This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Allow negative indexing in recursive patterns
[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
YO
248 \1 Backreference to a specific group.
249 '1' may actually be any positive integer.
81714fb9 250 \k<name> Named backreference
5d458dd8 251 \N{name} Named unicode character, or unicode escape
81714fb9
YO
252 \x12 Hexadecimal escape sequence
253 \x{1234} Long hexadecimal escape sequence
a0d0e21e 254
08ce8fc6
JH
255A C<\w> matches a single alphanumeric character (an alphabetic
256character, or a decimal digit) or C<_>, not a whole word. Use C<\w+>
257to match a string of Perl-identifier characters (which isn't the same
258as matching an English word). If C<use locale> is in effect, the list
259of alphabetic characters generated by C<\w> is taken from the current
260locale. See L<perllocale>. You may use C<\w>, C<\W>, C<\s>, C<\S>,
1209ba90 261C<\d>, and C<\D> within character classes, but if you try to use them
08ce8fc6
JH
262as endpoints of a range, that's not a range, the "-" is understood
263literally. If Unicode is in effect, C<\s> matches also "\x{85}",
264"\x{2028}, and "\x{2029}", see L<perlunicode> for more details about
491fd90a 265C<\pP>, C<\PP>, and C<\X>, and L<perluniintro> about Unicode in general.
fa11829f 266You can define your own C<\p> and C<\P> properties, see L<perlunicode>.
d74e8afc 267X<\w> X<\W> X<word>
a0d0e21e 268
b8c5462f 269The POSIX character class syntax
d74e8afc 270X<character class>
b8c5462f 271
820475bd 272 [:class:]
b8c5462f 273
5496314a
SP
274is also available. Note that the C<[> and C<]> braces are I<literal>;
275they must always be used within a character class expression.
276
277 # this is correct:
278 $string =~ /[[:alpha:]]/;
279
280 # this is not, and will generate a warning:
281 $string =~ /[:alpha:]/;
282
283The available classes and their backslash equivalents (if available) are
284as follows:
d74e8afc
ITB
285X<character class>
286X<alpha> X<alnum> X<ascii> X<blank> X<cntrl> X<digit> X<graph>
287X<lower> X<print> X<punct> X<space> X<upper> X<word> X<xdigit>
b8c5462f
JH
288
289 alpha
290 alnum
291 ascii
aaa51d5e 292 blank [1]
b8c5462f
JH
293 cntrl
294 digit \d
295 graph
296 lower
297 print
298 punct
aaa51d5e 299 space \s [2]
b8c5462f 300 upper
aaa51d5e 301 word \w [3]
b8c5462f
JH
302 xdigit
303
07698885
RGS
304=over
305
306=item [1]
307
b432a672 308A GNU extension equivalent to C<[ \t]>, "all horizontal whitespace".
07698885
RGS
309
310=item [2]
311
312Not exactly equivalent to C<\s> since the C<[[:space:]]> includes
b432a672 313also the (very rare) "vertical tabulator", "\ck", chr(11).
07698885
RGS
314
315=item [3]
316
08ce8fc6 317A Perl extension, see above.
07698885
RGS
318
319=back
aaa51d5e 320
26b44a0a 321For example use C<[:upper:]> to match all the uppercase characters.
aaa51d5e
JF
322Note that the C<[]> are part of the C<[::]> construct, not part of the
323whole character class. For example:
b8c5462f 324
820475bd 325 [01[:alpha:]%]
b8c5462f 326
593df60c 327matches zero, one, any alphabetic character, and the percentage sign.
b8c5462f 328
72ff2908
JH
329The following equivalences to Unicode \p{} constructs and equivalent
330backslash character classes (if available), will hold:
d74e8afc 331X<character class> X<\p> X<\p{}>
72ff2908 332
5496314a 333 [[:...:]] \p{...} backslash
b8c5462f
JH
334
335 alpha IsAlpha
336 alnum IsAlnum
337 ascii IsASCII
b432a672 338 blank IsSpace
b8c5462f 339 cntrl IsCntrl
3bec3564 340 digit IsDigit \d
b8c5462f
JH
341 graph IsGraph
342 lower IsLower
343 print IsPrint
344 punct IsPunct
345 space IsSpace
3bec3564 346 IsSpacePerl \s
b8c5462f
JH
347 upper IsUpper
348 word IsWord
349 xdigit IsXDigit
350
5496314a 351For example C<[[:lower:]]> and C<\p{IsLower}> are equivalent.
b8c5462f
JH
352
353If the C<utf8> pragma is not used but the C<locale> pragma is, the
aaa51d5e 354classes correlate with the usual isalpha(3) interface (except for
b432a672 355"word" and "blank").
b8c5462f
JH
356
357The assumedly non-obviously named classes are:
358
359=over 4
360
361=item cntrl
d74e8afc 362X<cntrl>
b8c5462f 363
820475bd
GS
364Any control character. Usually characters that don't produce output as
365such but instead control the terminal somehow: for example newline and
366backspace are control characters. All characters with ord() less than
593df60c 36732 are most often classified as control characters (assuming ASCII,
7be5a6cf
JF
368the ISO Latin character sets, and Unicode), as is the character with
369the ord() value of 127 (C<DEL>).
b8c5462f
JH
370
371=item graph
d74e8afc 372X<graph>
b8c5462f 373
f1cbbd6e 374Any alphanumeric or punctuation (special) character.
b8c5462f
JH
375
376=item print
d74e8afc 377X<print>
b8c5462f 378
f79b3095 379Any alphanumeric or punctuation (special) character or the space character.
b8c5462f
JH
380
381=item punct
d74e8afc 382X<punct>
b8c5462f 383
f1cbbd6e 384Any punctuation (special) character.
b8c5462f
JH
385
386=item xdigit
d74e8afc 387X<xdigit>
b8c5462f 388
593df60c 389Any hexadecimal digit. Though this may feel silly ([0-9A-Fa-f] would
820475bd 390work just fine) it is included for completeness.
b8c5462f 391
b8c5462f
JH
392=back
393
394You can negate the [::] character classes by prefixing the class name
395with a '^'. This is a Perl extension. For example:
d74e8afc 396X<character class, negation>
b8c5462f 397
5496314a 398 POSIX traditional Unicode
93733859 399
5496314a
SP
400 [[:^digit:]] \D \P{IsDigit}
401 [[:^space:]] \S \P{IsSpace}
402 [[:^word:]] \W \P{IsWord}
b8c5462f 403
54c18d04
MK
404Perl respects the POSIX standard in that POSIX character classes are
405only supported within a character class. The POSIX character classes
406[.cc.] and [=cc=] are recognized but B<not> supported and trying to
407use them will cause an error.
b8c5462f 408
04838cea
RGS
409=head3 Assertions
410
a0d0e21e 411Perl defines the following zero-width assertions:
d74e8afc
ITB
412X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
413X<regexp, zero-width assertion>
414X<regular expression, zero-width assertion>
415X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e
LW
416
417 \b Match a word boundary
418 \B Match a non-(word boundary)
b85d18e9
IZ
419 \A Match only at beginning of string
420 \Z Match only at end of string, or before newline at the end
421 \z Match only at end of string
9da458fc
IZ
422 \G Match only at pos() (e.g. at the end-of-match position
423 of prior m//g)
a0d0e21e 424
14218588 425A word boundary (C<\b>) is a spot between two characters
19799a22
GS
426that has a C<\w> on one side of it and a C<\W> on the other side
427of it (in either order), counting the imaginary characters off the
428beginning and end of the string as matching a C<\W>. (Within
429character classes C<\b> represents backspace rather than a word
430boundary, just as it normally does in any double-quoted string.)
