a22 706For look-behind see below.
c277df42 707
ee9b8eae
YO
708=item C<(?<=pattern)> C<\K>
709X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 710
c47ff5f1 711A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
712matches a word that follows a tab, without including the tab in C<$&>.
713Works only for fixed-width look-behind.
c277df42 714
ee9b8eae
YO
715There is a special form of this construct, called C<\K>, which causes the
716regex engine to "keep" everything it had matched prior to the C<\K> and
717not include it in C<$&>. This effectively provides variable length
718look-behind. The use of C<\K> inside of another look-around assertion
719is allowed, but the behaviour is currently not well defined.
720
c62285ac 721For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
722equivalent C<< (?<=...) >> construct, and it is especially useful in
723situations where you want to efficiently remove something following
724something else in a string. For instance
725
726 s/(foo)bar/$1/g;
727
728can be rewritten as the much more efficient
729
730 s/foo\Kbar//g;
731
5a964f20 732=item C<(?<!pattern)>
d74e8afc 733X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 734
19799a22
GS
735A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
736matches any occurrence of "foo" that does not follow "bar". Works
737only for fixed-width look-behind.
c277df42 738
ee9b8eae
YO
739=back
740
81714fb9
YO
741=item C<(?'NAME'pattern)>
742
743=item C<< (?<NAME>pattern) >>
744X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
745
746A named capture buffer. Identical in every respect to normal capturing
90a18110
RGS
747parentheses C<()> but for the additional fact that C<%+> or C<%-> may be
748used after a successful match to refer to a named buffer. See C<perlvar>
749for more details on the C<%+> and C<%-> hashes.
81714fb9
YO
750
751If multiple distinct capture buffers have the same name then the
752$+{NAME} will refer to the leftmost defined buffer in the match.
753
0d017f4d 754The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
755
756B<NOTE:> While the notation of this construct is the same as the similar
0d017f4d 757function in .NET regexes, the behavior is not. In Perl the buffers are
81714fb9
YO
758numbered sequentially regardless of being named or not. Thus in the
759pattern
760
761 /(x)(?<foo>y)(z)/
762
763$+{foo} will be the same as $2, and $3 will contain 'z' instead of
764the opposite which is what a .NET regex hacker might expect.
765
1f1031fe
YO
766Currently NAME is restricted to simple identifiers only.
767In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
768its Unicode extension (see L<utf8>),
769though it isn't extended by the locale (see L<perllocale>).
81714fb9 770
1f1031fe 771B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 772with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 773may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 774support the use of single quotes as a delimiter for the name.
81714fb9 775
1f1031fe
YO
776=item C<< \k<NAME> >>
777
778=item C<< \k'NAME' >>
81714fb9
YO
779
780Named backreference. Similar to numeric backreferences, except that
781the group is designated by name and not number. If multiple groups
782have the same name then it refers to the leftmost defined group in
783the current match.
784
0d017f4d 785It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
786earlier in the pattern.
787
788Both forms are equivalent.
789
1f1031fe 790B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 791with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 792may be used instead of C<< \k<NAME> >>.
1f1031fe 793
cc6b7395 794=item C<(?{ code })>
d74e8afc 795X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 796
19799a22 797B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
798experimental, and may be changed without notice. Code executed that
799has side effects may not perform identically from version to version
800due to the effect of future optimisations in the regex engine.
c277df42 801
cc46d5f2 802This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
803always succeeds, and its C<code> is not interpolated. Currently,
804the rules to determine where the C<code> ends are somewhat convoluted.
805
77ea4f6d
JV
806This feature can be used together with the special variable C<$^N> to
807capture the results of submatches in variables without having to keep
808track of the number of nested parentheses. For example:
809
810 $_ = "The brown fox jumps over the lazy dog";
811 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
812 print "color = $color, animal = $animal\n";
813
754091cb
RGS
814Inside the C<(?{...})> block, C<$_> refers to the string the regular
815expression is matching against. You can also use C<pos()> to know what is
fa11829f 816the current position of matching within this string.
754091cb 817
19799a22
GS
818The C<code> is properly scoped in the following sense: If the assertion
819is backtracked (compare L<"Backtracking">), all changes introduced after
820C<local>ization are undone, so that
b9ac3b5b
GS
821
822 $_ = 'a' x 8;
5d458dd8 823 m<
b9ac3b5b
GS
824 (?{ $cnt = 0 }) # Initialize $cnt.
825 (
5d458dd8 826 a
b9ac3b5b
GS
827 (?{
828 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
829 })
5d458dd8 830 )*
b9ac3b5b
GS
831 aaaa
832 (?{ $res = $cnt }) # On success copy to non-localized
833 # location.
834 >x;
835
0d017f4d 836will set C<$res = 4>. Note that after the match, C<$cnt> returns to the globally
14218588 837introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
838are unwound.
839
19799a22
GS
840This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
841switch. If I<not> used in this way, the result of evaluation of
842C<code> is put into the special variable C<$^R>. This happens
843immediately, so C<$^R> can be used from other C<(?{ code })> assertions
844inside the same regular expression.
b9ac3b5b 845
19799a22
GS
846The assignment to C<$^R> above is properly localized, so the old
847value of C<$^R> is restored if the assertion is backtracked; compare
848L<"Backtracking">.
b9ac3b5b 849
19799a22
GS
850For reasons of security, this construct is forbidden if the regular
851expression involves run-time interpolation of variables, unless the
852perilous C<use re 'eval'> pragma has been used (see L<re>), or the
853variables contain results of C<qr//> operator (see
9bb1f947 854L<perlop/"qr/STRINGE<sol>msixpo">).
871b0233 855
0d017f4d 856This restriction is due to the wide-spread and remarkably convenient
19799a22 857custom of using run-time determined strings as patterns. For example:
871b0233
IZ
858
859 $re = <>;
860 chomp $re;
861 $string =~ /$re/;
862
14218588
GS
863Before Perl knew how to execute interpolated code within a pattern,
864this operation was completely safe from a security point of view,
865although it could raise an exception from an illegal pattern. If
866you turn on the C<use re 'eval'>, though, it is no longer secure,
867so you should only do so if you are also using taint checking.
868Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 869compartment. See L<perlsec> for details about both these mechanisms.
871b0233 870
e95d7314
GG
871B<WARNING>: Use of lexical (C<my>) variables in these blocks is
872broken. The result is unpredictable and will make perl unstable. The
873workaround is to use global (C<our>) variables.
874
875B<WARNING>: Because Perl's regex engine is currently not re-entrant,
876interpolated code may not invoke the regex engine either directly with
877C<m//> or C<s///>), or indirectly with functions such as
878C<split>. Invoking the regex engine in these blocks will make perl
879unstable.
8988a1bb 880
14455d6c 881=item C<(??{ code })>
d74e8afc
ITB
882X<(??{})>
883X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 884
19799a22 885B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
886experimental, and may be changed without notice. Code executed that
887has side effects may not perform identically from version to version
888due to the effect of future optimisations in the regex engine.
0f5d15d6 889
19799a22
GS
890This is a "postponed" regular subexpression. The C<code> is evaluated
891at run time, at the moment this subexpression may match. The result
892of evaluation is considered as a regular expression and matched as
61528107 893if it were inserted instead of this construct. Note that this means
6bda09f9
YO
894that the contents of capture buffers defined inside an eval'ed pattern
895are not available outside of the pattern, and vice versa, there is no
896way for the inner pattern to refer to a capture buffer defined outside.
897Thus,
898
899 ('a' x 100)=~/(??{'(.)' x 100})/
900
81714fb9 901B<will> match, it will B<not> set $1.
0f5d15d6 902
428594d9 903The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
904where the C<code> ends are currently somewhat convoluted.
905
906The following pattern matches a parenthesized group:
0f5d15d6
IZ
907
908 $re = qr{
909 \(
910 (?:
911 (?> [^()]+ ) # Non-parens without backtracking
912 |
14455d6c 913 (??{ $re }) # Group with matching parens
0f5d15d6
IZ
914 )*
915 \)
916 }x;
917
6bda09f9
YO
918See also C<(?PARNO)> for a different, more efficient way to accomplish
919the same task.
920
0b370c0a
A
921For reasons of security, this construct is forbidden if the regular
922expression involves run-time interpolation of variables, unless the
923perilous C<use re 'eval'> pragma has been used (see L<re>), or the
924variables contain results of C<qr//> operator (see
9bb1f947 925L<perlop/"qrE<sol>STRINGE<sol>msixpo">).
0b370c0a 926
5d458dd8 927Because perl's regex engine is not currently re-entrant, delayed
8988a1bb
DD
928code may not invoke the regex engine either directly with C<m//> or C<s///>),
929or indirectly with functions such as C<split>.
930
5d458dd8
YO
931Recursing deeper than 50 times without consuming any input string will
932result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
933changing it requires a custom build.
934
542fa716
YO
935=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
936X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 937X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 938X<regex, relative recursion>
6bda09f9 939
81714fb9
YO
940Similar to C<(??{ code })> except it does not involve compiling any code,
941instead it treats the contents of a capture buffer as an independent
61528107 942pattern that must match at the current position. Capture buffers
81714fb9 943contained by the pattern will have the value as determined by the
6bda09f9
YO
944outermost recursion.
945
894be9b7
YO
946PARNO is a sequence of digits (not starting with 0) whose value reflects
947the paren-number of the capture buffer to recurse to. C<(?R)> recurses to
948the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716
YO
949C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
950to be relative, with negative numbers indicating preceding capture buffers
951and positive ones following. Thus C<(?-1)> refers to the most recently
952declared buffer, and C<(?+1)> indicates the next buffer to be declared.
c74340f9
YO
953Note that the counting for relative recursion differs from that of
954relative backreferences, in that with recursion unclosed buffers B<are>
955included.
6bda09f9 956
81714fb9 957The following pattern matches a function foo() which may contain
f145b7e9 958balanced parentheses as the argument.
6bda09f9
YO
959
960 $re = qr{ ( # paren group 1 (full function)
81714fb9 961 foo
6bda09f9
YO
962 ( # paren group 2 (parens)
963 \(
964 ( # paren group 3 (contents of parens)
965 (?:
966 (?> [^()]+ ) # Non-parens without backtracking
967 |
968 (?2) # Recurse to start of paren group 2
969 )*
970 )
971 \)
972 )
973 )
974 }x;
975
976If the pattern was used as follows
977
978 'foo(bar(baz)+baz(bop))'=~/$re/
979 and print "\$1 = $1\n",
980 "\$2 = $2\n",
981 "\$3 = $3\n";
982
983the output produced should be the following:
984
985 $1 = foo(bar(baz)+baz(bop))
986 $2 = (bar(baz)+baz(bop))
81714fb9 987 $3 = bar(baz)+baz(bop)
6bda09f9 988
81714fb9 989If there is no corresponding capture buffer defined, then it is a
61528107 990fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 991string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
992into perl, so changing it requires a custom build.
993
542fa716
YO
994The following shows how using negative indexing can make it
995easier to embed recursive patterns inside of a C<qr//> construct
996for later use:
997
998 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
999 if (/foo $parens \s+ + \s+ bar $parens/x) {
1000 # do something here...
1001 }
1002
81714fb9 1003B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1004PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1005a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1006as atomic. Also, modifiers are resolved at compile time, so constructs
1007like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1008be processed.
6bda09f9 1009
894be9b7
YO
1010=item C<(?&NAME)>
1011X<(?&NAME)>
1012
0d017f4d
WL
1013Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1014parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1015the same name, then it recurses to the leftmost.
1016
1017It is an error to refer to a name that is not declared somewhere in the
1018pattern.
1019
1f1031fe
YO
1020B<NOTE:> In order to make things easier for programmers with experience
1021with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1022may be used instead of C<< (?&NAME) >>.
