This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add Regexp::Keep \K functionality to regex engine as well as add \v and \V, cleanup...
[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
19799a22 19Matching operations can have various modifiers. Modifiers
5a964f20 20that relate to the interpretation of the regular expression inside
19799a22 21are listed below. Modifiers that alter the way a regular expression
5d458dd8 22is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
1e66bd83 23L<perlop/"Gory details of parsing quoted constructs">.
a0d0e21e 24
55497cff
PP
25=over 4
26
27=item i
d74e8afc
ITB
28X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
29X<regular expression, case-insensitive>
55497cff
PP
30
31Do case-insensitive pattern matching.
32
a034a98d
DD
33If C<use locale> is in effect, the case map is taken from the current
34locale. See L<perllocale>.
35
54310121 36=item m
d74e8afc 37X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff
PP
38
39Treat string as multiple lines. That is, change "^" and "$" from matching
14218588 40the start or end of the string to matching the start or end of any
7f761169 41line anywhere within the string.
55497cff 42
54310121 43=item s
d74e8afc
ITB
44X</s> X<regex, single-line> X<regexp, single-line>
45X<regular expression, single-line>
55497cff
PP
46
47Treat string as single line. That is, change "." to match any character
19799a22 48whatsoever, even a newline, which normally it would not match.
55497cff 49
f02c194e 50Used together, as /ms, they let the "." match any character whatsoever,
fb55449c 51while still allowing "^" and "$" to match, respectively, just after
19799a22 52and just before newlines within the string.
7b8d334a 53
54310121 54=item x
d74e8afc 55X</x>
55497cff
PP
56
57Extend your pattern's legibility by permitting whitespace and comments.
58
59=back
a0d0e21e
LW
60
61These are usually written as "the C</x> modifier", even though the delimiter
14218588 62in question might not really be a slash. Any of these
a0d0e21e 63modifiers may also be embedded within the regular expression itself using
14218588 64the C<(?...)> construct. See below.
a0d0e21e 65
4633a7c4 66The C</x> modifier itself needs a little more explanation. It tells
55497cff
PP
67the regular expression parser to ignore whitespace that is neither
68backslashed nor within a character class. You can use this to break up
4633a7c4 69your regular expression into (slightly) more readable parts. The C<#>
54310121 70character is also treated as a metacharacter introducing a comment,
55497cff 71just as in ordinary Perl code. This also means that if you want real
14218588 72whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a
RGS
73class, where they are unaffected by C</x>), then you'll either have to
74escape them (using backslashes or C<\Q...\E>) or encode them using octal
8933a740
RGS
75or hex escapes. Taken together, these features go a long way towards
76making Perl's regular expressions more readable. Note that you have to
77be careful not to include the pattern delimiter in the comment--perl has
78no way of knowing you did not intend to close the pattern early. See
79the C-comment deletion code in L<perlop>. Also note that anything inside
1031e5db 80a C<\Q...\E> stays unaffected by C</x>.
d74e8afc 81X</x>
a0d0e21e
LW
82
83=head2 Regular Expressions
84
04838cea
RGS
85=head3 Metacharacters
86
19799a22 87The patterns used in Perl pattern matching derive from supplied in
14218588 88the Version 8 regex routines. (The routines are derived
19799a22
GS
89(distantly) from Henry Spencer's freely redistributable reimplementation
90of the V8 routines.) See L<Version 8 Regular Expressions> for
91details.
a0d0e21e
LW
92
93In particular the following metacharacters have their standard I<egrep>-ish
94meanings:
d74e8afc
ITB
95X<metacharacter>
96X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
97
a0d0e21e 98
54310121 99 \ Quote the next metacharacter
a0d0e21e
LW
100 ^ Match the beginning of the line
101 . Match any character (except newline)
c07a80fd 102 $ Match the end of the line (or before newline at the end)
a0d0e21e
LW
103 | Alternation
104 () Grouping
105 [] Character class
106
14218588
GS
107By default, the "^" character is guaranteed to match only the
108beginning of the string, the "$" character only the end (or before the
109newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
110assumption that the string contains only one line. Embedded newlines
111will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 112string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
113newline within the string (except if the newline is the last character in
114the string), and "$" will match before any newline. At the
a0d0e21e
LW
115cost of a little more overhead, you can do this by using the /m modifier
116on the pattern match operator. (Older programs did this by setting C<$*>,
f02c194e 117but this practice has been removed in perl 5.9.)
d74e8afc 118X<^> X<$> X</m>
a0d0e21e 119
14218588 120To simplify multi-line substitutions, the "." character never matches a
55497cff 121newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 122the string is a single line--even if it isn't.
d74e8afc 123X<.> X</s>
a0d0e21e 124
04838cea
RGS
125=head3 Quantifiers
126
a0d0e21e 127The following standard quantifiers are recognized:
d74e8afc 128X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e
LW
129
130 * Match 0 or more times
131 + Match 1 or more times
132 ? Match 1 or 0 times
133 {n} Match exactly n times
134 {n,} Match at least n times
135 {n,m} Match at least n but not more than m times
136
137(If a curly bracket occurs in any other context, it is treated
b975c076
JL
138as a regular character. In particular, the lower bound
139is not optional.) The "*" modifier is equivalent to C<{0,}>, the "+"
25f94b33 140modifier to C<{1,}>, and the "?" modifier to C<{0,1}>. n and m are limited
9c79236d
GS
141to integral values less than a preset limit defined when perl is built.
142This is usually 32766 on the most common platforms. The actual limit can
143be seen in the error message generated by code such as this:
144
820475bd 145 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 146
54310121
PP
147By default, a quantified subpattern is "greedy", that is, it will match as
148many times as possible (given a particular starting location) while still
149allowing the rest of the pattern to match. If you want it to match the
150minimum number of times possible, follow the quantifier with a "?". Note
151that the meanings don't change, just the "greediness":
d74e8afc
ITB
152X<metacharacter> X<greedy> X<greedyness>
153X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e
LW
154
155 *? Match 0 or more times
156 +? Match 1 or more times
157 ?? Match 0 or 1 time
158 {n}? Match exactly n times
159 {n,}? Match at least n times
160 {n,m}? Match at least n but not more than m times
161
b9b4dddf
YO
162By default, when a quantified subpattern does not allow the rest of the
163overall pattern to match, Perl will backtrack. However, this behaviour is
164sometimes undesirable. Thus Perl provides the "possesive" quantifier form
165as well.
166
167 *+ Match 0 or more times and give nothing back
04838cea 168 ++ Match 1 or more times and give nothing back
b9b4dddf
YO
169 ?+ Match 0 or 1 time and give nothing back
170 {n}+ Match exactly n times and give nothing back (redundant)
04838cea
RGS
171 {n,}+ Match at least n times and give nothing back
172 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
173
174For instance,
175
176 'aaaa' =~ /a++a/
177
178will never match, as the C<a++> will gobble up all the C<a>'s in the
179string and won't leave any for the remaining part of the pattern. This
180feature can be extremely useful to give perl hints about where it
181shouldn't backtrack. For instance, the typical "match a double-quoted
182string" problem can be most efficiently performed when written as:
183
184 /"(?:[^"\\]++|\\.)*+"/
185
186as we know that if the final quote does not match, bactracking will not
187help. See the independent subexpression C<< (?>...) >> for more details;
188possessive quantifiers are just syntactic sugar for that construct. For
189instance the above example could also be written as follows:
190
191 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
192
04838cea
RGS
193=head3 Escape sequences
194
5f05dabc 195Because patterns are processed as double quoted strings, the following
a0d0e21e 196also work:
d74e8afc
ITB
197X<\t> X<\n> X<\r> X<\f> X<\a> X<\l> X<\u> X<\L> X<\U> X<\E> X<\Q>
198X<\0> X<\c> X<\N> X<\x>
a0d0e21e 199
0f36ee90
PP
200 \t tab (HT, TAB)
201 \n newline (LF, NL)
202 \r return (CR)
203 \f form feed (FF)
204 \a alarm (bell) (BEL)
205 \e escape (think troff) (ESC)
cb1a09d0
AD
206 \033 octal char (think of a PDP-11)
207 \x1B hex char
a0ed51b3 208 \x{263a} wide hex char (Unicode SMILEY)
a0d0e21e 209 \c[ control char
4a2d328f 210 \N{name} named char
cb1a09d0
AD
211 \l lowercase next char (think vi)
212 \u uppercase next char (think vi)
213 \L lowercase till \E (think vi)
214 \U uppercase till \E (think vi)
215 \E end case modification (think vi)
5a964f20 216 \Q quote (disable) pattern metacharacters till \E
a0d0e21e 217
a034a98d 218If C<use locale> is in effect, the case map used by C<\l>, C<\L>, C<\u>
423cee85 219and C<\U> is taken from the current locale. See L<perllocale>. For
4a2d328f 220documentation of C<\N{name}>, see L<charnames>.
a034a98d 221
1d2dff63
GS
222You cannot include a literal C<$> or C<@> within a C<\Q> sequence.
223An unescaped C<$> or C<@> interpolates the corresponding variable,
224while escaping will cause the literal string C<\$> to be matched.
225You'll need to write something like C<m/\Quser\E\@\Qhost/>.
226
04838cea
RGS
227=head3 Character classes
228
a0d0e21e 229In addition, Perl defines the following:
d74e8afc
ITB
230X<metacharacter>
231X<\w> X<\W> X<\s> X<\S> X<\d> X<\D> X<\X> X<\p> X<\P> X<\C>
232X<word> X<whitespace>
a0d0e21e 233
81714fb9
YO
234 \w Match a "word" character (alphanumeric plus "_")
235 \W Match a non-"word" character
236 \s Match a whitespace character
237 \S Match a non-whitespace character
238 \d Match a digit character
239 \D Match a non-digit character
240 \pP Match P, named property. Use \p{Prop} for longer names.
241 \PP Match non-P
242 \X Match eXtended Unicode "combining character sequence",
243 equivalent to (?:\PM\pM*)
244 \C Match a single C char (octet) even under Unicode.