431The C<\A> and C<\Z> are just like "^" and "$", except that they
432won't match multiple times when the C</m> modifier is used, while
433"^" and "$" will match at every internal line boundary. To match
434the actual end of the string and not ignore an optional trailing
435newline, use C<\z>.
d74e8afc 436X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
437
438The C<\G> assertion can be used to chain global matches (using
439C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
440It is also useful when writing C<lex>-like scanners, when you have
441several patterns that you want to match against consequent substrings
442of your string, see the previous reference. The actual location
443where C<\G> will match can also be influenced by using C<pos()> as
25cf8c22
HS
444an lvalue: see L<perlfunc/pos>. Currently C<\G> is only fully
445supported when anchored to the start of the pattern; while it
446is permitted to use it elsewhere, as in C</(?<=\G..)./g>, some
447such uses (C</.\G/g>, for example) currently cause problems, and
448it is recommended that you avoid such usage for now.
d74e8afc 449X<\G>
c47ff5f1 450
04838cea
RGS
451=head3 Capture buffers
452
14218588 453The bracketing construct C<( ... )> creates capture buffers. To
c47ff5f1 454refer to the digit'th buffer use \<digit> within the
14218588 455match. Outside the match use "$" instead of "\". (The
81714fb9 456\<digit> notation works in certain circumstances outside
14218588
GS
457the match. See the warning below about \1 vs $1 for details.)
458Referring back to another part of the match is called a
459I<backreference>.
d74e8afc
ITB
460X<regex, capture buffer> X<regexp, capture buffer>
461X<regular expression, capture buffer> X<backreference>
14218588
GS
462
463There is no limit to the number of captured substrings that you may
464use. However Perl also uses \10, \11, etc. as aliases for \010,
fb55449c
JH
465\011, etc. (Recall that 0 means octal, so \011 is the character at
466number 9 in your coded character set; which would be the 10th character,
81714fb9
YO
467a horizontal tab under ASCII.) Perl resolves this
468ambiguity by interpreting \10 as a backreference only if at least 10
469left parentheses have opened before it. Likewise \11 is a
470backreference only if at least 11 left parentheses have opened
471before it. And so on. \1 through \9 are always interpreted as
fb55449c 472backreferences.
14218588 473
81714fb9
YO
474Additionally, as of Perl 5.10 you may use named capture buffers and named
475backreferences. The notation is C<< (?<name>...) >> and C<< \k<name> >>
476(you may also use single quotes instead of angle brackets to quote the
477name). The only difference with named capture buffers and unnamed ones is
478that multiple buffers may have the same name and that the contents of
479named capture buffers is available via the C<%+> hash. When multiple
480groups share the same name C<$+{name}> and C<< \k<name> >> refer to the
481leftmost defined group, thus it's possible to do things with named capture
482buffers that would otherwise require C<(??{})> code to accomplish. Named
483capture buffers are numbered just as normal capture buffers are and may be
484referenced via the magic numeric variables or via numeric backreferences
485as well as by name.
486
14218588 487Examples:
a0d0e21e
LW
488
489 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
490
81714fb9
YO
491 /(.)\1/ # find first doubled char
492 and print "'$1' is the first doubled character\n";
493
494 /(?<char>.)\k<char>/ # ... a different way
495 and print "'$+{char}' is the first doubled character\n";
496
497 /(?<char>.)\1/ # ... mix and match
498 and print "'$1' is the first doubled character\n";
c47ff5f1 499
14218588 500 if (/Time: (..):(..):(..)/) { # parse out values
a0d0e21e
LW
501 $hours = $1;
502 $minutes = $2;
503 $seconds = $3;
504 }
c47ff5f1 505
14218588
GS
506Several special variables also refer back to portions of the previous
507match. C<$+> returns whatever the last bracket match matched.
508C<$&> returns the entire matched string. (At one point C<$0> did
509also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
510everything before the matched string. C<$'> returns everything
511after the matched string. And C<$^N> contains whatever was matched by
512the most-recently closed group (submatch). C<$^N> can be used in
513extended patterns (see below), for example to assign a submatch to a
81714fb9 514variable.
d74e8afc 515X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 516
665e98b9 517The numbered match variables ($1, $2, $3, etc.) and the related punctuation
77ea4f6d 518set (C<$+>, C<$&>, C<$`>, C<$'>, and C<$^N>) are all dynamically scoped
14218588
GS
519until the end of the enclosing block or until the next successful
520match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
521X<$+> X<$^N> X<$&> X<$`> X<$'>
522X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
523
14218588 524
665e98b9 525B<NOTE>: failed matches in Perl do not reset the match variables,
5146ce24 526which makes it easier to write code that tests for a series of more
665e98b9
JH
527specific cases and remembers the best match.
528
14218588
GS
529B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
530C<$'> anywhere in the program, it has to provide them for every
531pattern match. This may substantially slow your program. Perl
532uses the same mechanism to produce $1, $2, etc, so you also pay a
533price for each pattern that contains capturing parentheses. (To
534avoid this cost while retaining the grouping behaviour, use the
535extended regular expression C<(?: ... )> instead.) But if you never
536use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
537parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
538if you can, but if you can't (and some algorithms really appreciate
539them), once you've used them once, use them at will, because you've
540already paid the price. As of 5.005, C<$&> is not so costly as the
541other two.
d74e8afc 542X<$&> X<$`> X<$'>
68dc0745 543
19799a22
GS
544Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
545C<\w>, C<\n>. Unlike some other regular expression languages, there
546are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 547that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
548interpreted as a literal character, not a metacharacter. This was
549once used in a common idiom to disable or quote the special meanings
550of regular expression metacharacters in a string that you want to
36bbe248 551use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
552
553 $pattern =~ s/(\W)/\\$1/g;
554
f1cbbd6e 555(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
556Today it is more common to use the quotemeta() function or the C<\Q>
557metaquoting escape sequence to disable all metacharacters' special
558meanings like this:
a0d0e21e
LW
559
560 /$unquoted\Q$quoted\E$unquoted/
561
9da458fc
IZ
562Beware that if you put literal backslashes (those not inside
563interpolated variables) between C<\Q> and C<\E>, double-quotish
564backslash interpolation may lead to confusing results. If you
565I<need> to use literal backslashes within C<\Q...\E>,
566consult L<perlop/"Gory details of parsing quoted constructs">.
567
19799a22
GS
568=head2 Extended Patterns
569
14218588
GS
570Perl also defines a consistent extension syntax for features not
571found in standard tools like B<awk> and B<lex>. The syntax is a
572pair of parentheses with a question mark as the first thing within
573the parentheses. The character after the question mark indicates
574the extension.
19799a22 575
14218588
GS
576The stability of these extensions varies widely. Some have been
577part of the core language for many years. Others are experimental
578and may change without warning or be completely removed. Check
579the documentation on an individual feature to verify its current
580status.
19799a22 581
14218588
GS
582A question mark was chosen for this and for the minimal-matching
583construct because 1) question marks are rare in older regular
584expressions, and 2) whenever you see one, you should stop and
585"question" exactly what is going on. That's psychology...
a0d0e21e
LW
586
587=over 10
588
cc6b7395 589=item C<(?#text)>
d74e8afc 590X<(?#)>
a0d0e21e 591
14218588 592A comment. The text is ignored. If the C</x> modifier enables
19799a22 593whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
594the comment as soon as it sees a C<)>, so there is no way to put a literal
595C<)> in the comment.
a0d0e21e 596
19799a22 597=item C<(?imsx-imsx)>
d74e8afc 598X<(?)>
19799a22 599
0b6d1084
JH
600One or more embedded pattern-match modifiers, to be turned on (or
601turned off, if preceded by C<->) for the remainder of the pattern or
602the remainder of the enclosing pattern group (if any). This is
603particularly useful for dynamic patterns, such as those read in from a
604configuration file, read in as an argument, are specified in a table
605somewhere, etc. Consider the case that some of which want to be case
606sensitive and some do not. The case insensitive ones need to include
607merely C<(?i)> at the front of the pattern. For example:
19799a22
GS
608
609 $pattern = "foobar";
5d458dd8 610 if ( /$pattern/i ) { }
19799a22
GS
611
612 # more flexible:
613
614 $pattern = "(?i)foobar";
5d458dd8 615 if ( /$pattern/ ) { }
19799a22 616
0b6d1084 617These modifiers are restored at the end of the enclosing group. For example,
19799a22
GS
618
619 ( (?i) blah ) \s+ \1
620
621will match a repeated (I<including the case>!) word C<blah> in any
14218588 622case, assuming C<x> modifier, and no C<i> modifier outside this
19799a22
GS
623group.