1f1031fe 1023
e2e6a0f1
YO
1024=item C<(?(condition)yes-pattern|no-pattern)>
1025X<(?()>
286f584a 1026
e2e6a0f1 1027=item C<(?(condition)yes-pattern)>
286f584a 1028
e2e6a0f1
YO
1029Conditional expression. C<(condition)> should be either an integer in
1030parentheses (which is valid if the corresponding pair of parentheses
1031matched), a look-ahead/look-behind/evaluate zero-width assertion, a
1032name in angle brackets or single quotes (which is valid if a buffer
1033with the given name matched), or the special symbol (R) (true when
1034evaluated inside of recursion or eval). Additionally the R may be
1035followed by a number, (which will be true when evaluated when recursing
1036inside of the appropriate group), or by C<&NAME>, in which case it will
1037be true only when evaluated during recursion in the named group.
1038
1039Here's a summary of the possible predicates:
1040
1041=over 4
1042
1043=item (1) (2) ...
1044
1045Checks if the numbered capturing buffer has matched something.
1046
1047=item (<NAME>) ('NAME')
1048
1049Checks if a buffer with the given name has matched something.
1050
1051=item (?{ CODE })
1052
1053Treats the code block as the condition.
1054
1055=item (R)
1056
1057Checks if the expression has been evaluated inside of recursion.
1058
1059=item (R1) (R2) ...
1060
1061Checks if the expression has been evaluated while executing directly
1062inside of the n-th capture group. This check is the regex equivalent of
1063
1064 if ((caller(0))[3] eq 'subname') { ... }
1065
1066In other words, it does not check the full recursion stack.
1067
1068=item (R&NAME)
1069
1070Similar to C<(R1)>, this predicate checks to see if we're executing
1071directly inside of the leftmost group with a given name (this is the same
1072logic used by C<(?&NAME)> to disambiguate). It does not check the full
1073stack, but only the name of the innermost active recursion.
1074
1075=item (DEFINE)
1076
1077In this case, the yes-pattern is never directly executed, and no
1078no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1079See below for details.
1080
1081=back
1082
1083For example:
1084
1085 m{ ( \( )?
1086 [^()]+
1087 (?(1) \) )
1088 }x
1089
1090matches a chunk of non-parentheses, possibly included in parentheses
1091themselves.
1092
1093A special form is the C<(DEFINE)> predicate, which never executes directly
1094its yes-pattern, and does not allow a no-pattern. This allows to define
1095subpatterns which will be executed only by using the recursion mechanism.
1096This way, you can define a set of regular expression rules that can be
1097bundled into any pattern you choose.
1098
1099It is recommended that for this usage you put the DEFINE block at the
1100end of the pattern, and that you name any subpatterns defined within it.
1101
1102Also, it's worth noting that patterns defined this way probably will
1103not be as efficient, as the optimiser is not very clever about
1104handling them.
1105
1106An example of how this might be used is as follows:
1107
2bf803e2 1108 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1109 (?(DEFINE)
2bf803e2
YO
1110 (?<NAME_PAT>....)
1111 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1112 )/x
1113
1114Note that capture buffers matched inside of recursion are not accessible
0d017f4d 1115after the recursion returns, so the extra layer of capturing buffers is
e2e6a0f1
YO
1116necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1117C<$+{NAME}> would be.
286f584a 1118
c47ff5f1 1119=item C<< (?>pattern) >>
6bda09f9 1120X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1121
19799a22
GS
1122An "independent" subexpression, one which matches the substring
1123that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1124position, and it matches I<nothing other than this substring>. This
19799a22
GS
1125construct is useful for optimizations of what would otherwise be
1126"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1127It may also be useful in places where the "grab all you can, and do not
1128give anything back" semantic is desirable.
19799a22 1129
c47ff5f1 1130For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1131(anchored at the beginning of string, as above) will match I<all>
1132characters C<a> at the beginning of string, leaving no C<a> for
1133C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1134since the match of the subgroup C<a*> is influenced by the following
1135group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1136C<a*ab> will match fewer characters than a standalone C<a*>, since
1137this makes the tail match.
1138
c47ff5f1 1139An effect similar to C<< (?>pattern) >> may be achieved by writing
19799a22
GS
1140C<(?=(pattern))\1>. This matches the same substring as a standalone
1141C<a+>, and the following C<\1> eats the matched string; it therefore
c47ff5f1 1142makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1143(The difference between these two constructs is that the second one
1144uses a capturing group, thus shifting ordinals of backreferences
1145in the rest of a regular expression.)
1146
1147Consider this pattern:
c277df42 1148
871b0233 1149 m{ \(
e2e6a0f1
YO
1150 (
1151 [^()]+ # x+
1152 |
871b0233
IZ
1153 \( [^()]* \)
1154 )+
e2e6a0f1 1155 \)
871b0233 1156 }x
5a964f20 1157
19799a22
GS
1158That will efficiently match a nonempty group with matching parentheses
1159two levels deep or less. However, if there is no such group, it
1160will take virtually forever on a long string. That's because there
1161are so many different ways to split a long string into several
1162substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1163to a subpattern of the above pattern. Consider how the pattern
1164above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1165seconds, but that each extra letter doubles this time. This
1166exponential performance will make it appear that your program has
14218588 1167hung. However, a tiny change to this pattern
5a964f20 1168
e2e6a0f1
YO
1169 m{ \(
1170 (
1171 (?> [^()]+ ) # change x+ above to (?> x+ )
1172 |
871b0233
IZ
1173 \( [^()]* \)
1174 )+
e2e6a0f1 1175 \)
871b0233 1176 }x
c277df42 1177
c47ff5f1 1178which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1179this yourself would be a productive exercise), but finishes in a fourth
1180the time when used on a similar string with 1000000 C<a>s. Be aware,
1181however, that this pattern currently triggers a warning message under
9f1b1f2d 1182the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1183C<"matches null string many times in regex">.
c277df42 1184
c47ff5f1 1185On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1186effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1187This was only 4 times slower on a string with 1000000 C<a>s.
1188
9da458fc
IZ
1189The "grab all you can, and do not give anything back" semantic is desirable
1190in many situations where on the first sight a simple C<()*> looks like
1191the correct solution. Suppose we parse text with comments being delimited
1192by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1193its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1194the comment delimiter, because it may "give up" some whitespace if
1195the remainder of the pattern can be made to match that way. The correct
1196answer is either one of these:
1197
1198 (?>#[ \t]*)
1199 #[ \t]*(?![ \t])
1200
1201For example, to grab non-empty comments into $1, one should use either
1202one of these:
1203
1204 / (?> \# [ \t]* ) ( .+ ) /x;
1205 / \# [ \t]* ( [^ \t] .* ) /x;
1206
1207Which one you pick depends on which of these expressions better reflects
1208the above specification of comments.