245 NOTE: breaks up characters into their UTF-8 bytes,
246 so you may end up with malformed pieces of UTF-8.
247 Unsupported in lookbehind.
5d458dd8 248 \1 Backreference to a specific group.
c74340f9 249 '1' may actually be any positive integer.
2bf803e2
YO
250 \g1 Backreference to a specific or previous group,
251 \g{-1} number may be negative indicating a previous buffer and may
252 optionally be wrapped in curly brackets for safer parsing.
1f1031fe 253 \g{name} Named backreference
81714fb9 254 \k<name> Named backreference
5d458dd8 255 \N{name} Named unicode character, or unicode escape
81714fb9
YO
256 \x12 Hexadecimal escape sequence
257 \x{1234} Long hexadecimal escape sequence
ee9b8eae
YO
258 \K Keep the stuff left of the \K, don't include it in $&
259 \v Shortcut for (*PRUNE)
260 \V Shortcut for (*SKIP)
a0d0e21e 261
08ce8fc6
JH
262A C<\w> matches a single alphanumeric character (an alphabetic
263character, or a decimal digit) or C<_>, not a whole word. Use C<\w+>
264to match a string of Perl-identifier characters (which isn't the same
265as matching an English word). If C<use locale> is in effect, the list
266of alphabetic characters generated by C<\w> is taken from the current
267locale. See L<perllocale>. You may use C<\w>, C<\W>, C<\s>, C<\S>,
1209ba90 268C<\d>, and C<\D> within character classes, but if you try to use them
08ce8fc6
JH
269as endpoints of a range, that's not a range, the "-" is understood
270literally. If Unicode is in effect, C<\s> matches also "\x{85}",
271"\x{2028}, and "\x{2029}", see L<perlunicode> for more details about
491fd90a 272C<\pP>, C<\PP>, and C<\X>, and L<perluniintro> about Unicode in general.
fa11829f 273You can define your own C<\p> and C<\P> properties, see L<perlunicode>.
d74e8afc 274X<\w> X<\W> X<word>
a0d0e21e 275
b8c5462f 276The POSIX character class syntax
d74e8afc 277X<character class>
b8c5462f 278
820475bd 279 [:class:]
b8c5462f 280
5496314a
SP
281is also available. Note that the C<[> and C<]> braces are I<literal>;
282they must always be used within a character class expression.
283
284 # this is correct:
285 $string =~ /[[:alpha:]]/;
286
287 # this is not, and will generate a warning:
288 $string =~ /[:alpha:]/;
289
290The available classes and their backslash equivalents (if available) are
291as follows:
d74e8afc
ITB
292X<character class>
293X<alpha> X<alnum> X<ascii> X<blank> X<cntrl> X<digit> X<graph>
294X<lower> X<print> X<punct> X<space> X<upper> X<word> X<xdigit>
b8c5462f
JH
295
296 alpha
297 alnum
298 ascii
aaa51d5e 299 blank [1]
b8c5462f
JH
300 cntrl
301 digit \d
302 graph
303 lower
304 print
305 punct
aaa51d5e 306 space \s [2]
b8c5462f 307 upper
aaa51d5e 308 word \w [3]
b8c5462f
JH
309 xdigit
310
07698885
RGS
311=over
312
313=item [1]
314
b432a672 315A GNU extension equivalent to C<[ \t]>, "all horizontal whitespace".
07698885
RGS
316
317=item [2]
318
319Not exactly equivalent to C<\s> since the C<[[:space:]]> includes
b432a672 320also the (very rare) "vertical tabulator", "\ck", chr(11).
07698885
RGS
321
322=item [3]
323
08ce8fc6 324A Perl extension, see above.
07698885
RGS
325
326=back
aaa51d5e 327
26b44a0a 328For example use C<[:upper:]> to match all the uppercase characters.
aaa51d5e
JF
329Note that the C<[]> are part of the C<[::]> construct, not part of the
330whole character class. For example:
b8c5462f 331
820475bd 332 [01[:alpha:]%]
b8c5462f 333
593df60c 334matches zero, one, any alphabetic character, and the percentage sign.
b8c5462f 335
72ff2908
JH
336The following equivalences to Unicode \p{} constructs and equivalent
337backslash character classes (if available), will hold:
d74e8afc 338X<character class> X<\p> X<\p{}>
72ff2908 339
5496314a 340 [[:...:]] \p{...} backslash
b8c5462f
JH
341
342 alpha IsAlpha
343 alnum IsAlnum
344 ascii IsASCII
b432a672 345 blank IsSpace
b8c5462f 346 cntrl IsCntrl
3bec3564 347 digit IsDigit \d
b8c5462f
JH
348 graph IsGraph
349 lower IsLower
350 print IsPrint
351 punct IsPunct
352 space IsSpace
3bec3564 353 IsSpacePerl \s
b8c5462f
JH
354 upper IsUpper
355 word IsWord
356 xdigit IsXDigit
357
5496314a 358For example C<[[:lower:]]> and C<\p{IsLower}> are equivalent.
b8c5462f
JH
359
360If the C<utf8> pragma is not used but the C<locale> pragma is, the
aaa51d5e 361classes correlate with the usual isalpha(3) interface (except for
b432a672 362"word" and "blank").
b8c5462f
JH
363
364The assumedly non-obviously named classes are:
365
366=over 4
367
368=item cntrl
d74e8afc 369X<cntrl>
b8c5462f 370
820475bd
GS
371Any control character. Usually characters that don't produce output as
372such but instead control the terminal somehow: for example newline and
373backspace are control characters. All characters with ord() less than
593df60c 37432 are most often classified as control characters (assuming ASCII,
7be5a6cf
JF
375the ISO Latin character sets, and Unicode), as is the character with
376the ord() value of 127 (C<DEL>).
b8c5462f
JH
377
378=item graph
d74e8afc 379X<graph>
b8c5462f 380
f1cbbd6e 381Any alphanumeric or punctuation (special) character.
b8c5462f
JH
382
383=item print
d74e8afc 384X<print>
b8c5462f 385
f79b3095 386Any alphanumeric or punctuation (special) character or the space character.
b8c5462f
JH
387
388=item punct
d74e8afc 389X<punct>
b8c5462f 390
f1cbbd6e 391Any punctuation (special) character.
b8c5462f
JH
392
393=item xdigit
d74e8afc 394X<xdigit>
b8c5462f 395
593df60c 396Any hexadecimal digit. Though this may feel silly ([0-9A-Fa-f] would
820475bd 397work just fine) it is included for completeness.
b8c5462f 398
b8c5462f
JH
399=back
400
401You can negate the [::] character classes by prefixing the class name
402with a '^'. This is a Perl extension. For example:
d74e8afc 403X<character class, negation>
b8c5462f 404
5496314a 405 POSIX traditional Unicode
93733859 406
5496314a
SP
407 [[:^digit:]] \D \P{IsDigit}
408 [[:^space:]] \S \P{IsSpace}
409 [[:^word:]] \W \P{IsWord}
b8c5462f 410
54c18d04
MK
411Perl respects the POSIX standard in that POSIX character classes are
412only supported within a character class. The POSIX character classes
413[.cc.] and [=cc=] are recognized but B<not> supported and trying to
414use them will cause an error.
b8c5462f 415
04838cea
RGS
416=head3 Assertions
417
a0d0e21e 418Perl defines the following zero-width assertions:
d74e8afc
ITB
419X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
420X<regexp, zero-width assertion>
421X<regular expression, zero-width assertion>
422X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e
LW
423
424 \b Match a word boundary
425 \B Match a non-(word boundary)
b85d18e9
IZ
426 \A Match only at beginning of string
427 \Z Match only at end of string, or before newline at the end
428 \z Match only at end of string
9da458fc
IZ
429 \G Match only at pos() (e.g. at the end-of-match position
430 of prior m//g)
a0d0e21e 431
14218588 432A word boundary (C<\b>) is a spot between two characters
19799a22
GS
433that has a C<\w> on one side of it and a C<\W> on the other side
434of it (in either order), counting the imaginary characters off the
435beginning and end of the string as matching a C<\W>. (Within
436character classes C<\b> represents backspace rather than a word
437boundary, just as it normally does in any double-quoted string.)
438The C<\A> and C<\Z> are just like "^" and "$", except that they
439won't match multiple times when the C</m> modifier is used, while
440"^" and "$" will match at every internal line boundary. To match
441the actual end of the string and not ignore an optional trailing
442newline, use C<\z>.
d74e8afc 443X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
444
445The C<\G> assertion can be used to chain global matches (using
446C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
447It is also useful when writing C<lex>-like scanners, when you have
448several patterns that you want to match against consequent substrings
449of your string, see the previous reference. The actual location
450where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d
YO
451an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
452matches is modified somewhat, in that contents to the left of C<\G> is
453not counted when determining the length of the match. Thus the following
454will not match forever:
d74e8afc 455X<\G>
c47ff5f1 456
58e23c8d
YO
457 $str = 'ABC';
458 pos($str) = 1;
459 while (/.\G/g) {
460 print $&;
461 }
462
463It will print 'A' and then terminate, as it considers the match to
464be zero-width, and thus will not match at the same position twice in a
465row.
466
467It is worth noting that C<\G> improperly used can result in an infinite
468loop. Take care when using patterns that include C<\G> in an alternation.
469
04838cea
RGS
470=head3 Capture buffers
471
14218588 472The bracketing construct C<( ... )> creates capture buffers. To
c47ff5f1 473refer to the digit'th buffer use \<digit> within the
14218588 474match. Outside the match use "$" instead of "\". (The
81714fb9 475\<digit> notation works in certain circumstances outside
14218588
GS
476the match. See the warning below about \1 vs $1 for details.)