624
5a964f20 625=item C<(?:pattern)>
d74e8afc 626X<(?:)>
a0d0e21e 627
ca9dfc88
IZ
628=item C<(?imsx-imsx:pattern)>
629
5a964f20
TC
630This is for clustering, not capturing; it groups subexpressions like
631"()", but doesn't make backreferences as "()" does. So
a0d0e21e 632
5a964f20 633 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
634
635is like
636
5a964f20 637 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 638
19799a22
GS
639but doesn't spit out extra fields. It's also cheaper not to capture
640characters if you don't need to.
a0d0e21e 641
19799a22 642Any letters between C<?> and C<:> act as flags modifiers as with
5d458dd8 643C<(?imsx-imsx)>. For example,
ca9dfc88
IZ
644
645 /(?s-i:more.*than).*million/i
646
14218588 647is equivalent to the more verbose
ca9dfc88
IZ
648
649 /(?:(?s-i)more.*than).*million/i
650
5a964f20 651=item C<(?=pattern)>
d74e8afc 652X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 653
19799a22 654A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
655matches a word followed by a tab, without including the tab in C<$&>.
656
5a964f20 657=item C<(?!pattern)>
d74e8afc 658X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 659
19799a22 660A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 661matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
662however that look-ahead and look-behind are NOT the same thing. You cannot
663use this for look-behind.
7b8d334a 664
5a964f20 665If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
666will not do what you want. That's because the C<(?!foo)> is just saying that
667the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
668match. You would have to do something like C</(?!foo)...bar/> for that. We
669say "like" because there's the case of your "bar" not having three characters
670before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
671Sometimes it's still easier just to say:
a0d0e21e 672
a3cb178b 673 if (/bar/ && $` !~ /foo$/)
a0d0e21e 674
19799a22 675For look-behind see below.
c277df42 676
c47ff5f1 677=item C<(?<=pattern)>
d74e8afc 678X<(?<=)> X<look-behind, positive> X<lookbehind, positive>
c277df42 679
c47ff5f1 680A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
681matches a word that follows a tab, without including the tab in C<$&>.
682Works only for fixed-width look-behind.
c277df42 683
5a964f20 684=item C<(?<!pattern)>
d74e8afc 685X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 686
19799a22
GS
687A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
688matches any occurrence of "foo" that does not follow "bar". Works
689only for fixed-width look-behind.
c277df42 690
81714fb9
YO
691=item C<(?'NAME'pattern)>
692
693=item C<< (?<NAME>pattern) >>
694X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
695
696A named capture buffer. Identical in every respect to normal capturing
697parens C<()> but for the additional fact that C<%+> may be used after
698a succesful match to refer to a named buffer. See C<perlvar> for more
699details on the C<%+> hash.
700
701If multiple distinct capture buffers have the same name then the
702$+{NAME} will refer to the leftmost defined buffer in the match.
703
704The forms C<(?'NAME'pattern)> and C<(?<NAME>pattern)> are equivalent.
705
706B<NOTE:> While the notation of this construct is the same as the similar
707function in .NET regexes, the behavior is not, in Perl the buffers are
708numbered sequentially regardless of being named or not. Thus in the
709pattern
710
711 /(x)(?<foo>y)(z)/
712
713$+{foo} will be the same as $2, and $3 will contain 'z' instead of
714the opposite which is what a .NET regex hacker might expect.
715
716Currently NAME is restricted to word chars only. In other words, it
717must match C</^\w+$/>.
718
719=item C<< \k<name> >>
720
721=item C<< \k'name' >>
722
723Named backreference. Similar to numeric backreferences, except that
724the group is designated by name and not number. If multiple groups
725have the same name then it refers to the leftmost defined group in
726the current match.
727
728It is an error to refer to a name not defined by a C<(?<NAME>)>
729earlier in the pattern.
730
731Both forms are equivalent.
732
cc6b7395 733=item C<(?{ code })>
d74e8afc 734X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 735
19799a22 736B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
737experimental, and may be changed without notice. Code executed that
738has side effects may not perform identically from version to version
739due to the effect of future optimisations in the regex engine.
c277df42 740
cc46d5f2 741This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
742always succeeds, and its C<code> is not interpolated. Currently,
743the rules to determine where the C<code> ends are somewhat convoluted.
744
77ea4f6d
JV
745This feature can be used together with the special variable C<$^N> to
746capture the results of submatches in variables without having to keep
747track of the number of nested parentheses. For example:
748
749 $_ = "The brown fox jumps over the lazy dog";
750 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
751 print "color = $color, animal = $animal\n";
752
754091cb
RGS
753Inside the C<(?{...})> block, C<$_> refers to the string the regular
754expression is matching against. You can also use C<pos()> to know what is
fa11829f 755the current position of matching within this string.
754091cb 756
19799a22
GS
757The C<code> is properly scoped in the following sense: If the assertion
758is backtracked (compare L<"Backtracking">), all changes introduced after
759C<local>ization are undone, so that
b9ac3b5b
GS
760
761 $_ = 'a' x 8;
5d458dd8 762 m<
b9ac3b5b
GS
763 (?{ $cnt = 0 }) # Initialize $cnt.
764 (
5d458dd8 765 a
b9ac3b5b
GS
766 (?{
767 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
768 })
5d458dd8 769 )*
b9ac3b5b
GS
770 aaaa
771 (?{ $res = $cnt }) # On success copy to non-localized
772 # location.
773 >x;
774
19799a22 775will set C<$res = 4>. Note that after the match, $cnt returns to the globally
14218588 776introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
777are unwound.
778
19799a22
GS
779This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
780switch. If I<not> used in this way, the result of evaluation of
781C<code> is put into the special variable C<$^R>. This happens
782immediately, so C<$^R> can be used from other C<(?{ code })> assertions
783inside the same regular expression.
b9ac3b5b 784
19799a22
GS
785The assignment to C<$^R> above is properly localized, so the old
786value of C<$^R> is restored if the assertion is backtracked; compare
787L<"Backtracking">.
b9ac3b5b 788
61528107
SP
789Due to an unfortunate implementation issue, the Perl code contained in these
790blocks is treated as a compile time closure that can have seemingly bizarre
6bda09f9 791consequences when used with lexically scoped variables inside of subroutines
61528107
SP
792or loops. There are various workarounds for this, including simply using
793global variables instead. If you are using this construct and strange results
6bda09f9
YO
794occur then check for the use of lexically scoped variables.
795
19799a22
GS
796For reasons of security, this construct is forbidden if the regular
797expression involves run-time interpolation of variables, unless the
798perilous C<use re 'eval'> pragma has been used (see L<re>), or the
799variables contain results of C<qr//> operator (see
5d458dd8 800L<perlop/"qr/STRING/imosx">).
871b0233 801
14218588 802This restriction is because of the wide-spread and remarkably convenient
19799a22 803custom of using run-time determined strings as patterns. For example:
871b0233
IZ
804
805 $re = <>;
806 chomp $re;
807 $string =~ /$re/;
808
14218588
GS
809Before Perl knew how to execute interpolated code within a pattern,
810this operation was completely safe from a security point of view,
811although it could raise an exception from an illegal pattern. If
812you turn on the C<use re 'eval'>, though, it is no longer secure,
813so you should only do so if you are also using taint checking.
814Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 815compartment. See L<perlsec> for details about both these mechanisms.
871b0233 816
5d458dd8 817Because perl's regex engine is not currently re-entrant, interpolated
8988a1bb
DD
818code may not invoke the regex engine either directly with C<m//> or C<s///>),
819or indirectly with functions such as C<split>.
820
14455d6c 821=item C<(??{ code })>
d74e8afc
ITB
822X<(??{})>
823X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 824
19799a22 825B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
826experimental, and may be changed without notice. Code executed that
827has side effects may not perform identically from version to version
828due to the effect of future optimisations in the regex engine.