1209
6bda09f9
YO
1210In some literature this construct is called "atomic matching" or
1211"possessive matching".
1212
b9b4dddf
YO
1213Possessive quantifiers are equivalent to putting the item they are applied
1214to inside of one of these constructs. The following equivalences apply:
1215
1216 Quantifier Form Bracketing Form
1217 --------------- ---------------
1218 PAT*+ (?>PAT*)
1219 PAT++ (?>PAT+)
1220 PAT?+ (?>PAT?)
1221 PAT{min,max}+ (?>PAT{min,max})
1222
e2e6a0f1
YO
1223=back
1224
1225=head2 Special Backtracking Control Verbs
1226
1227B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1228removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1229be noted to avoid problems during upgrades.
1230
1231These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1232otherwise stated the ARG argument is optional; in some cases, it is
1233forbidden.
1234
1235Any pattern containing a special backtracking verb that allows an argument
e1020413 1236has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1237C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1238rules apply:
e2e6a0f1 1239
5d458dd8
YO
1240On failure, the C<$REGERROR> variable will be set to the ARG value of the
1241verb pattern, if the verb was involved in the failure of the match. If the
1242ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1243name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1244none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1245
5d458dd8
YO
1246On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1247the C<$REGMARK> variable will be set to the name of the last
1248C<(*MARK:NAME)> pattern executed. See the explanation for the
1249C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1250
5d458dd8
YO
1251B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1252and most other regex related variables. They are not local to a scope, nor
1253readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1254Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1255
1256If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1257argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1
YO
1258
1259=over 4
1260
1261=item Verbs that take an argument
1262
1263=over 4
1264
5d458dd8 1265=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1266X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1267
5d458dd8
YO
1268This zero-width pattern prunes the backtracking tree at the current point
1269when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1270where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1271A may backtrack as necessary to match. Once it is reached, matching
1272continues in B, which may also backtrack as necessary; however, should B
1273not match, then no further backtracking will take place, and the pattern
1274will fail outright at the current starting position.
54612592
YO
1275
1276The following example counts all the possible matching strings in a
1277pattern (without actually matching any of them).
1278
e2e6a0f1 1279 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1280 print "Count=$count\n";
1281
1282which produces:
1283
1284 aaab
1285 aaa
1286 aa
1287 a
1288 aab
1289 aa
1290 a
1291 ab
1292 a
1293 Count=9
1294
5d458dd8 1295If we add a C<(*PRUNE)> before the count like the following
54612592 1296
5d458dd8 1297 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1298 print "Count=$count\n";
1299
1300we prevent backtracking and find the count of the longest matching
353c6505 1301at each matching starting point like so:
54612592
YO
1302
1303 aaab
1304 aab
1305 ab
1306 Count=3
1307
5d458dd8 1308Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1309
5d458dd8
YO
1310See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1311control backtracking. In some cases, the use of C<(*PRUNE)> can be
1312replaced with a C<< (?>pattern) >> with no functional difference; however,
1313C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1314C<< (?>pattern) >> alone.
54612592 1315
e2e6a0f1 1316
5d458dd8
YO
1317=item C<(*SKIP)> C<(*SKIP:NAME)>
1318X<(*SKIP)>
e2e6a0f1 1319
5d458dd8 1320This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1321failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1322to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1323of this pattern. This effectively means that the regex engine "skips" forward
1324to this position on failure and tries to match again, (assuming that
1325there is sufficient room to match).
1326
1327The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1328C<(*MARK:NAME)> was encountered while matching, then it is that position
1329which is used as the "skip point". If no C<(*MARK)> of that name was
1330encountered, then the C<(*SKIP)> operator has no effect. When used
1331without a name the "skip point" is where the match point was when
1332executing the (*SKIP) pattern.
1333
1334Compare the following to the examples in C<(*PRUNE)>, note the string
24b23f37
YO
1335is twice as long:
1336
5d458dd8 1337 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1338 print "Count=$count\n";
1339
1340outputs
1341
1342 aaab
1343 aaab
1344 Count=2
1345
5d458dd8 1346Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1347executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1348C<(*SKIP)> was executed.
1349
5d458dd8
YO
1350=item C<(*MARK:NAME)> C<(*:NAME)>
1351X<(*MARK)> C<(*MARK:NAME)> C<(*:NAME)>
1352
1353This zero-width pattern can be used to mark the point reached in a string
1354when a certain part of the pattern has been successfully matched. This
1355mark may be given a name. A later C<(*SKIP)> pattern will then skip
1356forward to that point if backtracked into on failure. Any number of
b4222fa9 1357C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1358
1359In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1360can be used to "label" a pattern branch, so that after matching, the
1361program can determine which branches of the pattern were involved in the
1362match.
1363
1364When a match is successful, the C<$REGMARK> variable will be set to the
1365name of the most recently executed C<(*MARK:NAME)> that was involved
1366in the match.
1367
1368This can be used to determine which branch of a pattern was matched
c62285ac 1369without using a separate capture buffer for each branch, which in turn
5d458dd8
YO
1370can result in a performance improvement, as perl cannot optimize
1371C</(?:(x)|(y)|(z))/> as efficiently as something like
1372C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1373
1374When a match has failed, and unless another verb has been involved in
1375failing the match and has provided its own name to use, the C<$REGERROR>
1376variable will be set to the name of the most recently executed
1377C<(*MARK:NAME)>.
1378
1379See C<(*SKIP)> for more details.
1380
b62d2d15
YO
1381As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1382
5d458dd8
YO
1383=item C<(*THEN)> C<(*THEN:NAME)>
1384
241e7389 1385This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1386C<(*PRUNE)>, this verb always matches, and when backtracked into on
1387failure, it causes the regex engine to try the next alternation in the
1388innermost enclosing group (capturing or otherwise).