477Referring back to another part of the match is called a
478I<backreference>.
d74e8afc
ITB
479X<regex, capture buffer> X<regexp, capture buffer>
480X<regular expression, capture buffer> X<backreference>
14218588
GS
481
482There is no limit to the number of captured substrings that you may
483use. However Perl also uses \10, \11, etc. as aliases for \010,
fb55449c
JH
484\011, etc. (Recall that 0 means octal, so \011 is the character at
485number 9 in your coded character set; which would be the 10th character,
81714fb9
YO
486a horizontal tab under ASCII.) Perl resolves this
487ambiguity by interpreting \10 as a backreference only if at least 10
488left parentheses have opened before it. Likewise \11 is a
489backreference only if at least 11 left parentheses have opened
490before it. And so on. \1 through \9 are always interpreted as
5624f11d 491backreferences.
c74340f9 492
1f1031fe 493X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
2bf803e2
YO
494In order to provide a safer and easier way to construct patterns using
495backrefs, in Perl 5.10 the C<\g{N}> notation is provided. The curly
496brackets are optional, however omitting them is less safe as the meaning
497of the pattern can be changed by text (such as digits) following it.
498When N is a positive integer the C<\g{N}> notation is exactly equivalent
499to using normal backreferences. When N is a negative integer then it is
500a relative backreference referring to the previous N'th capturing group.
1f1031fe
YO
501When the bracket form is used and N is not an integer, it is treated as a
502reference to a named buffer.
2bf803e2
YO
503
504Thus C<\g{-1}> refers to the last buffer, C<\g{-2}> refers to the
505buffer before that. For example:
5624f11d
YO
506
507 /
508 (Y) # buffer 1
509 ( # buffer 2
510 (X) # buffer 3
2bf803e2
YO
511 \g{-1} # backref to buffer 3
512 \g{-3} # backref to buffer 1
5624f11d
YO
513 )
514 /x
515
2bf803e2 516and would match the same as C</(Y) ( (X) \3 \1 )/x>.
14218588 517
81714fb9 518Additionally, as of Perl 5.10 you may use named capture buffers and named
1f1031fe
YO
519backreferences. The notation is C<< (?<name>...) >> to declare and C<< \k<name> >>
520to reference. You may also use single quotes instead of angle brackets to quote the
521name; and you may use the bracketed C<< \g{name} >> back reference syntax.
522The only difference between named capture buffers and unnamed ones is
81714fb9 523that multiple buffers may have the same name and that the contents of
1f1031fe 524named capture buffers are available via the C<%+> hash. When multiple
81714fb9
YO
525groups share the same name C<$+{name}> and C<< \k<name> >> refer to the
526leftmost defined group, thus it's possible to do things with named capture
527buffers that would otherwise require C<(??{})> code to accomplish. Named
528capture buffers are numbered just as normal capture buffers are and may be
529referenced via the magic numeric variables or via numeric backreferences
530as well as by name.
531
14218588 532Examples:
a0d0e21e
LW
533
534 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
535
81714fb9
YO
536 /(.)\1/ # find first doubled char
537 and print "'$1' is the first doubled character\n";
538
539 /(?<char>.)\k<char>/ # ... a different way
540 and print "'$+{char}' is the first doubled character\n";
541
542 /(?<char>.)\1/ # ... mix and match
543 and print "'$1' is the first doubled character\n";
c47ff5f1 544
14218588 545 if (/Time: (..):(..):(..)/) { # parse out values
a0d0e21e
LW
546 $hours = $1;
547 $minutes = $2;
548 $seconds = $3;
549 }
c47ff5f1 550
14218588
GS
551Several special variables also refer back to portions of the previous
552match. C<$+> returns whatever the last bracket match matched.
553C<$&> returns the entire matched string. (At one point C<$0> did
554also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
555everything before the matched string. C<$'> returns everything
556after the matched string. And C<$^N> contains whatever was matched by
557the most-recently closed group (submatch). C<$^N> can be used in
558extended patterns (see below), for example to assign a submatch to a
81714fb9 559variable.
d74e8afc 560X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 561
665e98b9 562The numbered match variables ($1, $2, $3, etc.) and the related punctuation
77ea4f6d 563set (C<$+>, C<$&>, C<$`>, C<$'>, and C<$^N>) are all dynamically scoped
14218588
GS
564until the end of the enclosing block or until the next successful
565match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
566X<$+> X<$^N> X<$&> X<$`> X<$'>
567X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
568
14218588 569
665e98b9 570B<NOTE>: failed matches in Perl do not reset the match variables,
5146ce24 571which makes it easier to write code that tests for a series of more
665e98b9
JH
572specific cases and remembers the best match.
573
14218588
GS
574B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
575C<$'> anywhere in the program, it has to provide them for every
576pattern match. This may substantially slow your program. Perl
577uses the same mechanism to produce $1, $2, etc, so you also pay a
578price for each pattern that contains capturing parentheses. (To
579avoid this cost while retaining the grouping behaviour, use the
580extended regular expression C<(?: ... )> instead.) But if you never
581use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
582parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
583if you can, but if you can't (and some algorithms really appreciate
584them), once you've used them once, use them at will, because you've
585already paid the price. As of 5.005, C<$&> is not so costly as the
586other two.
d74e8afc 587X<$&> X<$`> X<$'>
68dc0745 588
19799a22
GS
589Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
590C<\w>, C<\n>. Unlike some other regular expression languages, there
591are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 592that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
593interpreted as a literal character, not a metacharacter. This was
594once used in a common idiom to disable or quote the special meanings
595of regular expression metacharacters in a string that you want to
36bbe248 596use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
597
598 $pattern =~ s/(\W)/\\$1/g;
599
f1cbbd6e 600(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
601Today it is more common to use the quotemeta() function or the C<\Q>
602metaquoting escape sequence to disable all metacharacters' special
603meanings like this:
a0d0e21e
LW
604
605 /$unquoted\Q$quoted\E$unquoted/
606
9da458fc
IZ
607Beware that if you put literal backslashes (those not inside
608interpolated variables) between C<\Q> and C<\E>, double-quotish
609backslash interpolation may lead to confusing results. If you
610I<need> to use literal backslashes within C<\Q...\E>,
611consult L<perlop/"Gory details of parsing quoted constructs">.
612
19799a22
GS
613=head2 Extended Patterns
614
14218588
GS
615Perl also defines a consistent extension syntax for features not
616found in standard tools like B<awk> and B<lex>. The syntax is a
617pair of parentheses with a question mark as the first thing within
618the parentheses. The character after the question mark indicates
619the extension.
19799a22 620
14218588
GS
621The stability of these extensions varies widely. Some have been
622part of the core language for many years. Others are experimental
623and may change without warning or be completely removed. Check
624the documentation on an individual feature to verify its current
625status.
19799a22 626
14218588
GS
627A question mark was chosen for this and for the minimal-matching
628construct because 1) question marks are rare in older regular
629expressions, and 2) whenever you see one, you should stop and
630"question" exactly what is going on. That's psychology...
a0d0e21e
LW
631
632=over 10
633
cc6b7395 634=item C<(?#text)>
d74e8afc 635X<(?#)>
a0d0e21e 636
14218588 637A comment. The text is ignored. If the C</x> modifier enables
19799a22 638whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
639the comment as soon as it sees a C<)>, so there is no way to put a literal
640C<)> in the comment.
a0d0e21e 641
19799a22 642=item C<(?imsx-imsx)>
d74e8afc 643X<(?)>
19799a22 644
0b6d1084
JH
645One or more embedded pattern-match modifiers, to be turned on (or
646turned off, if preceded by C<->) for the remainder of the pattern or
647the remainder of the enclosing pattern group (if any). This is
648particularly useful for dynamic patterns, such as those read in from a
649configuration file, read in as an argument, are specified in a table
650somewhere, etc. Consider the case that some of which want to be case
651sensitive and some do not. The case insensitive ones need to include
652merely C<(?i)> at the front of the pattern. For example:
19799a22
GS
653
654 $pattern = "foobar";
5d458dd8 655 if ( /$pattern/i ) { }
19799a22
GS
656
657 # more flexible:
658
659 $pattern = "(?i)foobar";
5d458dd8 660 if ( /$pattern/ ) { }
19799a22 661
0b6d1084 662These modifiers are restored at the end of the enclosing group. For example,
19799a22
GS
663
664 ( (?i) blah ) \s+ \1
665
666will match a repeated (I<including the case>!) word C<blah> in any
14218588 667case, assuming C<x> modifier, and no C<i> modifier outside this
19799a22
GS
668group.
669
5a964f20 670=item C<(?:pattern)>
d74e8afc 671X<(?:)>
a0d0e21e 672
ca9dfc88
IZ
673=item C<(?imsx-imsx:pattern)>
674
5a964f20
TC
675This is for clustering, not capturing; it groups subexpressions like
676"()", but doesn't make backreferences as "()" does. So
a0d0e21e 677
5a964f20 678 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
679
680is like
681
5a964f20 682 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 683
19799a22
GS
684but doesn't spit out extra fields. It's also cheaper not to capture
685characters if you don't need to.
a0d0e21e 686
19799a22 687Any letters between C<?> and C<:> act as flags modifiers as with
5d458dd8 688C<(?imsx-imsx)>. For example,
ca9dfc88
IZ
689
690 /(?s-i:more.*than).*million/i
691
14218588 692is equivalent to the more verbose
ca9dfc88
IZ
693
694 /(?:(?s-i)more.*than).*million/i
695
ee9b8eae
YO
696=item Look-Around Assertions
697X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
698
699Look-around assertions are zero width patterns which match a specific
700pattern without including it in C<$&>. Positive assertions match when
701their subpattern matches, negative assertions match when their subpattern
702fails. Look-behind matches text up to the current match position,
703look-ahead matches text following the current match position.
704
705=over 4
706
5a964f20 707=item C<(?=pattern)>
d74e8afc 708X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 709
19799a22 710A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
711matches a word followed by a tab, without including the tab in C<$&>.
712
5a964f20 713=item C<(?!pattern)>
d74e8afc 714X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 715
19799a22 716A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 717matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
718however that look-ahead and look-behind are NOT the same thing. You cannot
719use this for look-behind.