0f5d15d6 829
19799a22
GS
830This is a "postponed" regular subexpression. The C<code> is evaluated
831at run time, at the moment this subexpression may match. The result
832of evaluation is considered as a regular expression and matched as
61528107 833if it were inserted instead of this construct. Note that this means
6bda09f9
YO
834that the contents of capture buffers defined inside an eval'ed pattern
835are not available outside of the pattern, and vice versa, there is no
836way for the inner pattern to refer to a capture buffer defined outside.
837Thus,
838
839 ('a' x 100)=~/(??{'(.)' x 100})/
840
81714fb9 841B<will> match, it will B<not> set $1.
0f5d15d6 842
428594d9 843The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
844where the C<code> ends are currently somewhat convoluted.
845
846The following pattern matches a parenthesized group:
0f5d15d6
IZ
847
848 $re = qr{
849 \(
850 (?:
851 (?> [^()]+ ) # Non-parens without backtracking
852 |
14455d6c 853 (??{ $re }) # Group with matching parens
0f5d15d6
IZ
854 )*
855 \)
856 }x;
857
6bda09f9
YO
858See also C<(?PARNO)> for a different, more efficient way to accomplish
859the same task.
860
5d458dd8 861Because perl's regex engine is not currently re-entrant, delayed
8988a1bb
DD
862code may not invoke the regex engine either directly with C<m//> or C<s///>),
863or indirectly with functions such as C<split>.
864
5d458dd8
YO
865Recursing deeper than 50 times without consuming any input string will
866result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
867changing it requires a custom build.
868
542fa716
YO
869=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
870X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 871X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 872X<regex, relative recursion>
6bda09f9 873
81714fb9
YO
874Similar to C<(??{ code })> except it does not involve compiling any code,
875instead it treats the contents of a capture buffer as an independent
61528107 876pattern that must match at the current position. Capture buffers
81714fb9 877contained by the pattern will have the value as determined by the
6bda09f9
YO
878outermost recursion.
879
894be9b7
YO
880PARNO is a sequence of digits (not starting with 0) whose value reflects
881the paren-number of the capture buffer to recurse to. C<(?R)> recurses to
882the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716
YO
883C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
884to be relative, with negative numbers indicating preceding capture buffers
885and positive ones following. Thus C<(?-1)> refers to the most recently
886declared buffer, and C<(?+1)> indicates the next buffer to be declared.
6bda09f9 887
81714fb9 888The following pattern matches a function foo() which may contain
f145b7e9 889balanced parentheses as the argument.
6bda09f9
YO
890
891 $re = qr{ ( # paren group 1 (full function)
81714fb9 892 foo
6bda09f9
YO
893 ( # paren group 2 (parens)
894 \(
895 ( # paren group 3 (contents of parens)
896 (?:
897 (?> [^()]+ ) # Non-parens without backtracking
898 |
899 (?2) # Recurse to start of paren group 2
900 )*
901 )
902 \)
903 )
904 )
905 }x;
906
907If the pattern was used as follows
908
909 'foo(bar(baz)+baz(bop))'=~/$re/
910 and print "\$1 = $1\n",
911 "\$2 = $2\n",
912 "\$3 = $3\n";
913
914the output produced should be the following:
915
916 $1 = foo(bar(baz)+baz(bop))
917 $2 = (bar(baz)+baz(bop))
81714fb9 918 $3 = bar(baz)+baz(bop)
6bda09f9 919
81714fb9 920If there is no corresponding capture buffer defined, then it is a
61528107 921fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 922string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
923into perl, so changing it requires a custom build.
924
542fa716
YO
925The following shows how using negative indexing can make it
926easier to embed recursive patterns inside of a C<qr//> construct
927for later use:
928
929 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
930 if (/foo $parens \s+ + \s+ bar $parens/x) {
931 # do something here...
932 }
933
81714fb9 934B<Note> that this pattern does not behave the same way as the equivalent
6bda09f9
YO
935PCRE or Python construct of the same form. In perl you can backtrack into
936a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
937as atomic. Also, modifiers are resolved at compile time, so constructs
938like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
939be processed.
6bda09f9 940
894be9b7
YO
941=item C<(?&NAME)>
942X<(?&NAME)>
943
944Recurse to a named subpattern. Identical to (?PARNO) except that the
945parenthesis to recurse to is determined by name. If multiple parens have
946the same name, then it recurses to the leftmost.
947
948It is an error to refer to a name that is not declared somewhere in the
949pattern.
950
e2e6a0f1
YO
951=item C<(?(condition)yes-pattern|no-pattern)>
952X<(?()>
286f584a 953
e2e6a0f1 954=item C<(?(condition)yes-pattern)>
286f584a 955
e2e6a0f1
YO
956Conditional expression. C<(condition)> should be either an integer in
957parentheses (which is valid if the corresponding pair of parentheses
958matched), a look-ahead/look-behind/evaluate zero-width assertion, a
959name in angle brackets or single quotes (which is valid if a buffer
960with the given name matched), or the special symbol (R) (true when
961evaluated inside of recursion or eval). Additionally the R may be
962followed by a number, (which will be true when evaluated when recursing
963inside of the appropriate group), or by C<&NAME>, in which case it will
964be true only when evaluated during recursion in the named group.
965
966Here's a summary of the possible predicates:
967
968=over 4
969
970=item (1) (2) ...
971
972Checks if the numbered capturing buffer has matched something.
973
974=item (<NAME>) ('NAME')
975
976Checks if a buffer with the given name has matched something.
977
978=item (?{ CODE })
979
980Treats the code block as the condition.
981
982=item (R)
983
984Checks if the expression has been evaluated inside of recursion.
985
986=item (R1) (R2) ...
987
988Checks if the expression has been evaluated while executing directly
989inside of the n-th capture group. This check is the regex equivalent of
990
991 if ((caller(0))[3] eq 'subname') { ... }
992
993In other words, it does not check the full recursion stack.
994
995=item (R&NAME)
996
997Similar to C<(R1)>, this predicate checks to see if we're executing
998directly inside of the leftmost group with a given name (this is the same
999logic used by C<(?&NAME)> to disambiguate). It does not check the full
1000stack, but only the name of the innermost active recursion.
1001
1002=item (DEFINE)
1003
1004In this case, the yes-pattern is never directly executed, and no
1005no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1006See below for details.
1007
1008=back
1009
1010For example:
1011
1012 m{ ( \( )?
1013 [^()]+
1014 (?(1) \) )
1015 }x
1016
1017matches a chunk of non-parentheses, possibly included in parentheses
1018themselves.
1019
1020A special form is the C<(DEFINE)> predicate, which never executes directly
1021its yes-pattern, and does not allow a no-pattern. This allows to define
1022subpatterns which will be executed only by using the recursion mechanism.
1023This way, you can define a set of regular expression rules that can be
1024bundled into any pattern you choose.
1025
1026It is recommended that for this usage you put the DEFINE block at the
1027end of the pattern, and that you name any subpatterns defined within it.
1028
1029Also, it's worth noting that patterns defined this way probably will
1030not be as efficient, as the optimiser is not very clever about
1031handling them.
1032
1033An example of how this might be used is as follows:
1034
1035 /(?<NAME>(&NAME_PAT))(?<ADDR>(&ADDRESS_PAT))
1036 (?(DEFINE)
1037 (<NAME_PAT>....)
1038 (<ADRESS_PAT>....)
1039 )/x
1040
1041Note that capture buffers matched inside of recursion are not accessible
1042after the recursion returns, so the extra layer of capturing buffers are
1043necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1044C<$+{NAME}> would be.
286f584a 1045
c47ff5f1 1046=item C<< (?>pattern) >>
6bda09f9 1047X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1048
19799a22
GS
1049An "independent" subexpression, one which matches the substring
1050that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1051position, and it matches I<nothing other than this substring>. This
19799a22
GS
1052construct is useful for optimizations of what would otherwise be
1053"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1054It may also be useful in places where the "grab all you can, and do not
1055give anything back" semantic is desirable.
19799a22 1056
c47ff5f1 1057For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1058(anchored at the beginning of string, as above) will match I<all>
1059characters C<a> at the beginning of string, leaving no C<a> for
1060C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1061since the match of the subgroup C<a*> is influenced by the following
1062group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1063C<a*ab> will match fewer characters than a standalone C<a*>, since
1064this makes the tail match.