1389
1390Its name comes from the observation that this operation combined with the
1391alternation operator (C<|>) can be used to create what is essentially a
1392pattern-based if/then/else block:
1393
1394 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1395
1396Note that if this operator is used and NOT inside of an alternation then
1397it acts exactly like the C<(*PRUNE)> operator.
1398
1399 / A (*PRUNE) B /
1400
1401is the same as
1402
1403 / A (*THEN) B /
1404
1405but
1406
1407 / ( A (*THEN) B | C (*THEN) D ) /
1408
1409is not the same as
1410
1411 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1412
1413as after matching the A but failing on the B the C<(*THEN)> verb will
1414backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1415
e2e6a0f1
YO
1416=item C<(*COMMIT)>
1417X<(*COMMIT)>
24b23f37 1418
241e7389 1419This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1420zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1421into on failure it causes the match to fail outright. No further attempts
1422to find a valid match by advancing the start pointer will occur again.
1423For example,
24b23f37 1424
e2e6a0f1 1425 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1426 print "Count=$count\n";
1427
1428outputs
1429
1430 aaab
1431 Count=1
1432
e2e6a0f1
YO
1433In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1434does not match, the regex engine will not try any further matching on the
1435rest of the string.
c277df42 1436
e2e6a0f1 1437=back
9af228c6 1438
e2e6a0f1 1439=item Verbs without an argument
9af228c6
YO
1440
1441=over 4
1442
e2e6a0f1
YO
1443=item C<(*FAIL)> C<(*F)>
1444X<(*FAIL)> X<(*F)>
9af228c6 1445
e2e6a0f1
YO
1446This pattern matches nothing and always fails. It can be used to force the
1447engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1448fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1449
e2e6a0f1 1450It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1451
e2e6a0f1
YO
1452=item C<(*ACCEPT)>
1453X<(*ACCEPT)>
9af228c6 1454
e2e6a0f1
YO
1455B<WARNING:> This feature is highly experimental. It is not recommended
1456for production code.
9af228c6 1457
e2e6a0f1
YO
1458This pattern matches nothing and causes the end of successful matching at
1459the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1460whether there is actually more to match in the string. When inside of a
0d017f4d 1461nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1462via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1463
e2e6a0f1
YO
1464If the C<(*ACCEPT)> is inside of capturing buffers then the buffers are
1465marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1466For instance:
9af228c6 1467
e2e6a0f1 1468 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1469
e2e6a0f1 1470will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0d017f4d 1471be set. If another branch in the inner parentheses were matched, such as in the
e2e6a0f1 1472string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1473
1474=back
c277df42 1475
a0d0e21e
LW
1476=back
1477
c07a80fd 1478=head2 Backtracking
d74e8afc 1479X<backtrack> X<backtracking>
c07a80fd 1480
35a734be
IZ
1481NOTE: This section presents an abstract approximation of regular
1482expression behavior. For a more rigorous (and complicated) view of
1483the rules involved in selecting a match among possible alternatives,
0d017f4d 1484see L<Combining RE Pieces>.
35a734be 1485
c277df42 1486A fundamental feature of regular expression matching involves the
5a964f20 1487notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1488by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1489C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1490internally, but the general principle outlined here is valid.
c07a80fd
PP
1491
1492For a regular expression to match, the I<entire> regular expression must
1493match, not just part of it. So if the beginning of a pattern containing a
1494quantifier succeeds in a way that causes later parts in the pattern to
1495fail, the matching engine backs up and recalculates the beginning
1496part--that's why it's called backtracking.
1497
1498Here is an example of backtracking: Let's say you want to find the
1499word following "foo" in the string "Food is on the foo table.":
1500
1501 $_ = "Food is on the foo table.";
1502 if ( /\b(foo)\s+(\w+)/i ) {
1503 print "$2 follows $1.\n";
1504 }
1505
1506When the match runs, the first part of the regular expression (C<\b(foo)>)
1507finds a possible match right at the beginning of the string, and loads up
1508$1 with "Foo". However, as soon as the matching engine sees that there's
1509no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1510mistake and starts over again one character after where it had the
c07a80fd
PP
1511tentative match. This time it goes all the way until the next occurrence
1512of "foo". The complete regular expression matches this time, and you get
1513the expected output of "table follows foo."
1514
1515Sometimes minimal matching can help a lot. Imagine you'd like to match
1516everything between "foo" and "bar". Initially, you write something
1517like this:
1518
1519 $_ = "The food is under the bar in the barn.";
1520 if ( /foo(.*)bar/ ) {
1521 print "got <$1>\n";
1522 }
1523
1524Which perhaps unexpectedly yields:
1525
1526 got <d is under the bar in the >
1527
1528That's because C<.*> was greedy, so you get everything between the
14218588 1529I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1530to use minimal matching to make sure you get the text between a "foo"
1531and the first "bar" thereafter.
1532
1533 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1534 got <d is under the >
1535
0d017f4d 1536Here's another example. Let's say you'd like to match a number at the end
b6e13d97 1537of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
1538So you write this:
1539
1540 $_ = "I have 2 numbers: 53147";
1541 if ( /(.*)(\d*)/ ) { # Wrong!
1542 print "Beginning is <$1>, number is <$2>.\n";
1543 }
1544
1545That won't work at all, because C<.*> was greedy and gobbled up the
1546whole string. As C<\d*> can match on an empty string the complete
1547regular expression matched successfully.