7b8d334a 720
5a964f20 721If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
722will not do what you want. That's because the C<(?!foo)> is just saying that
723the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
724match. You would have to do something like C</(?!foo)...bar/> for that. We
725say "like" because there's the case of your "bar" not having three characters
726before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
727Sometimes it's still easier just to say:
a0d0e21e 728
a3cb178b 729 if (/bar/ && $` !~ /foo$/)
a0d0e21e 730
19799a22 731For look-behind see below.
c277df42 732
ee9b8eae
YO
733=item C<(?<=pattern)> C<\K>
734X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 735
c47ff5f1 736A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
737matches a word that follows a tab, without including the tab in C<$&>.
738Works only for fixed-width look-behind.
c277df42 739
ee9b8eae
YO
740There is a special form of this construct, called C<\K>, which causes the
741regex engine to "keep" everything it had matched prior to the C<\K> and
742not include it in C<$&>. This effectively provides variable length
743look-behind. The use of C<\K> inside of another look-around assertion
744is allowed, but the behaviour is currently not well defined.
745
746For various reasons C<\K> may be signifigantly more efficient than the
747equivalent C<< (?<=...) >> construct, and it is especially useful in
748situations where you want to efficiently remove something following
749something else in a string. For instance
750
751 s/(foo)bar/$1/g;
752
753can be rewritten as the much more efficient
754
755 s/foo\Kbar//g;
756
5a964f20 757=item C<(?<!pattern)>
d74e8afc 758X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 759
19799a22
GS
760A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
761matches any occurrence of "foo" that does not follow "bar". Works
762only for fixed-width look-behind.
c277df42 763
ee9b8eae
YO
764=back
765
81714fb9
YO
766=item C<(?'NAME'pattern)>
767
768=item C<< (?<NAME>pattern) >>
769X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
770
771A named capture buffer. Identical in every respect to normal capturing
772parens C<()> but for the additional fact that C<%+> may be used after
773a succesful match to refer to a named buffer. See C<perlvar> for more
774details on the C<%+> hash.
775
776If multiple distinct capture buffers have the same name then the
777$+{NAME} will refer to the leftmost defined buffer in the match.
778
779The forms C<(?'NAME'pattern)> and C<(?<NAME>pattern)> are equivalent.
780
781B<NOTE:> While the notation of this construct is the same as the similar
782function in .NET regexes, the behavior is not, in Perl the buffers are
783numbered sequentially regardless of being named or not. Thus in the
784pattern
785
786 /(x)(?<foo>y)(z)/
787
788$+{foo} will be the same as $2, and $3 will contain 'z' instead of
789the opposite which is what a .NET regex hacker might expect.
790
1f1031fe
YO
791Currently NAME is restricted to simple identifiers only.
792In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
793its Unicode extension (see L<utf8>),
794though it isn't extended by the locale (see L<perllocale>).
81714fb9 795
1f1031fe 796B<NOTE:> In order to make things easier for programmers with experience
ee9b8eae 797with the Python or PCRE regex engines the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe
YO
798maybe be used instead of C<< (?<NAME>pattern) >>; however this form does not
799support the use of single quotes as a delimiter for the name. This is
800only available in Perl 5.10 or later.
81714fb9 801
1f1031fe
YO
802=item C<< \k<NAME> >>
803
804=item C<< \k'NAME' >>
81714fb9
YO
805
806Named backreference. Similar to numeric backreferences, except that
807the group is designated by name and not number. If multiple groups
808have the same name then it refers to the leftmost defined group in
809the current match.
810
811It is an error to refer to a name not defined by a C<(?<NAME>)>
812earlier in the pattern.
813
814Both forms are equivalent.
815
1f1031fe
YO
816B<NOTE:> In order to make things easier for programmers with experience
817with the Python or PCRE regex engines the pattern C<< (?P=NAME) >>
818maybe be used instead of C<< \k<NAME> >> in Perl 5.10 or later.
819
cc6b7395 820=item C<(?{ code })>
d74e8afc 821X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 822
19799a22 823B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
824experimental, and may be changed without notice. Code executed that
825has side effects may not perform identically from version to version
826due to the effect of future optimisations in the regex engine.
c277df42 827
cc46d5f2 828This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
829always succeeds, and its C<code> is not interpolated. Currently,
830the rules to determine where the C<code> ends are somewhat convoluted.
831
77ea4f6d
JV
832This feature can be used together with the special variable C<$^N> to
833capture the results of submatches in variables without having to keep
834track of the number of nested parentheses. For example:
835
836 $_ = "The brown fox jumps over the lazy dog";
837 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
838 print "color = $color, animal = $animal\n";
839
754091cb
RGS
840Inside the C<(?{...})> block, C<$_> refers to the string the regular
841expression is matching against. You can also use C<pos()> to know what is
fa11829f 842the current position of matching within this string.
754091cb 843
19799a22
GS
844The C<code> is properly scoped in the following sense: If the assertion
845is backtracked (compare L<"Backtracking">), all changes introduced after
846C<local>ization are undone, so that
b9ac3b5b
GS
847
848 $_ = 'a' x 8;
5d458dd8 849 m<
b9ac3b5b
GS
850 (?{ $cnt = 0 }) # Initialize $cnt.
851 (
5d458dd8 852 a
b9ac3b5b
GS
853 (?{
854 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
855 })
5d458dd8 856 )*
b9ac3b5b
GS
857 aaaa
858 (?{ $res = $cnt }) # On success copy to non-localized
859 # location.
860 >x;
861
19799a22 862will set C<$res = 4>. Note that after the match, $cnt returns to the globally
14218588 863introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
864are unwound.
865
19799a22
GS
866This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
867switch. If I<not> used in this way, the result of evaluation of
868C<code> is put into the special variable C<$^R>. This happens
869immediately, so C<$^R> can be used from other C<(?{ code })> assertions
870inside the same regular expression.
b9ac3b5b 871
19799a22
GS
872The assignment to C<$^R> above is properly localized, so the old
873value of C<$^R> is restored if the assertion is backtracked; compare
874L<"Backtracking">.
b9ac3b5b 875
61528107
SP
876Due to an unfortunate implementation issue, the Perl code contained in these
877blocks is treated as a compile time closure that can have seemingly bizarre
6bda09f9 878consequences when used with lexically scoped variables inside of subroutines
61528107
SP
879or loops. There are various workarounds for this, including simply using
880global variables instead. If you are using this construct and strange results
6bda09f9
YO
881occur then check for the use of lexically scoped variables.
882
19799a22
GS
883For reasons of security, this construct is forbidden if the regular
884expression involves run-time interpolation of variables, unless the
885perilous C<use re 'eval'> pragma has been used (see L<re>), or the
886variables contain results of C<qr//> operator (see
5d458dd8 887L<perlop/"qr/STRING/imosx">).
871b0233 888
14218588 889This restriction is because of the wide-spread and remarkably convenient
19799a22 890custom of using run-time determined strings as patterns. For example:
871b0233
IZ
891
892 $re = <>;
893 chomp $re;
894 $string =~ /$re/;
895
14218588
GS
896Before Perl knew how to execute interpolated code within a pattern,
897this operation was completely safe from a security point of view,
898although it could raise an exception from an illegal pattern. If
899you turn on the C<use re 'eval'>, though, it is no longer secure,
900so you should only do so if you are also using taint checking.
901Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 902compartment. See L<perlsec> for details about both these mechanisms.
871b0233 903
5d458dd8 904Because perl's regex engine is not currently re-entrant, interpolated
8988a1bb
DD
905code may not invoke the regex engine either directly with C<m//> or C<s///>),
906or indirectly with functions such as C<split>.
907
14455d6c 908=item C<(??{ code })>
d74e8afc
ITB
909X<(??{})>
910X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 911
19799a22 912B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
913experimental, and may be changed without notice. Code executed that
914has side effects may not perform identically from version to version
915due to the effect of future optimisations in the regex engine.
0f5d15d6 916
19799a22
GS
917This is a "postponed" regular subexpression. The C<code> is evaluated
918at run time, at the moment this subexpression may match. The result
919of evaluation is considered as a regular expression and matched as
61528107 920if it were inserted instead of this construct. Note that this means
6bda09f9
YO
921that the contents of capture buffers defined inside an eval'ed pattern
922are not available outside of the pattern, and vice versa, there is no
923way for the inner pattern to refer to a capture buffer defined outside.
924Thus,
925
926 ('a' x 100)=~/(??{'(.)' x 100})/
927
81714fb9 928B<will> match, it will B<not> set $1.
0f5d15d6 929
428594d9 930The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
931where the C<code> ends are currently somewhat convoluted.
932
933The following pattern matches a parenthesized group:
0f5d15d6
IZ
934
935 $re = qr{
936 \(
937 (?:
938 (?> [^()]+ ) # Non-parens without backtracking
939 |
14455d6c 940 (??{ $re }) # Group with matching parens
0f5d15d6
IZ
941 )*
942 \)
943 }x;
944
6bda09f9
YO
945See also C<(?PARNO)> for a different, more efficient way to accomplish
946the same task.
947
5d458dd8 948Because perl's regex engine is not currently re-entrant, delayed
8988a1bb
DD
949code may not invoke the regex engine either directly with C<m//> or C<s///>),
950or indirectly with functions such as C<split>.
951
5d458dd8
YO
952Recursing deeper than 50 times without consuming any input string will
953result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
954changing it requires a custom build.
955
542fa716
YO
956=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
957X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 958X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 959X<regex, relative recursion>
6bda09f9 960
81714fb9
YO
961Similar to C<(??{ code })> except it does not involve compiling any code,
962instead it treats the contents of a capture buffer as an independent
61528107 963pattern that must match at the current position. Capture buffers
81714fb9 964contained by the pattern will have the value as determined by the
6bda09f9
YO
965outermost recursion.