1065
c47ff5f1 1066An effect similar to C<< (?>pattern) >> may be achieved by writing
19799a22
GS
1067C<(?=(pattern))\1>. This matches the same substring as a standalone
1068C<a+>, and the following C<\1> eats the matched string; it therefore
c47ff5f1 1069makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1070(The difference between these two constructs is that the second one
1071uses a capturing group, thus shifting ordinals of backreferences
1072in the rest of a regular expression.)
1073
1074Consider this pattern:
c277df42 1075
871b0233 1076 m{ \(
e2e6a0f1
YO
1077 (
1078 [^()]+ # x+
1079 |
871b0233
IZ
1080 \( [^()]* \)
1081 )+
e2e6a0f1 1082 \)
871b0233 1083 }x
5a964f20 1084
19799a22
GS
1085That will efficiently match a nonempty group with matching parentheses
1086two levels deep or less. However, if there is no such group, it
1087will take virtually forever on a long string. That's because there
1088are so many different ways to split a long string into several
1089substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1090to a subpattern of the above pattern. Consider how the pattern
1091above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1092seconds, but that each extra letter doubles this time. This
1093exponential performance will make it appear that your program has
14218588 1094hung. However, a tiny change to this pattern
5a964f20 1095
e2e6a0f1
YO
1096 m{ \(
1097 (
1098 (?> [^()]+ ) # change x+ above to (?> x+ )
1099 |
871b0233
IZ
1100 \( [^()]* \)
1101 )+
e2e6a0f1 1102 \)
871b0233 1103 }x
c277df42 1104
c47ff5f1 1105which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1106this yourself would be a productive exercise), but finishes in a fourth
1107the time when used on a similar string with 1000000 C<a>s. Be aware,
1108however, that this pattern currently triggers a warning message under
9f1b1f2d 1109the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1110C<"matches null string many times in regex">.
c277df42 1111
c47ff5f1 1112On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1113effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1114This was only 4 times slower on a string with 1000000 C<a>s.
1115
9da458fc
IZ
1116The "grab all you can, and do not give anything back" semantic is desirable
1117in many situations where on the first sight a simple C<()*> looks like
1118the correct solution. Suppose we parse text with comments being delimited
1119by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1120its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1121the comment delimiter, because it may "give up" some whitespace if
1122the remainder of the pattern can be made to match that way. The correct
1123answer is either one of these:
1124
1125 (?>#[ \t]*)
1126 #[ \t]*(?![ \t])
1127
1128For example, to grab non-empty comments into $1, one should use either
1129one of these:
1130
1131 / (?> \# [ \t]* ) ( .+ ) /x;
1132 / \# [ \t]* ( [^ \t] .* ) /x;
1133
1134Which one you pick depends on which of these expressions better reflects
1135the above specification of comments.
1136
6bda09f9
YO
1137In some literature this construct is called "atomic matching" or
1138"possessive matching".
1139
b9b4dddf
YO
1140Possessive quantifiers are equivalent to putting the item they are applied
1141to inside of one of these constructs. The following equivalences apply:
1142
1143 Quantifier Form Bracketing Form
1144 --------------- ---------------
1145 PAT*+ (?>PAT*)
1146 PAT++ (?>PAT+)
1147 PAT?+ (?>PAT?)
1148 PAT{min,max}+ (?>PAT{min,max})
1149
e2e6a0f1
YO
1150=back
1151
1152=head2 Special Backtracking Control Verbs
1153
1154B<WARNING:> These patterns are experimental and subject to change or
1155removal in a future version of perl. Their usage in production code should
1156be noted to avoid problems during upgrades.
1157
1158These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1159otherwise stated the ARG argument is optional; in some cases, it is
1160forbidden.
1161
1162Any pattern containing a special backtracking verb that allows an argument
1163has the special behaviour that when executed it sets the current packages'
5d458dd8
YO
1164C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1165rules apply:
e2e6a0f1 1166
5d458dd8
YO
1167On failure, the C<$REGERROR> variable will be set to the ARG value of the
1168verb pattern, if the verb was involved in the failure of the match. If the
1169ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1170name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1171none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1172
5d458dd8
YO
1173On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1174the C<$REGMARK> variable will be set to the name of the last
1175C<(*MARK:NAME)> pattern executed. See the explanation for the
1176C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1177
5d458dd8
YO
1178B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1179and most other regex related variables. They are not local to a scope, nor
1180readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1181Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1182
1183If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1184argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1
YO
1185
1186=over 4
1187
1188=item Verbs that take an argument
1189
1190=over 4
1191
5d458dd8
YO
1192=item C<(*PRUNE)> C<(*PRUNE:NAME)>
1193X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1194
5d458dd8
YO
1195This zero-width pattern prunes the backtracking tree at the current point
1196when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1197where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1198A may backtrack as necessary to match. Once it is reached, matching
1199continues in B, which may also backtrack as necessary; however, should B
1200not match, then no further backtracking will take place, and the pattern
1201will fail outright at the current starting position.
54612592
YO
1202
1203The following example counts all the possible matching strings in a
1204pattern (without actually matching any of them).
1205
e2e6a0f1 1206 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1207 print "Count=$count\n";
1208
1209which produces:
1210
1211 aaab
1212 aaa
1213 aa
1214 a
1215 aab
1216 aa
1217 a
1218 ab
1219 a
1220 Count=9
1221
5d458dd8 1222If we add a C<(*PRUNE)> before the count like the following
54612592 1223
5d458dd8 1224 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1225 print "Count=$count\n";
1226
1227we prevent backtracking and find the count of the longest matching
1228at each matching startpoint like so:
1229
1230 aaab
1231 aab
1232 ab
1233 Count=3
1234
5d458dd8 1235Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1236
5d458dd8
YO
1237See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1238control backtracking. In some cases, the use of C<(*PRUNE)> can be
1239replaced with a C<< (?>pattern) >> with no functional difference; however,
1240C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1241C<< (?>pattern) >> alone.
54612592 1242
e2e6a0f1 1243
5d458dd8
YO
1244=item C<(*SKIP)> C<(*SKIP:NAME)>
1245X<(*SKIP)>
e2e6a0f1 1246
5d458dd8 1247This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1248failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1249to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1250of this pattern. This effectively means that the regex engine "skips" forward
1251to this position on failure and tries to match again, (assuming that
1252there is sufficient room to match).
1253
1254The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1255C<(*MARK:NAME)> was encountered while matching, then it is that position
1256which is used as the "skip point". If no C<(*MARK)> of that name was
1257encountered, then the C<(*SKIP)> operator has no effect. When used
1258without a name the "skip point" is where the match point was when
1259executing the (*SKIP) pattern.
1260
1261Compare the following to the examples in C<(*PRUNE)>, note the string
24b23f37
YO
1262is twice as long:
1263
5d458dd8 1264 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1265 print "Count=$count\n";
1266
1267outputs
1268
1269 aaab
1270 aaab
1271 Count=2
1272
5d458dd8 1273Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
e2e6a0f1 1274executed, the next startpoint will be where the cursor was when the
5d458dd8
YO
1275C<(*SKIP)> was executed.
1276
1277As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1278
1279=item C<(*MARK:NAME)> C<(*:NAME)>
1280X<(*MARK)> C<(*MARK:NAME)> C<(*:NAME)>
1281
1282This zero-width pattern can be used to mark the point reached in a string
1283when a certain part of the pattern has been successfully matched. This
1284mark may be given a name. A later C<(*SKIP)> pattern will then skip
1285forward to that point if backtracked into on failure. Any number of
1286C<(*MARK)> patterns are allowed, and the NAME portion is optional and may
1287be duplicated.
1288
1289In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1290can be used to "label" a pattern branch, so that after matching, the
1291program can determine which branches of the pattern were involved in the
1292match.
1293
1294When a match is successful, the C<$REGMARK> variable will be set to the
1295name of the most recently executed C<(*MARK:NAME)> that was involved
1296in the match.