1548
8e1088bc 1549 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
1550
1551Here are some variants, most of which don't work:
1552
1553 $_ = "I have 2 numbers: 53147";
1554 @pats = qw{
1555 (.*)(\d*)
1556 (.*)(\d+)
1557 (.*?)(\d*)
1558 (.*?)(\d+)
1559 (.*)(\d+)$
1560 (.*?)(\d+)$
1561 (.*)\b(\d+)$
1562 (.*\D)(\d+)$
1563 };
1564
1565 for $pat (@pats) {
1566 printf "%-12s ", $pat;
1567 if ( /$pat/ ) {
1568 print "<$1> <$2>\n";
1569 } else {
1570 print "FAIL\n";
1571 }
1572 }
1573
1574That will print out:
1575
1576 (.*)(\d*) <I have 2 numbers: 53147> <>
1577 (.*)(\d+) <I have 2 numbers: 5314> <7>
1578 (.*?)(\d*) <> <>
1579 (.*?)(\d+) <I have > <2>
1580 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1581 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1582 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1583 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1584
1585As you see, this can be a bit tricky. It's important to realize that a
1586regular expression is merely a set of assertions that gives a definition
1587of success. There may be 0, 1, or several different ways that the
1588definition might succeed against a particular string. And if there are
5a964f20
TC
1589multiple ways it might succeed, you need to understand backtracking to
1590know which variety of success you will achieve.
c07a80fd 1591
19799a22 1592When using look-ahead assertions and negations, this can all get even
8b19b778 1593trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
1594followed by "123". You might try to write that as
1595
871b0233
IZ
1596 $_ = "ABC123";
1597 if ( /^\D*(?!123)/ ) { # Wrong!
1598 print "Yup, no 123 in $_\n";
1599 }
c07a80fd
PP
1600
1601But that isn't going to match; at least, not the way you're hoping. It
1602claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1603why that pattern matches, contrary to popular expectations:
c07a80fd 1604
4358a253
SS
1605 $x = 'ABC123';
1606 $y = 'ABC445';
c07a80fd 1607
4358a253
SS
1608 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1609 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1610
4358a253
SS
1611 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1612 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
1613
1614This prints
1615
1616 2: got ABC
1617 3: got AB
1618 4: got ABC
1619
5f05dabc 1620You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
1621general purpose version of test 1. The important difference between
1622them is that test 3 contains a quantifier (C<\D*>) and so can use
1623backtracking, whereas test 1 will not. What's happening is
1624that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1625non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1626let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1627fail.
14218588 1628
c07a80fd 1629The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1630try to match C<(?!123> with "123", which fails. But because
c07a80fd
PP
1631a quantifier (C<\D*>) has been used in the regular expression, the
1632search engine can backtrack and retry the match differently
54310121 1633in the hope of matching the complete regular expression.
c07a80fd 1634
5a964f20
TC
1635The pattern really, I<really> wants to succeed, so it uses the
1636standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1637time. Now there's indeed something following "AB" that is not
14218588 1638"123". It's "C123", which suffices.
c07a80fd 1639
14218588
GS
1640We can deal with this by using both an assertion and a negation.
1641We'll say that the first part in $1 must be followed both by a digit
1642and by something that's not "123". Remember that the look-aheads
1643are zero-width expressions--they only look, but don't consume any
1644of the string in their match. So rewriting this way produces what
c07a80fd
PP
1645you'd expect; that is, case 5 will fail, but case 6 succeeds:
1646
4358a253
SS
1647 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1648 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
1649
1650 6: got ABC
1651
5a964f20 1652In other words, the two zero-width assertions next to each other work as though
19799a22 1653they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
1654matches only if you're at the beginning of the line AND the end of the
1655line simultaneously. The deeper underlying truth is that juxtaposition in
1656regular expressions always means AND, except when you write an explicit OR
1657using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1658although the attempted matches are made at different positions because "a"
1659is not a zero-width assertion, but a one-width assertion.
1660
0d017f4d 1661B<WARNING>: Particularly complicated regular expressions can take
14218588 1662exponential time to solve because of the immense number of possible
0d017f4d 1663ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
1664internal optimizations done by the regular expression engine, this will
1665take a painfully long time to run:
c07a80fd 1666
e1901655
IZ
1667 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1668
1669And if you used C<*>'s in the internal groups instead of limiting them
1670to 0 through 5 matches, then it would take forever--or until you ran
1671out of stack space. Moreover, these internal optimizations are not
1672always applicable. For example, if you put C<{0,5}> instead of C<*>
1673on the external group, no current optimization is applicable, and the
1674match takes a long time to finish.
c07a80fd 1675
9da458fc
IZ
1676A powerful tool for optimizing such beasts is what is known as an
1677"independent group",
c47ff5f1 1678which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1679zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 1680the tail match, since they are in "logical" context: only
14218588 1681whether they match is considered relevant. For an example
9da458fc 1682where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1683following match, see L<C<< (?>pattern) >>>.
c277df42 1684
a0d0e21e 1685=head2 Version 8 Regular Expressions
d74e8afc 1686X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1687
5a964f20 1688In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1689routines, here are the pattern-matching rules not described above.
1690
54310121 1691Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1692with a special meaning described here or above. You can cause
5a964f20 1693characters that normally function as metacharacters to be interpreted
5f05dabc 1694literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
1695character; "\\" matches a "\"). This escape mechanism is also required
1696for the character used as the pattern delimiter.
1697
1698A series of characters matches that series of characters in the target
1699string, so the pattern C<blurfl> would match "blurfl" in the target
1700string.
a0d0e21e
LW
1701
1702You can specify a character class, by enclosing a list of characters
5d458dd8 1703in C<[]>, which will match any character from the list. If the
a0d0e21e 1704first character after the "[" is "^", the class matches any character not
14218588 1705in the list. Within a list, the "-" character specifies a
5a964f20 1706range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1707inclusive. If you want either "-" or "]" itself to be a member of a
1708class, put it at the start of the list (possibly after a "^"), or
1709escape it with a backslash. "-" is also taken literally when it is
1710at the end of the list, just before the closing "]". (The
84850974
DD
1711following all specify the same class of three characters: C<[-az]>,
1712C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
1713specifies a class containing twenty-six characters, even on EBCDIC-based
1714character sets.) Also, if you try to use the character
1715classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1716a range, the "-" is understood literally.
a0d0e21e 1717
8ada0baa
JH
1718Note also that the whole range idea is rather unportable between
1719character sets--and even within character sets they may cause results
1720you probably didn't expect. A sound principle is to use only ranges
0d017f4d 1721that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
1722[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1723spell out the character sets in full.
