This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
was RE: Perl_die() / Perl_croak()
[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
91e0c79e
MJD
8This page describes the syntax of regular expressions in Perl.
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
GS
21are listed below. Modifiers that alter the way a regular expression
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.
248 \1 Backreference to a a specific group.
249 '1' may actually be any positive integer
250 \k<name> Named backreference
251 \N{name} Named unicode character, or unicode escape.
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 $pattern = "foobar";
610 if ( /$pattern/i ) { }
611
612 # more flexible:
613
614 $pattern = "(?i)foobar";
615 if ( /$pattern/ ) { }
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
GS
642Any letters between C<?> and C<:> act as flags modifiers as with
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;
762 m<
763 (?{ $cnt = 0 }) # Initialize $cnt.
764 (
765 a
766 (?{
767 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
768 })
769 )*
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
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
8988a1bb
DD
817Because perl's regex engine is not currently re-entrant, interpolated
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
8988a1bb
DD
861Because perl's regex engine is not currently re-entrant, delayed
862code may not invoke the regex engine either directly with C<m//> or C<s///>),
863or indirectly with functions such as C<split>.
864
6bda09f9 865Recursing deeper than 50 times without consuming any input string will
61528107 866result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
867changing it requires a custom build.
868
894be9b7
YO
869=item C<(?PARNO)> C<(?R)> C<(?0)>
870X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)>
6bda09f9
YO
871X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
872
81714fb9
YO
873Similar to C<(??{ code })> except it does not involve compiling any code,
874instead it treats the contents of a capture buffer as an independent
61528107 875pattern that must match at the current position. Capture buffers
81714fb9 876contained by the pattern will have the value as determined by the
6bda09f9
YO
877outermost recursion.
878
894be9b7
YO
879PARNO is a sequence of digits (not starting with 0) whose value reflects
880the paren-number of the capture buffer to recurse to. C<(?R)> recurses to
881the beginning of the whole pattern. C<(?0)> is an alternate syntax for
882C<(?R)>.
6bda09f9 883
81714fb9 884The following pattern matches a function foo() which may contain
f145b7e9 885balanced parentheses as the argument.
6bda09f9
YO
886
887 $re = qr{ ( # paren group 1 (full function)
81714fb9 888 foo
6bda09f9
YO
889 ( # paren group 2 (parens)
890 \(
891 ( # paren group 3 (contents of parens)
892 (?:
893 (?> [^()]+ ) # Non-parens without backtracking
894 |
895 (?2) # Recurse to start of paren group 2
896 )*
897 )
898 \)
899 )
900 )
901 }x;
902
903If the pattern was used as follows
904
905 'foo(bar(baz)+baz(bop))'=~/$re/
906 and print "\$1 = $1\n",
907 "\$2 = $2\n",
908 "\$3 = $3\n";
909
910the output produced should be the following:
911
912 $1 = foo(bar(baz)+baz(bop))
913 $2 = (bar(baz)+baz(bop))
81714fb9 914 $3 = bar(baz)+baz(bop)
6bda09f9 915
81714fb9 916If there is no corresponding capture buffer defined, then it is a
61528107 917fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 918string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
919into perl, so changing it requires a custom build.
920
81714fb9 921B<Note> that this pattern does not behave the same way as the equivalent
6bda09f9
YO
922PCRE or Python construct of the same form. In perl you can backtrack into
923a recursed group, in PCRE and Python the recursed into group is treated
81714fb9
YO
924as atomic. Also, constructs like (?i:(?1)) or (?:(?i)(?1)) do not affect
925the pattern being recursed into.
6bda09f9 926
894be9b7
YO
927=item C<(?&NAME)>
928X<(?&NAME)>
929
930Recurse to a named subpattern. Identical to (?PARNO) except that the
931parenthesis to recurse to is determined by name. If multiple parens have
932the same name, then it recurses to the leftmost.
933
934It is an error to refer to a name that is not declared somewhere in the
935pattern.
936
e2e6a0f1
YO
937=item C<(?(condition)yes-pattern|no-pattern)>
938X<(?()>
286f584a 939
e2e6a0f1 940=item C<(?(condition)yes-pattern)>
286f584a 941
e2e6a0f1
YO
942Conditional expression. C<(condition)> should be either an integer in
943parentheses (which is valid if the corresponding pair of parentheses
944matched), a look-ahead/look-behind/evaluate zero-width assertion, a
945name in angle brackets or single quotes (which is valid if a buffer
946with the given name matched), or the special symbol (R) (true when
947evaluated inside of recursion or eval). Additionally the R may be
948followed by a number, (which will be true when evaluated when recursing
949inside of the appropriate group), or by C<&NAME>, in which case it will
950be true only when evaluated during recursion in the named group.
951
952Here's a summary of the possible predicates:
953
954=over 4
955
956=item (1) (2) ...
957
958Checks if the numbered capturing buffer has matched something.
959
960=item (<NAME>) ('NAME')
961
962Checks if a buffer with the given name has matched something.