966
894be9b7
YO
967PARNO is a sequence of digits (not starting with 0) whose value reflects
968the paren-number of the capture buffer to recurse to. C<(?R)> recurses to
969the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716
YO
970C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
971to be relative, with negative numbers indicating preceding capture buffers
972and positive ones following. Thus C<(?-1)> refers to the most recently
973declared buffer, and C<(?+1)> indicates the next buffer to be declared.
c74340f9
YO
974Note that the counting for relative recursion differs from that of
975relative backreferences, in that with recursion unclosed buffers B<are>
976included.
6bda09f9 977
81714fb9 978The following pattern matches a function foo() which may contain
f145b7e9 979balanced parentheses as the argument.
6bda09f9
YO
980
981 $re = qr{ ( # paren group 1 (full function)
81714fb9 982 foo
6bda09f9
YO
983 ( # paren group 2 (parens)
984 \(
985 ( # paren group 3 (contents of parens)
986 (?:
987 (?> [^()]+ ) # Non-parens without backtracking
988 |
989 (?2) # Recurse to start of paren group 2
990 )*
991 )
992 \)
993 )
994 )
995 }x;
996
997If the pattern was used as follows
998
999 'foo(bar(baz)+baz(bop))'=~/$re/
1000 and print "\$1 = $1\n",
1001 "\$2 = $2\n",
1002 "\$3 = $3\n";
1003
1004the output produced should be the following:
1005
1006 $1 = foo(bar(baz)+baz(bop))
1007 $2 = (bar(baz)+baz(bop))
81714fb9 1008 $3 = bar(baz)+baz(bop)
6bda09f9 1009
81714fb9 1010If there is no corresponding capture buffer defined, then it is a
61528107 1011fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1012string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1013into perl, so changing it requires a custom build.
1014
542fa716
YO
1015The following shows how using negative indexing can make it
1016easier to embed recursive patterns inside of a C<qr//> construct
1017for later use:
1018
1019 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1020 if (/foo $parens \s+ + \s+ bar $parens/x) {
1021 # do something here...
1022 }
1023
81714fb9 1024B<Note> that this pattern does not behave the same way as the equivalent
6bda09f9
YO
1025PCRE or Python construct of the same form. In perl you can backtrack into
1026a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1027as atomic. Also, modifiers are resolved at compile time, so constructs
1028like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1029be processed.
6bda09f9 1030
894be9b7
YO
1031=item C<(?&NAME)>
1032X<(?&NAME)>
1033
1034Recurse to a named subpattern. Identical to (?PARNO) except that the
1035parenthesis to recurse to is determined by name. If multiple parens have
1036the same name, then it recurses to the leftmost.
1037
1038It is an error to refer to a name that is not declared somewhere in the
1039pattern.
1040
1f1031fe
YO
1041B<NOTE:> In order to make things easier for programmers with experience
1042with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
1043maybe be used instead of C<< (?&NAME) >> as of Perl 5.10.
1044
e2e6a0f1
YO
1045=item C<(?(condition)yes-pattern|no-pattern)>
1046X<(?()>
286f584a 1047
e2e6a0f1 1048=item C<(?(condition)yes-pattern)>
286f584a 1049
e2e6a0f1
YO
1050Conditional expression. C<(condition)> should be either an integer in
1051parentheses (which is valid if the corresponding pair of parentheses
1052matched), a look-ahead/look-behind/evaluate zero-width assertion, a
1053name in angle brackets or single quotes (which is valid if a buffer
1054with the given name matched), or the special symbol (R) (true when
1055evaluated inside of recursion or eval). Additionally the R may be
1056followed by a number, (which will be true when evaluated when recursing
1057inside of the appropriate group), or by C<&NAME>, in which case it will
1058be true only when evaluated during recursion in the named group.
1059
1060Here's a summary of the possible predicates:
1061
1062=over 4
1063
1064=item (1) (2) ...
1065
1066Checks if the numbered capturing buffer has matched something.
1067
1068=item (<NAME>) ('NAME')
1069
1070Checks if a buffer with the given name has matched something.
1071
1072=item (?{ CODE })
1073
1074Treats the code block as the condition.
1075
1076=item (R)
1077
1078Checks if the expression has been evaluated inside of recursion.
1079
1080=item (R1) (R2) ...
1081
1082Checks if the expression has been evaluated while executing directly
1083inside of the n-th capture group. This check is the regex equivalent of
1084
1085 if ((caller(0))[3] eq 'subname') { ... }
1086
1087In other words, it does not check the full recursion stack.
1088
1089=item (R&NAME)
1090
1091Similar to C<(R1)>, this predicate checks to see if we're executing
1092directly inside of the leftmost group with a given name (this is the same
1093logic used by C<(?&NAME)> to disambiguate). It does not check the full
1094stack, but only the name of the innermost active recursion.
1095
1096=item (DEFINE)
1097
1098In this case, the yes-pattern is never directly executed, and no
1099no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1100See below for details.
1101
1102=back
1103
1104For example:
1105
1106 m{ ( \( )?
1107 [^()]+
1108 (?(1) \) )
1109 }x
1110
1111matches a chunk of non-parentheses, possibly included in parentheses
1112themselves.
1113
1114A special form is the C<(DEFINE)> predicate, which never executes directly
1115its yes-pattern, and does not allow a no-pattern. This allows to define
1116subpatterns which will be executed only by using the recursion mechanism.
1117This way, you can define a set of regular expression rules that can be
1118bundled into any pattern you choose.
1119
1120It is recommended that for this usage you put the DEFINE block at the
1121end of the pattern, and that you name any subpatterns defined within it.
1122
1123Also, it's worth noting that patterns defined this way probably will
1124not be as efficient, as the optimiser is not very clever about
1125handling them.
1126
1127An example of how this might be used is as follows:
1128
2bf803e2 1129 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1130 (?(DEFINE)
2bf803e2
YO
1131 (?<NAME_PAT>....)
1132 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1133 )/x
1134
1135Note that capture buffers matched inside of recursion are not accessible
1136after the recursion returns, so the extra layer of capturing buffers are
1137necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1138C<$+{NAME}> would be.
286f584a 1139
c47ff5f1 1140=item C<< (?>pattern) >>
6bda09f9 1141X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1142
19799a22
GS
1143An "independent" subexpression, one which matches the substring
1144that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1145position, and it matches I<nothing other than this substring>. This
19799a22
GS
1146construct is useful for optimizations of what would otherwise be
1147"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1148It may also be useful in places where the "grab all you can, and do not
1149give anything back" semantic is desirable.
19799a22 1150
c47ff5f1 1151For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1152(anchored at the beginning of string, as above) will match I<all>
1153characters C<a> at the beginning of string, leaving no C<a> for
1154C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1155since the match of the subgroup C<a*> is influenced by the following
1156group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1157C<a*ab> will match fewer characters than a standalone C<a*>, since
1158this makes the tail match.
1159
c47ff5f1 1160An effect similar to C<< (?>pattern) >> may be achieved by writing
19799a22
GS
1161C<(?=(pattern))\1>. This matches the same substring as a standalone
1162C<a+>, and the following C<\1> eats the matched string; it therefore
c47ff5f1 1163makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1164(The difference between these two constructs is that the second one
1165uses a capturing group, thus shifting ordinals of backreferences
1166in the rest of a regular expression.)
1167
1168Consider this pattern:
c277df42 1169
871b0233 1170 m{ \(
e2e6a0f1
YO
1171 (
1172 [^()]+ # x+
1173 |
871b0233
IZ
1174 \( [^()]* \)
1175 )+
e2e6a0f1 1176 \)
871b0233 1177 }x
5a964f20 1178
19799a22
GS
1179That will efficiently match a nonempty group with matching parentheses
1180two levels deep or less. However, if there is no such group, it
1181will take virtually forever on a long string. That's because there
1182are so many different ways to split a long string into several
1183substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1184to a subpattern of the above pattern. Consider how the pattern
1185above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1186seconds, but that each extra letter doubles this time. This
1187exponential performance will make it appear that your program has
14218588 1188hung. However, a tiny change to this pattern
5a964f20 1189
e2e6a0f1
YO
1190 m{ \(
1191 (
1192 (?> [^()]+ ) # change x+ above to (?> x+ )
1193 |
871b0233
IZ
1194 \( [^()]* \)
1195 )+
e2e6a0f1 1196 \)
871b0233 1197 }x
c277df42 1198
c47ff5f1 1199which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1200this yourself would be a productive exercise), but finishes in a fourth
1201the time when used on a similar string with 1000000 C<a>s. Be aware,
1202however, that this pattern currently triggers a warning message under
9f1b1f2d 1203the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1204C<"matches null string many times in regex">.
c277df42 1205
c47ff5f1 1206On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1207effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1208This was only 4 times slower on a string with 1000000 C<a>s.
1209
9da458fc
IZ
1210The "grab all you can, and do not give anything back" semantic is desirable
1211in many situations where on the first sight a simple C<()*> looks like
1212the correct solution. Suppose we parse text with comments being delimited
1213by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1214its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1215the comment delimiter, because it may "give up" some whitespace if
1216the remainder of the pattern can be made to match that way. The correct
1217answer is either one of these:
1218
1219 (?>#[ \t]*)
1220 #[ \t]*(?![ \t])
1221
1222For example, to grab non-empty comments into $1, one should use either
1223one of these:
1224
1225 / (?> \# [ \t]* ) ( .+ ) /x;
1226 / \# [ \t]* ( [^ \t] .* ) /x;
1227
1228Which one you pick depends on which of these expressions better reflects
1229the above specification of comments.
1230
6bda09f9
YO
1231In some literature this construct is called "atomic matching" or
1232"possessive matching".
1233
b9b4dddf
YO
1234Possessive quantifiers are equivalent to putting the item they are applied
1235to inside of one of these constructs. The following equivalences apply:
1236
1237 Quantifier Form Bracketing Form
1238 --------------- ---------------
1239 PAT*+ (?>PAT*)
1240 PAT++ (?>PAT+)
1241 PAT?+ (?>PAT?)
1242 PAT{min,max}+ (?>PAT{min,max})
1243
e2e6a0f1
YO
1244=back
1245
1246=head2 Special Backtracking Control Verbs
1247
1248B<WARNING:> These patterns are experimental and subject to change or
1249removal in a future version of perl. Their usage in production code should
1250be noted to avoid problems during upgrades.