1297
1298This can be used to determine which branch of a pattern was matched
1299without using a seperate capture buffer for each branch, which in turn
1300can result in a performance improvement, as perl cannot optimize
1301C</(?:(x)|(y)|(z))/> as efficiently as something like
1302C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1303
1304When a match has failed, and unless another verb has been involved in
1305failing the match and has provided its own name to use, the C<$REGERROR>
1306variable will be set to the name of the most recently executed
1307C<(*MARK:NAME)>.
1308
1309See C<(*SKIP)> for more details.
1310
1311=item C<(*THEN)> C<(*THEN:NAME)>
1312
1313This is similar to the "cut group" operator C<::> from Perl6. Like
1314C<(*PRUNE)>, this verb always matches, and when backtracked into on
1315failure, it causes the regex engine to try the next alternation in the
1316innermost enclosing group (capturing or otherwise).
1317
1318Its name comes from the observation that this operation combined with the
1319alternation operator (C<|>) can be used to create what is essentially a
1320pattern-based if/then/else block:
1321
1322 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1323
1324Note that if this operator is used and NOT inside of an alternation then
1325it acts exactly like the C<(*PRUNE)> operator.
1326
1327 / A (*PRUNE) B /
1328
1329is the same as
1330
1331 / A (*THEN) B /
1332
1333but
1334
1335 / ( A (*THEN) B | C (*THEN) D ) /
1336
1337is not the same as
1338
1339 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1340
1341as after matching the A but failing on the B the C<(*THEN)> verb will
1342backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1343
e2e6a0f1
YO
1344=item C<(*COMMIT)>
1345X<(*COMMIT)>
24b23f37 1346
5d458dd8
YO
1347This is the Perl6 "commit pattern" C<< <commit> >> or C<:::>. It's a
1348zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1349into on failure it causes the match to fail outright. No further attempts
1350to find a valid match by advancing the start pointer will occur again.
1351For example,
24b23f37 1352
e2e6a0f1 1353 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1354 print "Count=$count\n";
1355
1356outputs
1357
1358 aaab
1359 Count=1
1360
e2e6a0f1
YO
1361In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1362does not match, the regex engine will not try any further matching on the
1363rest of the string.
c277df42 1364
e2e6a0f1 1365=back
9af228c6 1366
e2e6a0f1 1367=item Verbs without an argument
9af228c6
YO
1368
1369=over 4
1370
e2e6a0f1
YO
1371=item C<(*FAIL)> C<(*F)>
1372X<(*FAIL)> X<(*F)>
9af228c6 1373
e2e6a0f1
YO
1374This pattern matches nothing and always fails. It can be used to force the
1375engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1376fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1377
e2e6a0f1 1378It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1379
e2e6a0f1
YO
1380=item C<(*ACCEPT)>
1381X<(*ACCEPT)>
9af228c6 1382
e2e6a0f1
YO
1383B<WARNING:> This feature is highly experimental. It is not recommended
1384for production code.
9af228c6 1385
e2e6a0f1
YO
1386This pattern matches nothing and causes the end of successful matching at
1387the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1388whether there is actually more to match in the string. When inside of a
1389nested pattern, such as recursion or a dynamically generated subbpattern
1390via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1391
e2e6a0f1
YO
1392If the C<(*ACCEPT)> is inside of capturing buffers then the buffers are
1393marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1394For instance:
9af228c6 1395
e2e6a0f1 1396 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1397
e2e6a0f1
YO
1398will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
1399be set. If another branch in the inner parens were matched, such as in the
1400string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1401
1402=back
c277df42 1403
a0d0e21e
LW
1404=back
1405
c07a80fd 1406=head2 Backtracking
d74e8afc 1407X<backtrack> X<backtracking>
c07a80fd 1408
35a734be
IZ
1409NOTE: This section presents an abstract approximation of regular
1410expression behavior. For a more rigorous (and complicated) view of
1411the rules involved in selecting a match among possible alternatives,
1412see L<Combining pieces together>.
1413
c277df42 1414A fundamental feature of regular expression matching involves the
5a964f20 1415notion called I<backtracking>, which is currently used (when needed)
c277df42 1416by all regular expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1417C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1418internally, but the general principle outlined here is valid.
c07a80fd
PP
1419
1420For a regular expression to match, the I<entire> regular expression must
1421match, not just part of it. So if the beginning of a pattern containing a
1422quantifier succeeds in a way that causes later parts in the pattern to
1423fail, the matching engine backs up and recalculates the beginning
1424part--that's why it's called backtracking.
1425
1426Here is an example of backtracking: Let's say you want to find the
1427word following "foo" in the string "Food is on the foo table.":
1428
1429 $_ = "Food is on the foo table.";
1430 if ( /\b(foo)\s+(\w+)/i ) {
1431 print "$2 follows $1.\n";
1432 }
1433
1434When the match runs, the first part of the regular expression (C<\b(foo)>)
1435finds a possible match right at the beginning of the string, and loads up
1436$1 with "Foo". However, as soon as the matching engine sees that there's
1437no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1438mistake and starts over again one character after where it had the
c07a80fd
PP
1439tentative match. This time it goes all the way until the next occurrence
1440of "foo". The complete regular expression matches this time, and you get
1441the expected output of "table follows foo."
1442
1443Sometimes minimal matching can help a lot. Imagine you'd like to match
1444everything between "foo" and "bar". Initially, you write something
1445like this:
1446
1447 $_ = "The food is under the bar in the barn.";
1448 if ( /foo(.*)bar/ ) {
1449 print "got <$1>\n";
1450 }
1451
1452Which perhaps unexpectedly yields:
1453
1454 got <d is under the bar in the >
1455
1456That's because C<.*> was greedy, so you get everything between the
14218588 1457I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1458to use minimal matching to make sure you get the text between a "foo"
1459and the first "bar" thereafter.
1460
1461 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1462 got <d is under the >
1463
1464Here's another example: let's say you'd like to match a number at the end
b6e13d97 1465of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
1466So you write this:
1467
1468 $_ = "I have 2 numbers: 53147";
1469 if ( /(.*)(\d*)/ ) { # Wrong!
1470 print "Beginning is <$1>, number is <$2>.\n";
1471 }
1472
1473That won't work at all, because C<.*> was greedy and gobbled up the
1474whole string. As C<\d*> can match on an empty string the complete
1475regular expression matched successfully.
1476
8e1088bc 1477 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
1478
1479Here are some variants, most of which don't work:
1480
1481 $_ = "I have 2 numbers: 53147";
1482 @pats = qw{
1483 (.*)(\d*)
1484 (.*)(\d+)
1485 (.*?)(\d*)
1486 (.*?)(\d+)
1487 (.*)(\d+)$
1488 (.*?)(\d+)$
1489 (.*)\b(\d+)$
1490 (.*\D)(\d+)$
1491 };
1492
1493 for $pat (@pats) {
1494 printf "%-12s ", $pat;
1495 if ( /$pat/ ) {
1496 print "<$1> <$2>\n";
1497 } else {
1498 print "FAIL\n";
1499 }
1500 }
1501
1502That will print out:
1503
1504 (.*)(\d*) <I have 2 numbers: 53147> <>
1505 (.*)(\d+) <I have 2 numbers: 5314> <7>
1506 (.*?)(\d*) <> <>
1507 (.*?)(\d+) <I have > <2>
1508 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1509 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1510 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1511 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1512
1513As you see, this can be a bit tricky. It's important to realize that a
1514regular expression is merely a set of assertions that gives a definition
1515of success. There may be 0, 1, or several different ways that the
1516definition might succeed against a particular string. And if there are
5a964f20
TC
1517multiple ways it might succeed, you need to understand backtracking to
1518know which variety of success you will achieve.
c07a80fd 1519
19799a22 1520When using look-ahead assertions and negations, this can all get even
8b19b778 1521trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
1522followed by "123". You might try to write that as
1523
871b0233
IZ
1524 $_ = "ABC123";
1525 if ( /^\D*(?!123)/ ) { # Wrong!