963
964=item (?{ CODE })
965
966Treats the code block as the condition.
967
968=item (R)
969
970Checks if the expression has been evaluated inside of recursion.
971
972=item (R1) (R2) ...
973
974Checks if the expression has been evaluated while executing directly
975inside of the n-th capture group. This check is the regex equivalent of
976
977 if ((caller(0))[3] eq 'subname') { ... }
978
979In other words, it does not check the full recursion stack.
980
981=item (R&NAME)
982
983Similar to C<(R1)>, this predicate checks to see if we're executing
984directly inside of the leftmost group with a given name (this is the same
985logic used by C<(?&NAME)> to disambiguate). It does not check the full
986stack, but only the name of the innermost active recursion.
987
988=item (DEFINE)
989
990In this case, the yes-pattern is never directly executed, and no
991no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
992See below for details.
993
994=back
995
996For example:
997
998 m{ ( \( )?
999 [^()]+
1000 (?(1) \) )
1001 }x
1002
1003matches a chunk of non-parentheses, possibly included in parentheses
1004themselves.
1005
1006A special form is the C<(DEFINE)> predicate, which never executes directly
1007its yes-pattern, and does not allow a no-pattern. This allows to define
1008subpatterns which will be executed only by using the recursion mechanism.
1009This way, you can define a set of regular expression rules that can be
1010bundled into any pattern you choose.
1011
1012It is recommended that for this usage you put the DEFINE block at the
1013end of the pattern, and that you name any subpatterns defined within it.
1014
1015Also, it's worth noting that patterns defined this way probably will
1016not be as efficient, as the optimiser is not very clever about
1017handling them.
1018
1019An example of how this might be used is as follows:
1020
1021 /(?<NAME>(&NAME_PAT))(?<ADDR>(&ADDRESS_PAT))
1022 (?(DEFINE)
1023 (<NAME_PAT>....)
1024 (<ADRESS_PAT>....)
1025 )/x
1026
1027Note that capture buffers matched inside of recursion are not accessible
1028after the recursion returns, so the extra layer of capturing buffers are
1029necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1030C<$+{NAME}> would be.
286f584a 1031
c47ff5f1 1032=item C<< (?>pattern) >>
6bda09f9 1033X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1034
19799a22
GS
1035An "independent" subexpression, one which matches the substring
1036that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1037position, and it matches I<nothing other than this substring>. This
19799a22
GS
1038construct is useful for optimizations of what would otherwise be
1039"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1040It may also be useful in places where the "grab all you can, and do not
1041give anything back" semantic is desirable.
19799a22 1042
c47ff5f1 1043For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1044(anchored at the beginning of string, as above) will match I<all>
1045characters C<a> at the beginning of string, leaving no C<a> for
1046C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1047since the match of the subgroup C<a*> is influenced by the following
1048group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1049C<a*ab> will match fewer characters than a standalone C<a*>, since
1050this makes the tail match.
1051
c47ff5f1 1052An effect similar to C<< (?>pattern) >> may be achieved by writing
19799a22
GS
1053C<(?=(pattern))\1>. This matches the same substring as a standalone
1054C<a+>, and the following C<\1> eats the matched string; it therefore
c47ff5f1 1055makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1056(The difference between these two constructs is that the second one
1057uses a capturing group, thus shifting ordinals of backreferences
1058in the rest of a regular expression.)
1059
1060Consider this pattern:
c277df42 1061
871b0233 1062 m{ \(
e2e6a0f1
YO
1063 (
1064 [^()]+ # x+
1065 |
871b0233
IZ
1066 \( [^()]* \)
1067 )+
e2e6a0f1 1068 \)
871b0233 1069 }x
5a964f20 1070
19799a22
GS
1071That will efficiently match a nonempty group with matching parentheses
1072two levels deep or less. However, if there is no such group, it
1073will take virtually forever on a long string. That's because there
1074are so many different ways to split a long string into several
1075substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1076to a subpattern of the above pattern. Consider how the pattern
1077above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1078seconds, but that each extra letter doubles this time. This
1079exponential performance will make it appear that your program has
14218588 1080hung. However, a tiny change to this pattern
5a964f20 1081
e2e6a0f1
YO
1082 m{ \(
1083 (
1084 (?> [^()]+ ) # change x+ above to (?> x+ )
1085 |
871b0233
IZ
1086 \( [^()]* \)
1087 )+
e2e6a0f1 1088 \)
871b0233 1089 }x
c277df42 1090
c47ff5f1 1091which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1092this yourself would be a productive exercise), but finishes in a fourth
1093the time when used on a similar string with 1000000 C<a>s. Be aware,
1094however, that this pattern currently triggers a warning message under
9f1b1f2d 1095the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1096C<"matches null string many times in regex">.
c277df42 1097
c47ff5f1 1098On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1099effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1100This was only 4 times slower on a string with 1000000 C<a>s.