1251
1252These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1253otherwise stated the ARG argument is optional; in some cases, it is
1254forbidden.
1255
1256Any pattern containing a special backtracking verb that allows an argument
1257has the special behaviour that when executed it sets the current packages'
5d458dd8
YO
1258C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1259rules apply:
e2e6a0f1 1260
5d458dd8
YO
1261On failure, the C<$REGERROR> variable will be set to the ARG value of the
1262verb pattern, if the verb was involved in the failure of the match. If the
1263ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1264name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1265none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1266
5d458dd8
YO
1267On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1268the C<$REGMARK> variable will be set to the name of the last
1269C<(*MARK:NAME)> pattern executed. See the explanation for the
1270C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1271
5d458dd8
YO
1272B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1273and most other regex related variables. They are not local to a scope, nor
1274readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1275Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1276
1277If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1278argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1
YO
1279
1280=over 4
1281
1282=item Verbs that take an argument
1283
1284=over 4
1285
5d458dd8 1286=item C<(*PRUNE)> C<(*PRUNE:NAME)>
ee9b8eae 1287X<(*PRUNE)> X<(*PRUNE:NAME)> X<\v>
54612592 1288
5d458dd8
YO
1289This zero-width pattern prunes the backtracking tree at the current point
1290when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1291where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1292A may backtrack as necessary to match. Once it is reached, matching
1293continues in B, which may also backtrack as necessary; however, should B
1294not match, then no further backtracking will take place, and the pattern
1295will fail outright at the current starting position.
54612592 1296
ee9b8eae
YO
1297As a shortcut, X<\v> is exactly equivalent to C<(*PRUNE)>.
1298
54612592
YO
1299The following example counts all the possible matching strings in a
1300pattern (without actually matching any of them).
1301
e2e6a0f1 1302 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1303 print "Count=$count\n";
1304
1305which produces:
1306
1307 aaab
1308 aaa
1309 aa
1310 a
1311 aab
1312 aa
1313 a
1314 ab
1315 a
1316 Count=9
1317
5d458dd8 1318If we add a C<(*PRUNE)> before the count like the following
54612592 1319
5d458dd8 1320 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1321 print "Count=$count\n";
1322
1323we prevent backtracking and find the count of the longest matching
1324at each matching startpoint like so:
1325
1326 aaab
1327 aab
1328 ab
1329 Count=3
1330
5d458dd8 1331Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1332
5d458dd8
YO
1333See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1334control backtracking. In some cases, the use of C<(*PRUNE)> can be
1335replaced with a C<< (?>pattern) >> with no functional difference; however,
1336C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1337C<< (?>pattern) >> alone.
54612592 1338
e2e6a0f1 1339
5d458dd8
YO
1340=item C<(*SKIP)> C<(*SKIP:NAME)>
1341X<(*SKIP)>
e2e6a0f1 1342
5d458dd8 1343This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1344failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1345to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1346of this pattern. This effectively means that the regex engine "skips" forward
1347to this position on failure and tries to match again, (assuming that
1348there is sufficient room to match).
1349
ee9b8eae
YO
1350As a shortcut X<\V> is exactly equivalent to C<(*SKIP)>.
1351
5d458dd8
YO
1352The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1353C<(*MARK:NAME)> was encountered while matching, then it is that position
1354which is used as the "skip point". If no C<(*MARK)> of that name was
1355encountered, then the C<(*SKIP)> operator has no effect. When used
1356without a name the "skip point" is where the match point was when
1357executing the (*SKIP) pattern.
1358
1359Compare the following to the examples in C<(*PRUNE)>, note the string
24b23f37
YO
1360is twice as long:
1361
5d458dd8 1362 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1363 print "Count=$count\n";
1364
1365outputs
1366
1367 aaab
1368 aaab
1369 Count=2
1370
5d458dd8 1371Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
e2e6a0f1 1372executed, the next startpoint will be where the cursor was when the
5d458dd8
YO
1373C<(*SKIP)> was executed.
1374
5d458dd8
YO
1375=item C<(*MARK:NAME)> C<(*:NAME)>
1376X<(*MARK)> C<(*MARK:NAME)> C<(*:NAME)>
1377
1378This zero-width pattern can be used to mark the point reached in a string
1379when a certain part of the pattern has been successfully matched. This
1380mark may be given a name. A later C<(*SKIP)> pattern will then skip
1381forward to that point if backtracked into on failure. Any number of
1382C<(*MARK)> patterns are allowed, and the NAME portion is optional and may
1383be duplicated.
1384
1385In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1386can be used to "label" a pattern branch, so that after matching, the
1387program can determine which branches of the pattern were involved in the
1388match.
1389
1390When a match is successful, the C<$REGMARK> variable will be set to the
1391name of the most recently executed C<(*MARK:NAME)> that was involved
1392in the match.
1393
1394This can be used to determine which branch of a pattern was matched
1395without using a seperate capture buffer for each branch, which in turn
1396can result in a performance improvement, as perl cannot optimize
1397C</(?:(x)|(y)|(z))/> as efficiently as something like
1398C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1399
1400When a match has failed, and unless another verb has been involved in
1401failing the match and has provided its own name to use, the C<$REGERROR>
1402variable will be set to the name of the most recently executed
1403C<(*MARK:NAME)>.
1404
1405See C<(*SKIP)> for more details.
1406
b62d2d15
YO
1407As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1408
5d458dd8
YO
1409=item C<(*THEN)> C<(*THEN:NAME)>
1410
1411This is similar to the "cut group" operator C<::> from Perl6. Like
1412C<(*PRUNE)>, this verb always matches, and when backtracked into on
1413failure, it causes the regex engine to try the next alternation in the
1414innermost enclosing group (capturing or otherwise).
1415
1416Its name comes from the observation that this operation combined with the
1417alternation operator (C<|>) can be used to create what is essentially a
1418pattern-based if/then/else block:
1419
1420 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1421
1422Note that if this operator is used and NOT inside of an alternation then
1423it acts exactly like the C<(*PRUNE)> operator.
1424
1425 / A (*PRUNE) B /
1426
1427is the same as
1428
1429 / A (*THEN) B /
1430
1431but
1432
1433 / ( A (*THEN) B | C (*THEN) D ) /
1434
1435is not the same as
1436
1437 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1438
1439as after matching the A but failing on the B the C<(*THEN)> verb will
1440backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1441
e2e6a0f1
YO
1442=item C<(*COMMIT)>
1443X<(*COMMIT)>
24b23f37 1444
5d458dd8
YO
1445This is the Perl6 "commit pattern" C<< <commit> >> or C<:::>. It's a
1446zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1447into on failure it causes the match to fail outright. No further attempts
1448to find a valid match by advancing the start pointer will occur again.
1449For example,
24b23f37 1450
e2e6a0f1 1451 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1452 print "Count=$count\n";
1453
1454outputs
1455
1456 aaab
1457 Count=1
1458
e2e6a0f1
YO
1459In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1460does not match, the regex engine will not try any further matching on the
1461rest of the string.
c277df42 1462
e2e6a0f1 1463=back
9af228c6 1464
e2e6a0f1 1465=item Verbs without an argument
9af228c6
YO
1466
1467=over 4
1468
e2e6a0f1
YO
1469=item C<(*FAIL)> C<(*F)>
1470X<(*FAIL)> X<(*F)>
9af228c6 1471
e2e6a0f1
YO
1472This pattern matches nothing and always fails. It can be used to force the
1473engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1474fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1475
e2e6a0f1 1476It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1477
e2e6a0f1
YO
1478=item C<(*ACCEPT)>
1479X<(*ACCEPT)>
9af228c6 1480
e2e6a0f1
YO
1481B<WARNING:> This feature is highly experimental. It is not recommended
1482for production code.
9af228c6 1483
e2e6a0f1
YO
1484This pattern matches nothing and causes the end of successful matching at
1485the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1486whether there is actually more to match in the string. When inside of a
1487nested pattern, such as recursion or a dynamically generated subbpattern
1488via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1489
e2e6a0f1
YO
1490If the C<(*ACCEPT)> is inside of capturing buffers then the buffers are
1491marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1492For instance:
9af228c6 1493
e2e6a0f1 1494 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1495
e2e6a0f1
YO
1496will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
1497be set. If another branch in the inner parens were matched, such as in the
1498string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1499
1500=back
c277df42 1501
a0d0e21e
LW
1502=back
1503
c07a80fd 1504=head2 Backtracking
d74e8afc 1505X<backtrack> X<backtracking>
c07a80fd 1506
35a734be
IZ
1507NOTE: This section presents an abstract approximation of regular
1508expression behavior. For a more rigorous (and complicated) view of
1509the rules involved in selecting a match among possible alternatives,
1510see L<Combining pieces together>.
1511
c277df42 1512A fundamental feature of regular expression matching involves the
5a964f20 1513notion called I<backtracking>, which is currently used (when needed)
c277df42 1514by all regular expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1515C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1516internally, but the general principle outlined here is valid.
c07a80fd
PP
1517
1518For a regular expression to match, the I<entire> regular expression must
1519match, not just part of it. So if the beginning of a pattern containing a
1520quantifier succeeds in a way that causes later parts in the pattern to
1521fail, the matching engine backs up and recalculates the beginning
1522part--that's why it's called backtracking.
1523
1524Here is an example of backtracking: Let's say you want to find the
1525word following "foo" in the string "Food is on the foo table.":
1526
1527 $_ = "Food is on the foo table.";
1528 if ( /\b(foo)\s+(\w+)/i ) {
1529 print "$2 follows $1.\n";
1530 }
1531
1532When the match runs, the first part of the regular expression (C<\b(foo)>)
1533finds a possible match right at the beginning of the string, and loads up
1534$1 with "Foo". However, as soon as the matching engine sees that there's
1535no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1536mistake and starts over again one character after where it had the
c07a80fd
PP
1537tentative match. This time it goes all the way until the next occurrence
1538of "foo". The complete regular expression matches this time, and you get
1539the expected output of "table follows foo."