1724
54310121 1725Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1726used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1727"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
5d458dd8
YO
1728of octal digits, matches the character whose coded character set value
1729is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
1730matches the character whose numeric value is I<nn>. The expression \cI<x>
1731matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 1732matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1733
1734You can specify a series of alternatives for a pattern using "|" to
1735separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1736or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1737first alternative includes everything from the last pattern delimiter
1738("(", "[", or the beginning of the pattern) up to the first "|", and
1739the last alternative contains everything from the last "|" to the next
14218588
GS
1740pattern delimiter. That's why it's common practice to include
1741alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1742start and end.
1743
5a964f20 1744Alternatives are tried from left to right, so the first
a3cb178b
GS
1745alternative found for which the entire expression matches, is the one that
1746is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1747example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1748part will match, as that is the first alternative tried, and it successfully
1749matches the target string. (This might not seem important, but it is
1750important when you are capturing matched text using parentheses.)
1751
5a964f20 1752Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1753so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1754
14218588
GS
1755Within a pattern, you may designate subpatterns for later reference
1756by enclosing them in parentheses, and you may refer back to the
1757I<n>th subpattern later in the pattern using the metacharacter
1758\I<n>. Subpatterns are numbered based on the left to right order
1759of their opening parenthesis. A backreference matches whatever
1760actually matched the subpattern in the string being examined, not
1761the rules for that subpattern. Therefore, C<(0|0x)\d*\s\1\d*> will
1762match "0x1234 0x4321", but not "0x1234 01234", because subpattern
17631 matched "0x", even though the rule C<0|0x> could potentially match
1764the leading 0 in the second number.
cb1a09d0 1765
0d017f4d 1766=head2 Warning on \1 Instead of $1
cb1a09d0 1767
5a964f20 1768Some people get too used to writing things like:
cb1a09d0
AD
1769
1770 $pattern =~ s/(\W)/\\\1/g;
1771
1772This is grandfathered for the RHS of a substitute to avoid shocking the
1773B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1774PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1775the usual double-quoted string means a control-A. The customary Unix
1776meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1777of doing that, you get yourself into trouble if you then add an C</e>
1778modifier.
1779
5a964f20 1780 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1781
1782Or if you try to do
1783
1784 s/(\d+)/\1000/;
1785
1786You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1787C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1788with the operation of matching a backreference. Certainly they mean two
1789different things on the I<left> side of the C<s///>.
9fa51da4 1790
0d017f4d 1791=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 1792
19799a22 1793B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1794
1795Regular expressions provide a terse and powerful programming language. As
1796with most other power tools, power comes together with the ability
1797to wreak havoc.
1798
1799A common abuse of this power stems from the ability to make infinite
628afcb5 1800loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1801
1802 'foo' =~ m{ ( o? )* }x;
1803
0d017f4d 1804The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 1805in the string is not moved by the match, C<o?> would match again and again
527e91da 1806because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
1807is with the looping modifier C<//g>:
1808
1809 @matches = ( 'foo' =~ m{ o? }xg );
1810
1811or
1812
1813 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1814
1815or the loop implied by split().
1816
1817However, long experience has shown that many programming tasks may
14218588
GS
1818be significantly simplified by using repeated subexpressions that
1819may match zero-length substrings. Here's a simple example being:
c84d73f1
IZ
1820
1821 @chars = split //, $string; # // is not magic in split
1822 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1823
9da458fc 1824Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 1825the infinite loop>. The rules for this are different for lower-level
527e91da 1826loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
1827ones like the C</g> modifier or split() operator.
1828
19799a22
GS
1829The lower-level loops are I<interrupted> (that is, the loop is
1830broken) when Perl detects that a repeated expression matched a
1831zero-length substring. Thus
c84d73f1
IZ
1832
1833 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1834
5d458dd8 1835is made equivalent to
c84d73f1 1836
5d458dd8
YO
1837 m{ (?: NON_ZERO_LENGTH )*
1838 |
1839 (?: ZERO_LENGTH )?
c84d73f1
IZ
1840 }x;
1841
1842The higher level-loops preserve an additional state between iterations:
5d458dd8 1843whether the last match was zero-length. To break the loop, the following
c84d73f1 1844match after a zero-length match is prohibited to have a length of zero.
5d458dd8 1845This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
1846and so the I<second best> match is chosen if the I<best> match is of
1847zero length.
1848
19799a22 1849For example:
c84d73f1
IZ
1850
1851 $_ = 'bar';
1852 s/\w??/<$&>/g;
1853
20fb949f 1854results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 1855match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
1856best> match is what is matched by C<\w>. Thus zero-length matches
1857alternate with one-character-long matches.
1858
5d458dd8 1859Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
1860position one notch further in the string.
1861
19799a22 1862The additional state of being I<matched with zero-length> is associated with
c84d73f1 1863the matched string, and is reset by each assignment to pos().
9da458fc
IZ
1864Zero-length matches at the end of the previous match are ignored
1865during C<split>.
c84d73f1 1866
0d017f4d 1867=head2 Combining RE Pieces
35a734be
IZ
1868
1869Each of the elementary pieces of regular expressions which were described
1870before (such as C<ab> or C<\Z>) could match at most one substring
1871at the given position of the input string. However, in a typical regular
1872expression these elementary pieces are combined into more complicated
1873patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1874(in these examples C<S> and C<T> are regular subexpressions).
1875
1876Such combinations can include alternatives, leading to a problem of choice:
1877if we match a regular expression C<a|ab> against C<"abc">, will it match
1878substring C<"a"> or C<"ab">? One way to describe which substring is
1879actually matched is the concept of backtracking (see L<"Backtracking">).
1880However, this description is too low-level and makes you think
1881in terms of a particular implementation.
1882
1883Another description starts with notions of "better"/"worse". All the
1884substrings which may be matched by the given regular expression can be
1885sorted from the "best" match to the "worst" match, and it is the "best"
1886match which is chosen. This substitutes the question of "what is chosen?"
1887by the question of "which matches are better, and which are worse?".
1888
1889Again, for elementary pieces there is no such question, since at most
1890one match at a given position is possible. This section describes the
1891notion of better/worse for combining operators. In the description
1892below C<S> and C<T> are regular subexpressions.
1893
13a2d996 1894=over 4
35a734be
IZ
1895
1896=item C<ST>
1897
1898Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
1899substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 1900which can be matched by C<T>.