1101
9da458fc
IZ
1102The "grab all you can, and do not give anything back" semantic is desirable
1103in many situations where on the first sight a simple C<()*> looks like
1104the correct solution. Suppose we parse text with comments being delimited
1105by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1106its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1107the comment delimiter, because it may "give up" some whitespace if
1108the remainder of the pattern can be made to match that way. The correct
1109answer is either one of these:
1110
1111 (?>#[ \t]*)
1112 #[ \t]*(?![ \t])
1113
1114For example, to grab non-empty comments into $1, one should use either
1115one of these:
1116
1117 / (?> \# [ \t]* ) ( .+ ) /x;
1118 / \# [ \t]* ( [^ \t] .* ) /x;
1119
1120Which one you pick depends on which of these expressions better reflects
1121the above specification of comments.
1122
6bda09f9
YO
1123In some literature this construct is called "atomic matching" or
1124"possessive matching".
1125
b9b4dddf
YO
1126Possessive quantifiers are equivalent to putting the item they are applied
1127to inside of one of these constructs. The following equivalences apply:
1128
1129 Quantifier Form Bracketing Form
1130 --------------- ---------------
1131 PAT*+ (?>PAT*)
1132 PAT++ (?>PAT+)
1133 PAT?+ (?>PAT?)
1134 PAT{min,max}+ (?>PAT{min,max})
1135
e2e6a0f1
YO
1136=back
1137
1138=head2 Special Backtracking Control Verbs
1139
1140B<WARNING:> These patterns are experimental and subject to change or
1141removal in a future version of perl. Their usage in production code should
1142be noted to avoid problems during upgrades.
1143
1144These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1145otherwise stated the ARG argument is optional; in some cases, it is
1146forbidden.
1147
1148Any pattern containing a special backtracking verb that allows an argument
1149has the special behaviour that when executed it sets the current packages'
1150C<$REGERROR> variable. In this case, the following rules apply:
1151
1152On failure, this variable will be set to the ARG value of the verb
1153pattern, if the verb was involved in the failure of the match. If the ARG
1154part of the pattern was omitted, then C<$REGERROR> will be set to TRUE.
1155
1156On a successful match this variable will be set to FALSE.
1157
1158B<NOTE:> C<$REGERROR> is not a magic variable in the same sense than
1159C<$1> and most other regex related variables. It is not local to a
1160scope, nor readonly but instead a volatile package variable similar to
1161C<$AUTOLOAD>. Use C<local> to localize changes to it to a specific scope
1162if necessary.
1163
1164If a pattern does not contain a special backtracking verb that allows an
1165argument, then C<$REGERROR> is not touched at all.
1166
1167=over 4
1168
1169=item Verbs that take an argument
1170
1171=over 4
1172
1173=item C<(*NOMATCH)> C<(*NOMATCH:NAME)>
1174X<(*NOMATCH)> X<(*NOMATCH:NAME)>
54612592
YO
1175
1176This zero-width pattern commits the match at the current point, preventing
e2e6a0f1
YO
1177the engine from backtracking on failure to the left of the this point.
1178Consider the pattern C<A (*NOMATCH) B>, where A and B are complex patterns.
1179Until the C<(*NOMATCH)> is reached, A may backtrack as necessary to match.
54612592
YO
1180Once it is reached, matching continues in B, which may also backtrack as
1181necessary; however, should B not match, then no further backtracking will
1182take place, and the pattern will fail outright at that starting position.
1183
1184The following example counts all the possible matching strings in a
1185pattern (without actually matching any of them).
1186
e2e6a0f1 1187 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1188 print "Count=$count\n";
1189
1190which produces:
1191
1192 aaab
1193 aaa
1194 aa
1195 a
1196 aab
1197 aa
1198 a
1199 ab
1200 a
1201 Count=9
1202
e2e6a0f1 1203If we add a C<(*NOMATCH)> before the count like the following
54612592 1204
e2e6a0f1 1205 'aaab' =~ /a+b?(*NOMATCH)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1206 print "Count=$count\n";
1207
1208we prevent backtracking and find the count of the longest matching
1209at each matching startpoint like so:
1210
1211 aaab
1212 aab
1213 ab
1214 Count=3
1215
e2e6a0f1 1216Any number of C<(*NOMATCH)> assertions may be used in a pattern.
54612592
YO
1217
1218See also C<< (?>pattern) >> and possessive quantifiers for other
1219ways to control backtracking.
1220
e2e6a0f1
YO
1221=item C<(*MARK)> C<(*MARK:NAME)>
1222X<(*MARK)>
1223
1224This zero-width pattern can be used to mark the point in a string
1225reached when a certain part of the pattern has been successfully
1226matched. This mark may be given a name. A later C<(*CUT)> pattern
1227will then cut at that point if backtracked into on failure. Any
1228number of (*MARK) patterns are allowed, and the NAME portion is
1229optional and may be duplicated.
1230
1231See C<*CUT> for more detail.