1540
1541Sometimes minimal matching can help a lot. Imagine you'd like to match
1542everything between "foo" and "bar". Initially, you write something
1543like this:
1544
1545 $_ = "The food is under the bar in the barn.";
1546 if ( /foo(.*)bar/ ) {
1547 print "got <$1>\n";
1548 }
1549
1550Which perhaps unexpectedly yields:
1551
1552 got <d is under the bar in the >
1553
1554That's because C<.*> was greedy, so you get everything between the
14218588 1555I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1556to use minimal matching to make sure you get the text between a "foo"
1557and the first "bar" thereafter.
1558
1559 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1560 got <d is under the >
1561
1562Here's another example: let's say you'd like to match a number at the end
b6e13d97 1563of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
1564So you write this:
1565
1566 $_ = "I have 2 numbers: 53147";
1567 if ( /(.*)(\d*)/ ) { # Wrong!
1568 print "Beginning is <$1>, number is <$2>.\n";
1569 }
1570
1571That won't work at all, because C<.*> was greedy and gobbled up the
1572whole string. As C<\d*> can match on an empty string the complete
1573regular expression matched successfully.
1574
8e1088bc 1575 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
1576
1577Here are some variants, most of which don't work:
1578
1579 $_ = "I have 2 numbers: 53147";
1580 @pats = qw{
1581 (.*)(\d*)
1582 (.*)(\d+)
1583 (.*?)(\d*)
1584 (.*?)(\d+)
1585 (.*)(\d+)$
1586 (.*?)(\d+)$
1587 (.*)\b(\d+)$
1588 (.*\D)(\d+)$
1589 };
1590
1591 for $pat (@pats) {
1592 printf "%-12s ", $pat;
1593 if ( /$pat/ ) {
1594 print "<$1> <$2>\n";
1595 } else {
1596 print "FAIL\n";
1597 }
1598 }
1599
1600That will print out:
1601
1602 (.*)(\d*) <I have 2 numbers: 53147> <>
1603 (.*)(\d+) <I have 2 numbers: 5314> <7>
1604 (.*?)(\d*) <> <>
1605 (.*?)(\d+) <I have > <2>
1606 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1607 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1608 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1609 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1610
1611As you see, this can be a bit tricky. It's important to realize that a
1612regular expression is merely a set of assertions that gives a definition
1613of success. There may be 0, 1, or several different ways that the
1614definition might succeed against a particular string. And if there are
5a964f20
TC
1615multiple ways it might succeed, you need to understand backtracking to
1616know which variety of success you will achieve.
c07a80fd 1617
19799a22 1618When using look-ahead assertions and negations, this can all get even
8b19b778 1619trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
1620followed by "123". You might try to write that as
1621
871b0233
IZ
1622 $_ = "ABC123";
1623 if ( /^\D*(?!123)/ ) { # Wrong!
1624 print "Yup, no 123 in $_\n";
1625 }
c07a80fd
PP
1626
1627But that isn't going to match; at least, not the way you're hoping. It
1628claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1629why that pattern matches, contrary to popular expectations:
c07a80fd 1630
4358a253
SS
1631 $x = 'ABC123';
1632 $y = 'ABC445';
c07a80fd 1633
4358a253
SS
1634 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1635 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1636
4358a253
SS
1637 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1638 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
1639
1640This prints
1641
1642 2: got ABC
1643 3: got AB
1644 4: got ABC
1645
5f05dabc 1646You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
1647general purpose version of test 1. The important difference between
1648them is that test 3 contains a quantifier (C<\D*>) and so can use
1649backtracking, whereas test 1 will not. What's happening is
1650that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1651non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1652let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1653fail.
14218588 1654
c07a80fd 1655The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1656try to match C<(?!123> with "123", which fails. But because
c07a80fd
PP
1657a quantifier (C<\D*>) has been used in the regular expression, the
1658search engine can backtrack and retry the match differently
54310121 1659in the hope of matching the complete regular expression.
c07a80fd 1660
5a964f20
TC
1661The pattern really, I<really> wants to succeed, so it uses the
1662standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1663time. Now there's indeed something following "AB" that is not
14218588 1664"123". It's "C123", which suffices.
c07a80fd 1665
14218588
GS
1666We can deal with this by using both an assertion and a negation.
1667We'll say that the first part in $1 must be followed both by a digit
1668and by something that's not "123". Remember that the look-aheads
1669are zero-width expressions--they only look, but don't consume any
1670of the string in their match. So rewriting this way produces what
c07a80fd
PP
1671you'd expect; that is, case 5 will fail, but case 6 succeeds:
1672
4358a253
SS
1673 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1674 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
1675
1676 6: got ABC
1677
5a964f20 1678In other words, the two zero-width assertions next to each other work as though
19799a22 1679they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
1680matches only if you're at the beginning of the line AND the end of the
1681line simultaneously. The deeper underlying truth is that juxtaposition in
1682regular expressions always means AND, except when you write an explicit OR
1683using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1684although the attempted matches are made at different positions because "a"
1685is not a zero-width assertion, but a one-width assertion.
1686
19799a22 1687B<WARNING>: particularly complicated regular expressions can take
14218588 1688exponential time to solve because of the immense number of possible
9da458fc
IZ
1689ways they can use backtracking to try match. For example, without
1690internal optimizations done by the regular expression engine, this will
1691take a painfully long time to run:
c07a80fd 1692
e1901655
IZ
1693 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1694
1695And if you used C<*>'s in the internal groups instead of limiting them
1696to 0 through 5 matches, then it would take forever--or until you ran
1697out of stack space. Moreover, these internal optimizations are not
1698always applicable. For example, if you put C<{0,5}> instead of C<*>
1699on the external group, no current optimization is applicable, and the
1700match takes a long time to finish.
c07a80fd 1701
9da458fc
IZ
1702A powerful tool for optimizing such beasts is what is known as an
1703"independent group",
c47ff5f1 1704which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1705zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 1706the tail match, since they are in "logical" context: only
14218588 1707whether they match is considered relevant. For an example
9da458fc 1708where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1709following match, see L<C<< (?>pattern) >>>.
c277df42 1710
a0d0e21e 1711=head2 Version 8 Regular Expressions
d74e8afc 1712X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1713
5a964f20 1714In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1715routines, here are the pattern-matching rules not described above.
1716
54310121 1717Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1718with a special meaning described here or above. You can cause
5a964f20 1719characters that normally function as metacharacters to be interpreted
5f05dabc 1720literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
a0d0e21e
LW
1721character; "\\" matches a "\"). A series of characters matches that
1722series of characters in the target string, so the pattern C<blurfl>
1723would match "blurfl" in the target string.
1724
1725You can specify a character class, by enclosing a list of characters
5d458dd8 1726in C<[]>, which will match any character from the list. If the
a0d0e21e 1727first character after the "[" is "^", the class matches any character not
14218588 1728in the list. Within a list, the "-" character specifies a
5a964f20 1729range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1730inclusive. If you want either "-" or "]" itself to be a member of a
1731class, put it at the start of the list (possibly after a "^"), or
1732escape it with a backslash. "-" is also taken literally when it is
1733at the end of the list, just before the closing "]". (The
84850974
DD
1734following all specify the same class of three characters: C<[-az]>,
1735C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
1736specifies a class containing twenty-six characters, even on EBCDIC-based
1737character sets.) Also, if you try to use the character
1738classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1739a range, the "-" is understood literally.
a0d0e21e 1740
8ada0baa
JH
1741Note also that the whole range idea is rather unportable between
1742character sets--and even within character sets they may cause results
1743you probably didn't expect. A sound principle is to use only ranges
1744that begin from and end at either alphabets of equal case ([a-e],
1745[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1746spell out the character sets in full.
1747
54310121 1748Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1749used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1750"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
5d458dd8
YO
1751of octal digits, matches the character whose coded character set value
1752is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
1753matches the character whose numeric value is I<nn>. The expression \cI<x>
1754matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 1755matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1756
1757You can specify a series of alternatives for a pattern using "|" to
1758separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1759or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1760first alternative includes everything from the last pattern delimiter
1761("(", "[", or the beginning of the pattern) up to the first "|", and
1762the last alternative contains everything from the last "|" to the next
14218588
GS
1763pattern delimiter. That's why it's common practice to include
1764alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1765start and end.
1766
5a964f20 1767Alternatives are tried from left to right, so the first
a3cb178b
GS
1768alternative found for which the entire expression matches, is the one that
1769is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1770example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1771part will match, as that is the first alternative tried, and it successfully
1772matches the target string. (This might not seem important, but it is
1773important when you are capturing matched text using parentheses.)
1774
5a964f20 1775Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1776so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1777
14218588
GS
1778Within a pattern, you may designate subpatterns for later reference
1779by enclosing them in parentheses, and you may refer back to the
1780I<n>th subpattern later in the pattern using the metacharacter
1781\I<n>. Subpatterns are numbered based on the left to right order
1782of their opening parenthesis. A backreference matches whatever
1783actually matched the subpattern in the string being examined, not
1784the rules for that subpattern. Therefore, C<(0|0x)\d*\s\1\d*> will
1785match "0x1234 0x4321", but not "0x1234 01234", because subpattern
17861 matched "0x", even though the rule C<0|0x> could potentially match
1787the leading 0 in the second number.
cb1a09d0 1788
19799a22 1789=head2 Warning on \1 vs $1
cb1a09d0 1790
5a964f20 1791Some people get too used to writing things like:
cb1a09d0
AD
1792
1793 $pattern =~ s/(\W)/\\\1/g;
1794
1795This is grandfathered for the RHS of a substitute to avoid shocking the
1796B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1797PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1798the usual double-quoted string means a control-A. The customary Unix
1799meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1800of doing that, you get yourself into trouble if you then add an C</e>
1801modifier.