1526 print "Yup, no 123 in $_\n";
1527 }
c07a80fd
PP
1528
1529But that isn't going to match; at least, not the way you're hoping. It
1530claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1531why that pattern matches, contrary to popular expectations:
c07a80fd 1532
4358a253
SS
1533 $x = 'ABC123';
1534 $y = 'ABC445';
c07a80fd 1535
4358a253
SS
1536 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1537 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1538
4358a253
SS
1539 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1540 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
1541
1542This prints
1543
1544 2: got ABC
1545 3: got AB
1546 4: got ABC
1547
5f05dabc 1548You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
1549general purpose version of test 1. The important difference between
1550them is that test 3 contains a quantifier (C<\D*>) and so can use
1551backtracking, whereas test 1 will not. What's happening is
1552that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1553non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1554let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1555fail.
14218588 1556
c07a80fd 1557The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1558try to match C<(?!123> with "123", which fails. But because
c07a80fd
PP
1559a quantifier (C<\D*>) has been used in the regular expression, the
1560search engine can backtrack and retry the match differently
54310121 1561in the hope of matching the complete regular expression.
c07a80fd 1562
5a964f20
TC
1563The pattern really, I<really> wants to succeed, so it uses the
1564standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1565time. Now there's indeed something following "AB" that is not
14218588 1566"123". It's "C123", which suffices.
c07a80fd 1567
14218588
GS
1568We can deal with this by using both an assertion and a negation.
1569We'll say that the first part in $1 must be followed both by a digit
1570and by something that's not "123". Remember that the look-aheads
1571are zero-width expressions--they only look, but don't consume any
1572of the string in their match. So rewriting this way produces what
c07a80fd
PP
1573you'd expect; that is, case 5 will fail, but case 6 succeeds:
1574
4358a253
SS
1575 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1576 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
1577
1578 6: got ABC
1579
5a964f20 1580In other words, the two zero-width assertions next to each other work as though
19799a22 1581they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
1582matches only if you're at the beginning of the line AND the end of the
1583line simultaneously. The deeper underlying truth is that juxtaposition in
1584regular expressions always means AND, except when you write an explicit OR
1585using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1586although the attempted matches are made at different positions because "a"
1587is not a zero-width assertion, but a one-width assertion.
1588
19799a22 1589B<WARNING>: particularly complicated regular expressions can take
14218588 1590exponential time to solve because of the immense number of possible
9da458fc
IZ
1591ways they can use backtracking to try match. For example, without
1592internal optimizations done by the regular expression engine, this will
1593take a painfully long time to run:
c07a80fd 1594
e1901655
IZ
1595 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1596
1597And if you used C<*>'s in the internal groups instead of limiting them
1598to 0 through 5 matches, then it would take forever--or until you ran
1599out of stack space. Moreover, these internal optimizations are not
1600always applicable. For example, if you put C<{0,5}> instead of C<*>
1601on the external group, no current optimization is applicable, and the
1602match takes a long time to finish.
c07a80fd 1603
9da458fc
IZ
1604A powerful tool for optimizing such beasts is what is known as an
1605"independent group",
c47ff5f1 1606which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1607zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 1608the tail match, since they are in "logical" context: only
14218588 1609whether they match is considered relevant. For an example
9da458fc 1610where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1611following match, see L<C<< (?>pattern) >>>.
c277df42 1612
a0d0e21e 1613=head2 Version 8 Regular Expressions
d74e8afc 1614X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1615
5a964f20 1616In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1617routines, here are the pattern-matching rules not described above.
1618
54310121 1619Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1620with a special meaning described here or above. You can cause
5a964f20 1621characters that normally function as metacharacters to be interpreted
5f05dabc 1622literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
a0d0e21e
LW
1623character; "\\" matches a "\"). A series of characters matches that
1624series of characters in the target string, so the pattern C<blurfl>
1625would match "blurfl" in the target string.
1626
1627You can specify a character class, by enclosing a list of characters
5d458dd8 1628in C<[]>, which will match any character from the list. If the
a0d0e21e 1629first character after the "[" is "^", the class matches any character not
14218588 1630in the list. Within a list, the "-" character specifies a
5a964f20 1631range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1632inclusive. If you want either "-" or "]" itself to be a member of a
1633class, put it at the start of the list (possibly after a "^"), or
1634escape it with a backslash. "-" is also taken literally when it is
1635at the end of the list, just before the closing "]". (The
84850974
DD
1636following all specify the same class of three characters: C<[-az]>,
1637C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
1638specifies a class containing twenty-six characters, even on EBCDIC-based
1639character sets.) Also, if you try to use the character
1640classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1641a range, the "-" is understood literally.
a0d0e21e 1642
8ada0baa
JH
1643Note also that the whole range idea is rather unportable between
1644character sets--and even within character sets they may cause results
1645you probably didn't expect. A sound principle is to use only ranges
1646that begin from and end at either alphabets of equal case ([a-e],
1647[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1648spell out the character sets in full.
1649
54310121 1650Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1651used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1652"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
5d458dd8
YO
1653of octal digits, matches the character whose coded character set value
1654is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
1655matches the character whose numeric value is I<nn>. The expression \cI<x>
1656matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 1657matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1658
1659You can specify a series of alternatives for a pattern using "|" to
1660separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1661or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1662first alternative includes everything from the last pattern delimiter
1663("(", "[", or the beginning of the pattern) up to the first "|", and
1664the last alternative contains everything from the last "|" to the next
14218588
GS
1665pattern delimiter. That's why it's common practice to include
1666alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1667start and end.
1668
5a964f20 1669Alternatives are tried from left to right, so the first
a3cb178b
GS
1670alternative found for which the entire expression matches, is the one that
1671is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1672example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1673part will match, as that is the first alternative tried, and it successfully
1674matches the target string. (This might not seem important, but it is
1675important when you are capturing matched text using parentheses.)
1676
5a964f20 1677Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1678so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1679
14218588
GS
1680Within a pattern, you may designate subpatterns for later reference
1681by enclosing them in parentheses, and you may refer back to the
1682I<n>th subpattern later in the pattern using the metacharacter
1683\I<n>. Subpatterns are numbered based on the left to right order
1684of their opening parenthesis. A backreference matches whatever
1685actually matched the subpattern in the string being examined, not
1686the rules for that subpattern. Therefore, C<(0|0x)\d*\s\1\d*> will
1687match "0x1234 0x4321", but not "0x1234 01234", because subpattern
16881 matched "0x", even though the rule C<0|0x> could potentially match
1689the leading 0 in the second number.
cb1a09d0 1690
19799a22 1691=head2 Warning on \1 vs $1
cb1a09d0 1692
5a964f20 1693Some people get too used to writing things like:
cb1a09d0
AD
1694
1695 $pattern =~ s/(\W)/\\\1/g;
1696
1697This is grandfathered for the RHS of a substitute to avoid shocking the
1698B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1699PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1700the usual double-quoted string means a control-A. The customary Unix
1701meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1702of doing that, you get yourself into trouble if you then add an C</e>
1703modifier.
1704
5a964f20 1705 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1706
1707Or if you try to do
1708
1709 s/(\d+)/\1000/;
1710
1711You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1712C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1713with the operation of matching a backreference. Certainly they mean two
1714different things on the I<left> side of the C<s///>.
9fa51da4 1715
c84d73f1
IZ
1716=head2 Repeated patterns matching zero-length substring
1717
19799a22 1718B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1719
1720Regular expressions provide a terse and powerful programming language. As
1721with most other power tools, power comes together with the ability
1722to wreak havoc.
1723
1724A common abuse of this power stems from the ability to make infinite
628afcb5 1725loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1726
1727 'foo' =~ m{ ( o? )* }x;
1728
1729The C<o?> can match at the beginning of C<'foo'>, and since the position
1730in the string is not moved by the match, C<o?> would match again and again
14218588 1731because of the C<*> modifier. Another common way to create a similar cycle
c84d73f1
IZ
1732is with the looping modifier C<//g>:
1733
1734 @matches = ( 'foo' =~ m{ o? }xg );
1735
1736or
1737
1738 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1739
1740or the loop implied by split().