1232
1233=item C<(*CUT)> C<(*CUT:NAME)>
1234X<(*CUT)>
1235
1236This zero-width pattern is similar to C<(*NOMATCH)>, except that on
1237failure it also signifies that whatever text that was matched leading up
1238to the C<(*CUT)> pattern being executed cannot be part of a match, I<even
1239if started from a later point>. This effectively means that the regex
1240engine moves forward to this position on failure and tries to match
1241again, (assuming that there is sufficient room to match).
1242
1243The name of the C<(*CUT:NAME)> pattern has special significance. If a
1244C<(*MARK:NAME)> was encountered while matching, then it is the position
1245where that pattern was executed that is used for the "cut point" in the
1246string. If no mark of that name was encountered, then the cut is done at
1247the point where the C<(*CUT)> was. Similarly if no NAME is specified in
1248the C<(*CUT)>, and if a C<(*MARK)> with any name (or none) is encountered,
1249then that C<(*MARK)>'s cursor point will be used. If the C<(*CUT)> is not
1250preceded by a C<(*MARK)>, then the cut point is where the string was when
1251the C<(*CUT)> was encountered.
1252
1253Compare the following to the examples in C<(*NOMATCH)>, note the string
24b23f37
YO
1254is twice as long:
1255
e2e6a0f1 1256 'aaabaaab' =~ /a+b?(*CUT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1257 print "Count=$count\n";
1258
1259outputs
1260
1261 aaab
1262 aaab
1263 Count=2
1264
e2e6a0f1
YO
1265Once the 'aaab' at the start of the string has matched, and the C<(*CUT)>
1266executed, the next startpoint will be where the cursor was when the
1267C<(*CUT)> was executed.
24b23f37 1268
e2e6a0f1
YO
1269=item C<(*COMMIT)>
1270X<(*COMMIT)>
24b23f37 1271
e2e6a0f1 1272This zero-width pattern is similar to C<(*CUT)> except that it causes
24b23f37
YO
1273the match to fail outright. No attempts to match will occur again.
1274
e2e6a0f1 1275 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1276 print "Count=$count\n";
1277
1278outputs
1279
1280 aaab
1281 Count=1
1282
e2e6a0f1
YO
1283In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1284does not match, the regex engine will not try any further matching on the
1285rest of the string.
c277df42 1286
e2e6a0f1 1287=back
9af228c6 1288
e2e6a0f1 1289=item Verbs without an argument
9af228c6
YO
1290
1291=over 4
1292
e2e6a0f1
YO
1293=item C<(*FAIL)> C<(*F)>
1294X<(*FAIL)> X<(*F)>
9af228c6 1295
e2e6a0f1
YO
1296This pattern matches nothing and always fails. It can be used to force the
1297engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1298fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1299
e2e6a0f1 1300It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1301
e2e6a0f1
YO
1302=item C<(*ACCEPT)>
1303X<(*ACCEPT)>
9af228c6 1304
e2e6a0f1
YO
1305B<WARNING:> This feature is highly experimental. It is not recommended
1306for production code.
9af228c6 1307
e2e6a0f1
YO
1308This pattern matches nothing and causes the end of successful matching at
1309the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1310whether there is actually more to match in the string. When inside of a
1311nested pattern, such as recursion or a dynamically generated subbpattern
1312via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1313
e2e6a0f1
YO
1314If the C<(*ACCEPT)> is inside of capturing buffers then the buffers are
1315marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1316For instance:
9af228c6 1317
e2e6a0f1 1318 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1319
e2e6a0f1
YO
1320will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
1321be set. If another branch in the inner parens were matched, such as in the
1322string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1323
1324=back
c277df42 1325
a0d0e21e
LW
1326=back
1327
c07a80fd 1328=head2 Backtracking
d74e8afc 1329X<backtrack> X<backtracking>
c07a80fd 1330
35a734be
IZ
1331NOTE: This section presents an abstract approximation of regular
1332expression behavior. For a more rigorous (and complicated) view of
1333the rules involved in selecting a match among possible alternatives,
1334see L<Combining pieces together>.
1335
c277df42 1336A fundamental feature of regular expression matching involves the
5a964f20 1337notion called I<backtracking>, which is currently used (when needed)
c277df42 1338by all regular expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1339C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1340internally, but the general principle outlined here is valid.
c07a80fd
PP
1341
1342For a regular expression to match, the I<entire> regular expression must
1343match, not just part of it. So if the beginning of a pattern containing a
1344quantifier succeeds in a way that causes later parts in the pattern to
1345fail, the matching engine backs up and recalculates the beginning
1346part--that's why it's called backtracking.
1347
1348Here is an example of backtracking: Let's say you want to find the
1349word following "foo" in the string "Food is on the foo table.":
1350
1351 $_ = "Food is on the foo table.";
1352 if ( /\b(foo)\s+(\w+)/i ) {
1353 print "$2 follows $1.\n";
1354 }
1355
1356When the match runs, the first part of the regular expression (C<\b(foo)>)
1357finds a possible match right at the beginning of the string, and loads up
1358$1 with "Foo". However, as soon as the matching engine sees that there's
1359no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1360mistake and starts over again one character after where it had the
c07a80fd
PP
1361tentative match. This time it goes all the way until the next occurrence
1362of "foo". The complete regular expression matches this time, and you get
1363the expected output of "table follows foo."