1802
5a964f20 1803 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1804
1805Or if you try to do
1806
1807 s/(\d+)/\1000/;
1808
1809You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1810C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1811with the operation of matching a backreference. Certainly they mean two
1812different things on the I<left> side of the C<s///>.
9fa51da4 1813
c84d73f1
IZ
1814=head2 Repeated patterns matching zero-length substring
1815
19799a22 1816B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1817
1818Regular expressions provide a terse and powerful programming language. As
1819with most other power tools, power comes together with the ability
1820to wreak havoc.
1821
1822A common abuse of this power stems from the ability to make infinite
628afcb5 1823loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1824
1825 'foo' =~ m{ ( o? )* }x;
1826
1827The C<o?> can match at the beginning of C<'foo'>, and since the position
1828in the string is not moved by the match, C<o?> would match again and again
14218588 1829because of the C<*> modifier. Another common way to create a similar cycle
c84d73f1
IZ
1830is with the looping modifier C<//g>:
1831
1832 @matches = ( 'foo' =~ m{ o? }xg );
1833
1834or
1835
1836 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1837
1838or the loop implied by split().
1839
1840However, long experience has shown that many programming tasks may
14218588
GS
1841be significantly simplified by using repeated subexpressions that
1842may match zero-length substrings. Here's a simple example being:
c84d73f1
IZ
1843
1844 @chars = split //, $string; # // is not magic in split
1845 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1846
9da458fc 1847Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1
IZ
1848the infinite loop>. The rules for this are different for lower-level
1849loops given by the greedy modifiers C<*+{}>, and for higher-level
1850ones like the C</g> modifier or split() operator.
1851
19799a22
GS
1852The lower-level loops are I<interrupted> (that is, the loop is
1853broken) when Perl detects that a repeated expression matched a
1854zero-length substring. Thus
c84d73f1
IZ
1855
1856 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1857
5d458dd8 1858is made equivalent to
c84d73f1 1859
5d458dd8
YO
1860 m{ (?: NON_ZERO_LENGTH )*
1861 |
1862 (?: ZERO_LENGTH )?
c84d73f1
IZ
1863 }x;
1864
1865The higher level-loops preserve an additional state between iterations:
5d458dd8 1866whether the last match was zero-length. To break the loop, the following
c84d73f1 1867match after a zero-length match is prohibited to have a length of zero.
5d458dd8 1868This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
1869and so the I<second best> match is chosen if the I<best> match is of
1870zero length.
1871
19799a22 1872For example:
c84d73f1
IZ
1873
1874 $_ = 'bar';
1875 s/\w??/<$&>/g;
1876
20fb949f 1877results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 1878match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
1879best> match is what is matched by C<\w>. Thus zero-length matches
1880alternate with one-character-long matches.
1881
5d458dd8 1882Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
1883position one notch further in the string.
1884
19799a22 1885The additional state of being I<matched with zero-length> is associated with
c84d73f1 1886the matched string, and is reset by each assignment to pos().
9da458fc
IZ
1887Zero-length matches at the end of the previous match are ignored
1888during C<split>.
c84d73f1 1889
35a734be
IZ
1890=head2 Combining pieces together
1891
1892Each of the elementary pieces of regular expressions which were described
1893before (such as C<ab> or C<\Z>) could match at most one substring
1894at the given position of the input string. However, in a typical regular
1895expression these elementary pieces are combined into more complicated
1896patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1897(in these examples C<S> and C<T> are regular subexpressions).
1898
1899Such combinations can include alternatives, leading to a problem of choice:
1900if we match a regular expression C<a|ab> against C<"abc">, will it match
1901substring C<"a"> or C<"ab">? One way to describe which substring is
1902actually matched is the concept of backtracking (see L<"Backtracking">).
1903However, this description is too low-level and makes you think
1904in terms of a particular implementation.
1905
1906Another description starts with notions of "better"/"worse". All the
1907substrings which may be matched by the given regular expression can be
1908sorted from the "best" match to the "worst" match, and it is the "best"
1909match which is chosen. This substitutes the question of "what is chosen?"
1910by the question of "which matches are better, and which are worse?".
1911
1912Again, for elementary pieces there is no such question, since at most
1913one match at a given position is possible. This section describes the
1914notion of better/worse for combining operators. In the description
1915below C<S> and C<T> are regular subexpressions.
1916
13a2d996 1917=over 4
35a734be
IZ
1918
1919=item C<ST>
1920
1921Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
1922substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 1923which can be matched by C<T>.
35a734be
IZ
1924
1925If C<A> is better match for C<S> than C<A'>, C<AB> is a better
1926match than C<A'B'>.
1927
1928If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
1929C<B> is better match for C<T> than C<B'>.
1930
1931=item C<S|T>
1932
1933When C<S> can match, it is a better match than when only C<T> can match.
1934
1935Ordering of two matches for C<S> is the same as for C<S>. Similar for
1936two matches for C<T>.
1937
1938=item C<S{REPEAT_COUNT}>
1939
1940Matches as C<SSS...S> (repeated as many times as necessary).
1941
1942=item C<S{min,max}>
1943
1944Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
1945
1946=item C<S{min,max}?>
1947
1948Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
1949
1950=item C<S?>, C<S*>, C<S+>
1951
1952Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
1953
1954=item C<S??>, C<S*?>, C<S+?>
1955
1956Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
1957
c47ff5f1 1958=item C<< (?>S) >>
35a734be
IZ
1959
1960Matches the best match for C<S> and only that.
1961
1962=item C<(?=S)>, C<(?<=S)>
1963
1964Only the best match for C<S> is considered. (This is important only if
1965C<S> has capturing parentheses, and backreferences are used somewhere
1966else in the whole regular expression.)
1967
1968=item C<(?!S)>, C<(?<!S)>
1969
1970For this grouping operator there is no need to describe the ordering, since
1971only whether or not C<S> can match is important.
1972
6bda09f9 1973=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
1974
1975The ordering is the same as for the regular expression which is
6bda09f9 1976the result of EXPR, or the pattern contained by capture buffer PARNO.
35a734be
IZ
1977
1978=item C<(?(condition)yes-pattern|no-pattern)>
1979
1980Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
1981already determined. The ordering of the matches is the same as for the
1982chosen subexpression.
1983
1984=back
1985
1986The above recipes describe the ordering of matches I<at a given position>.
1987One more rule is needed to understand how a match is determined for the
1988whole regular expression: a match at an earlier position is always better
1989than a match at a later position.
1990
c84d73f1
IZ
1991=head2 Creating custom RE engines
1992
1993Overloaded constants (see L<overload>) provide a simple way to extend
1994the functionality of the RE engine.
1995
1996Suppose that we want to enable a new RE escape-sequence C<\Y|> which
6b0ac556 1997matches at boundary between whitespace characters and non-whitespace
c84d73f1
IZ
1998characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
1999at these positions, so we want to have each C<\Y|> in the place of the
2000more complicated version. We can create a module C<customre> to do
2001this:
2002
2003 package customre;
2004 use overload;
2005
2006 sub import {
2007 shift;
2008 die "No argument to customre::import allowed" if @_;
2009 overload::constant 'qr' => \&convert;
2010 }
2011
2012 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2013
580a9fe1
RGS
2014 # We must also take care of not escaping the legitimate \\Y|
2015 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2016 my %rules = ( '\\' => '\\\\',
c84d73f1
IZ
2017 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
2018 sub convert {
2019 my $re = shift;
5d458dd8 2020 $re =~ s{
c84d73f1
IZ
2021 \\ ( \\ | Y . )
2022 }
5d458dd8 2023 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2024 return $re;
2025 }
2026
2027Now C<use customre> enables the new escape in constant regular
2028expressions, i.e., those without any runtime variable interpolations.
2029As documented in L<overload>, this conversion will work only over
2030literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2031part of this regular expression needs to be converted explicitly
2032(but only if the special meaning of C<\Y|> should be enabled inside $re):
2033
2034 use customre;
2035 $re = <>;
2036 chomp $re;
2037 $re = customre::convert $re;
2038 /\Y|$re\Y|/;
2039
1f1031fe
YO
2040=head1 PCRE/Python Support
2041
2042As of Perl 5.10 Perl supports several Python/PCRE specific extensions
2043to the regex syntax. While Perl programmers are encouraged to use the
2044Perl specific syntax, the following are legal in Perl 5.10:
2045
2046=over 4
2047
ee9b8eae 2048=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe
YO
2049
2050Define a named capture buffer. Equivalent to C<< (?<NAME>pattern) >>.
2051
2052=item C<< (?P=NAME) >>
2053
2054Backreference to a named capture buffer. Equivalent to C<< \g{NAME} >>.
2055
2056=item C<< (?P>NAME) >>
2057
2058Subroutine call to a named capture buffer. Equivalent to C<< (?&NAME) >>.
2059
ee9b8eae 2060=back
1f1031fe 2061
19799a22
GS
2062=head1 BUGS
2063
9da458fc
IZ
2064This document varies from difficult to understand to completely
2065and utterly opaque. The wandering prose riddled with jargon is
2066hard to fathom in several places.
2067
2068This document needs a rewrite that separates the tutorial content
2069from the reference content.
19799a22
GS
2070
2071=head1 SEE ALSO
9fa51da4 2072
91e0c79e
MJD
2073L<perlrequick>.
2074
2075L<perlretut>.
2076
9b599b2a
GS
2077L<perlop/"Regexp Quote-Like Operators">.
2078
1e66bd83
PP
2079L<perlop/"Gory details of parsing quoted constructs">.
2080
14218588
GS
2081L<perlfaq6>.
2082
9b599b2a
GS
2083L<perlfunc/pos>.
2084
2085L<perllocale>.
2086
fb55449c
JH
2087L<perlebcdic>.
2088
14218588
GS
2089I<Mastering Regular Expressions> by Jeffrey Friedl, published
2090by O'Reilly and Associates.