1741
1742However, long experience has shown that many programming tasks may
14218588
GS
1743be significantly simplified by using repeated subexpressions that
1744may match zero-length substrings. Here's a simple example being:
c84d73f1
IZ
1745
1746 @chars = split //, $string; # // is not magic in split
1747 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1748
9da458fc 1749Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1
IZ
1750the infinite loop>. The rules for this are different for lower-level
1751loops given by the greedy modifiers C<*+{}>, and for higher-level
1752ones like the C</g> modifier or split() operator.
1753
19799a22
GS
1754The lower-level loops are I<interrupted> (that is, the loop is
1755broken) when Perl detects that a repeated expression matched a
1756zero-length substring. Thus
c84d73f1
IZ
1757
1758 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1759
5d458dd8 1760is made equivalent to
c84d73f1 1761
5d458dd8
YO
1762 m{ (?: NON_ZERO_LENGTH )*
1763 |
1764 (?: ZERO_LENGTH )?
c84d73f1
IZ
1765 }x;
1766
1767The higher level-loops preserve an additional state between iterations:
5d458dd8 1768whether the last match was zero-length. To break the loop, the following
c84d73f1 1769match after a zero-length match is prohibited to have a length of zero.
5d458dd8 1770This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
1771and so the I<second best> match is chosen if the I<best> match is of
1772zero length.
1773
19799a22 1774For example:
c84d73f1
IZ
1775
1776 $_ = 'bar';
1777 s/\w??/<$&>/g;
1778
20fb949f 1779results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 1780match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
1781best> match is what is matched by C<\w>. Thus zero-length matches
1782alternate with one-character-long matches.
1783
5d458dd8 1784Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
1785position one notch further in the string.
1786
19799a22 1787The additional state of being I<matched with zero-length> is associated with
c84d73f1 1788the matched string, and is reset by each assignment to pos().
9da458fc
IZ
1789Zero-length matches at the end of the previous match are ignored
1790during C<split>.
c84d73f1 1791
35a734be
IZ
1792=head2 Combining pieces together
1793
1794Each of the elementary pieces of regular expressions which were described
1795before (such as C<ab> or C<\Z>) could match at most one substring
1796at the given position of the input string. However, in a typical regular
1797expression these elementary pieces are combined into more complicated
1798patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1799(in these examples C<S> and C<T> are regular subexpressions).
1800
1801Such combinations can include alternatives, leading to a problem of choice:
1802if we match a regular expression C<a|ab> against C<"abc">, will it match
1803substring C<"a"> or C<"ab">? One way to describe which substring is
1804actually matched is the concept of backtracking (see L<"Backtracking">).
1805However, this description is too low-level and makes you think
1806in terms of a particular implementation.
1807
1808Another description starts with notions of "better"/"worse". All the
1809substrings which may be matched by the given regular expression can be
1810sorted from the "best" match to the "worst" match, and it is the "best"
1811match which is chosen. This substitutes the question of "what is chosen?"
1812by the question of "which matches are better, and which are worse?".
1813
1814Again, for elementary pieces there is no such question, since at most
1815one match at a given position is possible. This section describes the
1816notion of better/worse for combining operators. In the description
1817below C<S> and C<T> are regular subexpressions.
1818
13a2d996 1819=over 4
35a734be
IZ
1820
1821=item C<ST>
1822
1823Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
1824substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 1825which can be matched by C<T>.
35a734be
IZ
1826
1827If C<A> is better match for C<S> than C<A'>, C<AB> is a better
1828match than C<A'B'>.
1829
1830If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
1831C<B> is better match for C<T> than C<B'>.
1832
1833=item C<S|T>
1834
1835When C<S> can match, it is a better match than when only C<T> can match.
1836
1837Ordering of two matches for C<S> is the same as for C<S>. Similar for
1838two matches for C<T>.
1839
1840=item C<S{REPEAT_COUNT}>
1841
1842Matches as C<SSS...S> (repeated as many times as necessary).
1843
1844=item C<S{min,max}>
1845
1846Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
1847
1848=item C<S{min,max}?>
1849
1850Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
1851
1852=item C<S?>, C<S*>, C<S+>
1853
1854Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
1855
1856=item C<S??>, C<S*?>, C<S+?>
1857
1858Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
1859
c47ff5f1 1860=item C<< (?>S) >>
35a734be
IZ
1861
1862Matches the best match for C<S> and only that.
1863
1864=item C<(?=S)>, C<(?<=S)>
1865
1866Only the best match for C<S> is considered. (This is important only if
1867C<S> has capturing parentheses, and backreferences are used somewhere
1868else in the whole regular expression.)
1869
1870=item C<(?!S)>, C<(?<!S)>
1871
1872For this grouping operator there is no need to describe the ordering, since
1873only whether or not C<S> can match is important.
1874
6bda09f9 1875=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
1876
1877The ordering is the same as for the regular expression which is
6bda09f9 1878the result of EXPR, or the pattern contained by capture buffer PARNO.
35a734be
IZ
1879
1880=item C<(?(condition)yes-pattern|no-pattern)>
1881
1882Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
1883already determined. The ordering of the matches is the same as for the
1884chosen subexpression.
1885
1886=back
1887
1888The above recipes describe the ordering of matches I<at a given position>.
1889One more rule is needed to understand how a match is determined for the
1890whole regular expression: a match at an earlier position is always better
1891than a match at a later position.
1892
c84d73f1
IZ
1893=head2 Creating custom RE engines
1894
1895Overloaded constants (see L<overload>) provide a simple way to extend
1896the functionality of the RE engine.
1897
1898Suppose that we want to enable a new RE escape-sequence C<\Y|> which
6b0ac556 1899matches at boundary between whitespace characters and non-whitespace
c84d73f1
IZ
1900characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
1901at these positions, so we want to have each C<\Y|> in the place of the
1902more complicated version. We can create a module C<customre> to do
1903this:
1904
1905 package customre;
1906 use overload;
1907
1908 sub import {
1909 shift;
1910 die "No argument to customre::import allowed" if @_;
1911 overload::constant 'qr' => \&convert;
1912 }
1913
1914 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
1915
580a9fe1
RGS
1916 # We must also take care of not escaping the legitimate \\Y|
1917 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 1918 my %rules = ( '\\' => '\\\\',
c84d73f1
IZ
1919 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
1920 sub convert {
1921 my $re = shift;
5d458dd8 1922 $re =~ s{
c84d73f1
IZ
1923 \\ ( \\ | Y . )
1924 }
5d458dd8 1925 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
1926 return $re;
1927 }
1928
1929Now C<use customre> enables the new escape in constant regular
1930expressions, i.e., those without any runtime variable interpolations.
1931As documented in L<overload>, this conversion will work only over
1932literal parts of regular expressions. For C<\Y|$re\Y|> the variable
1933part of this regular expression needs to be converted explicitly
1934(but only if the special meaning of C<\Y|> should be enabled inside $re):
1935
1936 use customre;
1937 $re = <>;
1938 chomp $re;
1939 $re = customre::convert $re;
1940 /\Y|$re\Y|/;
1941
19799a22
GS
1942=head1 BUGS
1943
9da458fc
IZ
1944This document varies from difficult to understand to completely
1945and utterly opaque. The wandering prose riddled with jargon is
1946hard to fathom in several places.
1947
1948This document needs a rewrite that separates the tutorial content
1949from the reference content.
19799a22
GS
1950
1951=head1 SEE ALSO
9fa51da4 1952
91e0c79e
MJD
1953L<perlrequick>.
1954
1955L<perlretut>.
1956
9b599b2a
GS
1957L<perlop/"Regexp Quote-Like Operators">.
1958
1e66bd83
PP
1959L<perlop/"Gory details of parsing quoted constructs">.
1960
14218588
GS
1961L<perlfaq6>.
1962
9b599b2a
GS
1963L<perlfunc/pos>.
1964
1965L<perllocale>.
1966
fb55449c
JH
1967L<perlebcdic>.
1968
14218588
GS
1969I<Mastering Regular Expressions> by Jeffrey Friedl, published
1970by O'Reilly and Associates.