1364
1365Sometimes minimal matching can help a lot. Imagine you'd like to match
1366everything between "foo" and "bar". Initially, you write something
1367like this:
1368
1369 $_ = "The food is under the bar in the barn.";
1370 if ( /foo(.*)bar/ ) {
1371 print "got <$1>\n";
1372 }
1373
1374Which perhaps unexpectedly yields:
1375
1376 got <d is under the bar in the >
1377
1378That's because C<.*> was greedy, so you get everything between the
14218588 1379I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1380to use minimal matching to make sure you get the text between a "foo"
1381and the first "bar" thereafter.
1382
1383 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1384 got <d is under the >
1385
1386Here's another example: let's say you'd like to match a number at the end
b6e13d97 1387of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
1388So you write this:
1389
1390 $_ = "I have 2 numbers: 53147";
1391 if ( /(.*)(\d*)/ ) { # Wrong!
1392 print "Beginning is <$1>, number is <$2>.\n";
1393 }
1394
1395That won't work at all, because C<.*> was greedy and gobbled up the
1396whole string. As C<\d*> can match on an empty string the complete
1397regular expression matched successfully.
1398
8e1088bc 1399 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
1400
1401Here are some variants, most of which don't work:
1402
1403 $_ = "I have 2 numbers: 53147";
1404 @pats = qw{
1405 (.*)(\d*)
1406 (.*)(\d+)
1407 (.*?)(\d*)
1408 (.*?)(\d+)
1409 (.*)(\d+)$
1410 (.*?)(\d+)$
1411 (.*)\b(\d+)$
1412 (.*\D)(\d+)$
1413 };
1414
1415 for $pat (@pats) {
1416 printf "%-12s ", $pat;
1417 if ( /$pat/ ) {
1418 print "<$1> <$2>\n";
1419 } else {
1420 print "FAIL\n";
1421 }
1422 }
1423
1424That will print out:
1425
1426 (.*)(\d*) <I have 2 numbers: 53147> <>
1427 (.*)(\d+) <I have 2 numbers: 5314> <7>
1428 (.*?)(\d*) <> <>
1429 (.*?)(\d+) <I have > <2>
1430 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1431 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1432 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1433 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1434
1435As you see, this can be a bit tricky. It's important to realize that a
1436regular expression is merely a set of assertions that gives a definition
1437of success. There may be 0, 1, or several different ways that the
1438definition might succeed against a particular string. And if there are
5a964f20
TC
1439multiple ways it might succeed, you need to understand backtracking to
1440know which variety of success you will achieve.
c07a80fd 1441
19799a22 1442When using look-ahead assertions and negations, this can all get even
8b19b778 1443trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
1444followed by "123". You might try to write that as
1445
871b0233
IZ
1446 $_ = "ABC123";
1447 if ( /^\D*(?!123)/ ) { # Wrong!
1448 print "Yup, no 123 in $_\n";
1449 }
c07a80fd
PP
1450
1451But that isn't going to match; at least, not the way you're hoping. It
1452claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1453why that pattern matches, contrary to popular expectations:
c07a80fd 1454
4358a253
SS
1455 $x = 'ABC123';
1456 $y = 'ABC445';
c07a80fd 1457
4358a253
SS
1458 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1459 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1460
4358a253
SS
1461 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1462 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
1463
1464This prints
1465
1466 2: got ABC
1467 3: got AB
1468 4: got ABC
1469
5f05dabc 1470You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
1471general purpose version of test 1. The important difference between
1472them is that test 3 contains a quantifier (C<\D*>) and so can use
1473backtracking, whereas test 1 will not. What's happening is
1474that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1475non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1476let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1477fail.
14218588 1478
c07a80fd 1479The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1480try to match C<(?!123> with "123", which fails. But because
c07a80fd
PP
1481a quantifier (C<\D*>) has been used in the regular expression, the
1482search engine can backtrack and retry the match differently
54310121 1483in the hope of matching the complete regular expression.
c07a80fd 1484
5a964f20
TC
1485The pattern really, I<really> wants to succeed, so it uses the
1486standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1487time. Now there's indeed something following "AB" that is not
14218588 1488"123". It's "C123", which suffices.
c07a80fd 1489
14218588
GS
1490We can deal with this by using both an assertion and a negation.