35a734be
IZ
1901
1902If C<A> is better match for C<S> than C<A'>, C<AB> is a better
1903match than C<A'B'>.
1904
1905If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
1906C<B> is better match for C<T> than C<B'>.
1907
1908=item C<S|T>
1909
1910When C<S> can match, it is a better match than when only C<T> can match.
1911
1912Ordering of two matches for C<S> is the same as for C<S>. Similar for
1913two matches for C<T>.
1914
1915=item C<S{REPEAT_COUNT}>
1916
1917Matches as C<SSS...S> (repeated as many times as necessary).
1918
1919=item C<S{min,max}>
1920
1921Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
1922
1923=item C<S{min,max}?>
1924
1925Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
1926
1927=item C<S?>, C<S*>, C<S+>
1928
1929Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
1930
1931=item C<S??>, C<S*?>, C<S+?>
1932
1933Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
1934
c47ff5f1 1935=item C<< (?>S) >>
35a734be
IZ
1936
1937Matches the best match for C<S> and only that.
1938
1939=item C<(?=S)>, C<(?<=S)>
1940
1941Only the best match for C<S> is considered. (This is important only if
1942C<S> has capturing parentheses, and backreferences are used somewhere
1943else in the whole regular expression.)
1944
1945=item C<(?!S)>, C<(?<!S)>
1946
1947For this grouping operator there is no need to describe the ordering, since
1948only whether or not C<S> can match is important.
1949
6bda09f9 1950=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
1951
1952The ordering is the same as for the regular expression which is
6bda09f9 1953the result of EXPR, or the pattern contained by capture buffer PARNO.
35a734be
IZ
1954
1955=item C<(?(condition)yes-pattern|no-pattern)>
1956
1957Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
1958already determined. The ordering of the matches is the same as for the
1959chosen subexpression.
1960
1961=back
1962
1963The above recipes describe the ordering of matches I<at a given position>.
1964One more rule is needed to understand how a match is determined for the
1965whole regular expression: a match at an earlier position is always better
1966than a match at a later position.
1967
0d017f4d 1968=head2 Creating Custom RE Engines
c84d73f1
IZ
1969
1970Overloaded constants (see L<overload>) provide a simple way to extend
1971the functionality of the RE engine.
1972
1973Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 1974matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
1975characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
1976at these positions, so we want to have each C<\Y|> in the place of the
1977more complicated version. We can create a module C<customre> to do
1978this:
1979
1980 package customre;
1981 use overload;
1982
1983 sub import {
1984 shift;
1985 die "No argument to customre::import allowed" if @_;
1986 overload::constant 'qr' => \&convert;
1987 }
1988
1989 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
1990
580a9fe1
RGS
1991 # We must also take care of not escaping the legitimate \\Y|
1992 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 1993 my %rules = ( '\\' => '\\\\',
c84d73f1
IZ
1994 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
1995 sub convert {
1996 my $re = shift;
5d458dd8 1997 $re =~ s{
c84d73f1
IZ
1998 \\ ( \\ | Y . )
1999 }
5d458dd8 2000 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2001 return $re;
2002 }
2003
2004Now C<use customre> enables the new escape in constant regular
2005expressions, i.e., those without any runtime variable interpolations.
2006As documented in L<overload>, this conversion will work only over
2007literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2008part of this regular expression needs to be converted explicitly
2009(but only if the special meaning of C<\Y|> should be enabled inside $re):
2010
2011 use customre;
2012 $re = <>;
2013 chomp $re;
2014 $re = customre::convert $re;
2015 /\Y|$re\Y|/;
2016
1f1031fe
YO
2017=head1 PCRE/Python Support
2018
99d59c4d 2019As of Perl 5.10.0, Perl supports several Python/PCRE specific extensions
1f1031fe 2020to the regex syntax. While Perl programmers are encouraged to use the
99d59c4d 2021Perl specific syntax, the following are also accepted:
1f1031fe
YO
2022
2023=over 4
2024
ae5648b3 2025=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe
YO
2026
2027Define a named capture buffer. Equivalent to C<< (?<NAME>pattern) >>.
2028
2029=item C<< (?P=NAME) >>
2030
2031Backreference to a named capture buffer. Equivalent to C<< \g{NAME} >>.
2032
2033=item C<< (?P>NAME) >>
2034
2035Subroutine call to a named capture buffer. Equivalent to C<< (?&NAME) >>.
2036
ee9b8eae 2037=back
1f1031fe 2038
19799a22
GS
2039=head1 BUGS
2040
78288b8e
KW
2041There are numerous problems with case insensitive matching of characters
2042outside the ASCII range, especially with those whose folds are multiple
2043characters, such as ligatures like C<LATIN SMALL LIGATURE FF>.
2044
f253210b
KW
2045In a bracketed character class with case insensitive matching, ranges only work
2046for ASCII characters. For example,
2047C<m/[\N{CYRILLIC CAPITAL LETTER A}-\N{CYRILLIC CAPITAL LETTER YA}]/i>
2048doesn't match all the Russian upper and lower case letters.
2049
88c9975e
KW
2050Many regular expression constructs don't work on EBCDIC platforms.
2051
9da458fc
IZ
2052This document varies from difficult to understand to completely
2053and utterly opaque. The wandering prose riddled with jargon is
2054hard to fathom in several places.
2055
2056This document needs a rewrite that separates the tutorial content
2057from the reference content.
19799a22
GS
2058
2059=head1 SEE ALSO
9fa51da4 2060
91e0c79e
MJD
2061L<perlrequick>.
2062
2063L<perlretut>.
2064
9b599b2a
GS
2065L<perlop/"Regexp Quote-Like Operators">.
2066
1e66bd83
PP
2067L<perlop/"Gory details of parsing quoted constructs">.
2068
14218588
GS
2069L<perlfaq6>.
2070
9b599b2a
GS
2071L<perlfunc/pos>.
2072
2073L<perllocale>.
2074
fb55449c
JH
2075L<perlebcdic>.
2076
14218588
GS
2077I<Mastering Regular Expressions> by Jeffrey Friedl, published
2078by O'Reilly and Associates.