1491We'll say that the first part in $1 must be followed both by a digit
1492and by something that's not "123". Remember that the look-aheads
1493are zero-width expressions--they only look, but don't consume any
1494of the string in their match. So rewriting this way produces what
c07a80fd
PP
1495you'd expect; that is, case 5 will fail, but case 6 succeeds:
1496
4358a253
SS
1497 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1498 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
1499
1500 6: got ABC
1501
5a964f20 1502In other words, the two zero-width assertions next to each other work as though
19799a22 1503they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
1504matches only if you're at the beginning of the line AND the end of the
1505line simultaneously. The deeper underlying truth is that juxtaposition in
1506regular expressions always means AND, except when you write an explicit OR
1507using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1508although the attempted matches are made at different positions because "a"
1509is not a zero-width assertion, but a one-width assertion.
1510
19799a22 1511B<WARNING>: particularly complicated regular expressions can take
14218588 1512exponential time to solve because of the immense number of possible
9da458fc
IZ
1513ways they can use backtracking to try match. For example, without
1514internal optimizations done by the regular expression engine, this will
1515take a painfully long time to run:
c07a80fd 1516
e1901655
IZ
1517 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1518
1519And if you used C<*>'s in the internal groups instead of limiting them
1520to 0 through 5 matches, then it would take forever--or until you ran
1521out of stack space. Moreover, these internal optimizations are not
1522always applicable. For example, if you put C<{0,5}> instead of C<*>
1523on the external group, no current optimization is applicable, and the
1524match takes a long time to finish.
c07a80fd 1525
9da458fc
IZ
1526A powerful tool for optimizing such beasts is what is known as an
1527"independent group",
c47ff5f1 1528which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1529zero-length look-ahead/look-behind assertions will not backtrack to make
14218588
GS
1530the tail match, since they are in "logical" context: only
1531whether they match is considered relevant. For an example
9da458fc 1532where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1533following match, see L<C<< (?>pattern) >>>.
c277df42 1534
a0d0e21e 1535=head2 Version 8 Regular Expressions
d74e8afc 1536X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1537
5a964f20 1538In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1539routines, here are the pattern-matching rules not described above.
1540
54310121 1541Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1542with a special meaning described here or above. You can cause
5a964f20 1543characters that normally function as metacharacters to be interpreted
5f05dabc 1544literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
a0d0e21e
LW
1545character; "\\" matches a "\"). A series of characters matches that
1546series of characters in the target string, so the pattern C<blurfl>
1547would match "blurfl" in the target string.
1548
1549You can specify a character class, by enclosing a list of characters
5a964f20 1550in C<[]>, which will match any one character from the list. If the
a0d0e21e 1551first character after the "[" is "^", the class matches any character not
14218588 1552in the list. Within a list, the "-" character specifies a
5a964f20 1553range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1554inclusive. If you want either "-" or "]" itself to be a member of a
1555class, put it at the start of the list (possibly after a "^"), or
1556escape it with a backslash. "-" is also taken literally when it is
1557at the end of the list, just before the closing "]". (The
84850974
DD
1558following all specify the same class of three characters: C<[-az]>,
1559C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
fb55449c
JH
1560specifies a class containing twenty-six characters, even on EBCDIC
1561based coded character sets.) Also, if you try to use the character
1562classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1563a range, that's not a range, the "-" is understood literally.
a0d0e21e 1564
8ada0baa
JH
1565Note also that the whole range idea is rather unportable between
1566character sets--and even within character sets they may cause results
1567you probably didn't expect. A sound principle is to use only ranges
1568that begin from and end at either alphabets of equal case ([a-e],
1569[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1570spell out the character sets in full.
1571
54310121 1572Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1573used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1574"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
fb55449c
JH
1575of octal digits, matches the character whose coded character set value
1576is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
1577matches the character whose numeric value is I<nn>. The expression \cI<x>
1578matches the character control-I<x>. Finally, the "." metacharacter
1579matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1580
1581You can specify a series of alternatives for a pattern using "|" to
1582separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1583or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1584first alternative includes everything from the last pattern delimiter
1585("(", "[", or the beginning of the pattern) up to the first "|", and
1586the last alternative contains everything from the last "|" to the next
14218588
GS
1587pattern delimiter. That's why it's common practice to include
1588alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1589start and end.
1590
5a964f20 1591Alternatives are tried from left to right, so the first
a3cb178b
GS
1592alternative found for which the entire expression matches, is the one that
1593is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1594example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1595part will match, as that is the first alternative tried, and it successfully
1596matches the target string. (This might not seem important, but it is
1597important when you are capturing matched text using parentheses.)
1598
5a964f20 1599Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1600so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1601
14218588
GS
1602Within a pattern, you may designate subpatterns for later reference
1603by enclosing them in parentheses, and you may refer back to the
1604I<n>th subpattern later in the pattern using the metacharacter
1605\I<n>. Subpatterns are numbered based on the left to right order
1606of their opening parenthesis. A backreference matches whatever
1607actually matched the subpattern in the string being examined, not
1608the rules for that subpattern. Therefore, C<(0|0x)\d*\s\1\d*> will
1609match "0x1234 0x4321", but not "0x1234 01234", because subpattern
16101 matched "0x", even though the rule C<0|0x> could potentially match
1611the leading 0 in the second number.
cb1a09d0 1612
19799a22 1613=head2 Warning on \1 vs $1
cb1a09d0 1614
5a964f20 1615Some people get too used to writing things like:
cb1a09d0
AD
1616
1617 $pattern =~ s/(\W)/\\\1/g;
1618
1619This is grandfathered for the RHS of a substitute to avoid shocking the
1620B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1621PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1622the usual double-quoted string means a control-A. The customary Unix
1623meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1624of doing that, you get yourself into trouble if you then add an C</e>
1625modifier.
1626
5a964f20 1627 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1628
1629Or if you try to do
1630
1631 s/(\d+)/\1000/;
1632
1633You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1634C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1635with the operation of matching a backreference. Certainly they mean two
1636different things on the I<left> side of the C<s///>.
9fa51da4 1637
c84d73f1
IZ
1638=head2 Repeated patterns matching zero-length substring
1639
19799a22 1640B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1641
1642Regular expressions provide a terse and powerful programming language. As
1643with most other power tools, power comes together with the ability
1644to wreak havoc.
1645
1646A common abuse of this power stems from the ability to make infinite
628afcb5 1647loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1648
1649 'foo' =~ m{ ( o? )* }x;
1650
1651The C<o?> can match at the beginning of C<'foo'>, and since the position
1652in the string is not moved by the match, C<o?> would match again and again
14218588 1653because of the C<*> modifier. Another common way to create a similar cycle
c84d73f1
IZ
1654is with the looping modifier C<//g>:
1655
1656 @matches = ( 'foo' =~ m{ o? }xg );
1657
1658or
1659
1660 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1661
1662or the loop implied by split().
1663
1664However, long experience has shown that many programming tasks may
14218588
GS
1665be significantly simplified by using repeated subexpressions that
1666may match zero-length substrings. Here's a simple example being:
c84d73f1
IZ
1667
1668 @chars = split //, $string; # // is not magic in split
1669 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1670
9da458fc 1671Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1
IZ
1672the infinite loop>. The rules for this are different for lower-level
1673loops given by the greedy modifiers C<*+{}>, and for higher-level
1674ones like the C</g> modifier or split() operator.
1675
19799a22
GS
1676The lower-level loops are I<interrupted> (that is, the loop is
1677broken) when Perl detects that a repeated expression matched a
1678zero-length substring. Thus
c84d73f1
IZ
1679
1680 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1681
1682is made equivalent to
1683
1684 m{ (?: NON_ZERO_LENGTH )*
1685 |
1686 (?: ZERO_LENGTH )?
1687 }x;
1688
1689The higher level-loops preserve an additional state between iterations:
1690whether the last match was zero-length. To break the loop, the following
1691match after a zero-length match is prohibited to have a length of zero.
1692This prohibition interacts with backtracking (see L<"Backtracking">),
1693and so the I<second best> match is chosen if the I<best> match is of
1694zero length.
1695
19799a22 1696For example:
c84d73f1
IZ
1697
1698 $_ = 'bar';
1699 s/\w??/<$&>/g;
1700
20fb949f 1701results in C<< <><b><><a><><r><> >>. At each position of the string the best
c84d73f1
IZ
1702match given by non-greedy C<??> is the zero-length match, and the I<second
1703best> match is what is matched by C<\w>. Thus zero-length matches
1704alternate with one-character-long matches.
1705
1706Similarly, for repeated C<m/()/g> the second-best match is the match at the
1707position one notch further in the string.
1708
19799a22 1709The additional state of being I<matched with zero-length> is associated with
c84d73f1 1710the matched string, and is reset by each assignment to pos().
9da458fc
IZ
1711Zero-length matches at the end of the previous match are ignored
1712during C<split>.
c84d73f1 1713
35a734be
IZ
1714=head2 Combining pieces together
1715
1716Each of the elementary pieces of regular expressions which were described
1717before (such as C<ab> or C<\Z>) could match at most one substring
1718at the given position of the input string. However, in a typical regular
1719expression these elementary pieces are combined into more complicated
1720patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1721(in these examples C<S> and C<T> are regular subexpressions).
1722
1723Such combinations can include alternatives, leading to a problem of choice:
1724if we match a regular expression C<a|ab> against C<"abc">, will it match
1725substring C<"a"> or C<"ab">? One way to describe which substring is
1726actually matched is the concept of backtracking (see L<"Backtracking">).
1727However, this description is too low-level and makes you think
1728in terms of a particular implementation.
1729
1730Another description starts with notions of "better"/"worse". All the
1731substrings which may be matched by the given regular expression can be
1732sorted from the "best" match to the "worst" match, and it is the "best"
1733match which is chosen. This substitutes the question of "what is chosen?"
1734by the question of "which matches are better, and which are worse?".
1735
1736Again, for elementary pieces there is no such question, since at most
1737one match at a given position is possible. This section describes the
1738notion of better/worse for combining operators. In the description
1739below C<S> and C<T> are regular subexpressions.
1740
13a2d996 1741=over 4
35a734be
IZ
1742
1743=item C<ST>
1744
1745Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
1746substrings which can be matched by C<S>, C<B> and C<B'> are substrings
1747which can be matched by C<T>.
1748
1749If C<A> is better match for C<S> than C<A'>, C<AB> is a better
1750match than C<A'B'>.
1751
1752If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
1753C<B> is better match for C<T> than C<B'>.
1754
1755=item C<S|T>
1756
1757When C<S> can match, it is a better match than when only C<T> can match.
1758
1759Ordering of two matches for C<S> is the same as for C<S>. Similar for
1760two matches for C<T>.
1761
1762=item C<S{REPEAT_COUNT}>
1763
1764Matches as C<SSS...S> (repeated as many times as necessary).
1765
1766=item C<S{min,max}>
1767
1768Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
1769
1770=item C<S{min,max}?>
1771
1772Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
1773
1774=item C<S?>, C<S*>, C<S+>
1775
1776Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
1777
1778=item C<S??>, C<S*?>, C<S+?>
1779
1780Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
1781
c47ff5f1 1782=item C<< (?>S) >>
35a734be
IZ
1783
1784Matches the best match for C<S> and only that.
1785
1786=item C<(?=S)>, C<(?<=S)>
1787
1788Only the best match for C<S> is considered. (This is important only if
1789C<S> has capturing parentheses, and backreferences are used somewhere
1790else in the whole regular expression.)
1791
1792=item C<(?!S)>, C<(?<!S)>
1793
1794For this grouping operator there is no need to describe the ordering, since
1795only whether or not C<S> can match is important.
1796
6bda09f9 1797=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
1798
1799The ordering is the same as for the regular expression which is
6bda09f9 1800the result of EXPR, or the pattern contained by capture buffer PARNO.
35a734be
IZ
1801
1802=item C<(?(condition)yes-pattern|no-pattern)>
1803
1804Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
1805already determined. The ordering of the matches is the same as for the
1806chosen subexpression.
1807
1808=back
1809
1810The above recipes describe the ordering of matches I<at a given position>.
1811One more rule is needed to understand how a match is determined for the
1812whole regular expression: a match at an earlier position is always better
1813than a match at a later position.
1814
c84d73f1
IZ
1815=head2 Creating custom RE engines
1816
1817Overloaded constants (see L<overload>) provide a simple way to extend
1818the functionality of the RE engine.
1819
1820Suppose that we want to enable a new RE escape-sequence C<\Y|> which
6b0ac556 1821matches at boundary between whitespace characters and non-whitespace
c84d73f1
IZ
1822characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
1823at these positions, so we want to have each C<\Y|> in the place of the
1824more complicated version. We can create a module C<customre> to do
1825this:
1826
1827 package customre;
1828 use overload;
1829
1830 sub import {
1831 shift;
1832 die "No argument to customre::import allowed" if @_;
1833 overload::constant 'qr' => \&convert;
1834 }
1835
1836 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
1837
580a9fe1
RGS
1838 # We must also take care of not escaping the legitimate \\Y|
1839 # sequence, hence the presence of '\\' in the conversion rules.
141db969 1840 my %rules = ( '\\' => '\\\\',
c84d73f1
IZ
1841 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
1842 sub convert {
1843 my $re = shift;
1844 $re =~ s{
1845 \\ ( \\ | Y . )
1846 }
1847 { $rules{$1} or invalid($re,$1) }sgex;
1848 return $re;
1849 }
1850
1851Now C<use customre> enables the new escape in constant regular
1852expressions, i.e., those without any runtime variable interpolations.
1853As documented in L<overload>, this conversion will work only over
1854literal parts of regular expressions. For C<\Y|$re\Y|> the variable
1855part of this regular expression needs to be converted explicitly
1856(but only if the special meaning of C<\Y|> should be enabled inside $re):
1857
1858 use customre;
1859 $re = <>;
1860 chomp $re;
1861 $re = customre::convert $re;
1862 /\Y|$re\Y|/;
1863
19799a22
GS
1864=head1 BUGS
1865
9da458fc
IZ
1866This document varies from difficult to understand to completely
1867and utterly opaque. The wandering prose riddled with jargon is
1868hard to fathom in several places.
1869
1870This document needs a rewrite that separates the tutorial content
1871from the reference content.
19799a22
GS
1872
1873=head1 SEE ALSO
9fa51da4 1874
91e0c79e
MJD
1875L<perlrequick>.
1876
1877L<perlretut>.
1878
9b599b2a
GS
1879L<perlop/"Regexp Quote-Like Operators">.
1880
1e66bd83
PP
1881L<perlop/"Gory details of parsing quoted constructs">.
1882
14218588
GS
1883L<perlfaq6>.
1884
9b599b2a
GS
1885L<perlfunc/pos>.
1886
1887L<perllocale>.
1888
fb55449c
JH
1889L<perlebcdic>.
1890
14218588
GS
1891I<Mastering Regular Expressions> by Jeffrey Friedl, published
1892by O'Reilly and Associates.