This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlre.pod: corrections for /a
[perl5.git] / pod / perlre.pod
CommitLineData
a0d0e21e 1=head1 NAME
d74e8afc 2X<regular expression> X<regex> X<regexp>
a0d0e21e
LW
3
4perlre - Perl regular expressions
5
6=head1 DESCRIPTION
7
5d458dd8 8This page describes the syntax of regular expressions in Perl.
91e0c79e 9
cc46d5f2 10If you haven't used regular expressions before, a quick-start
91e0c79e
MJD
11introduction is available in L<perlrequick>, and a longer tutorial
12introduction is available in L<perlretut>.
13
14For reference on how regular expressions are used in matching
15operations, plus various examples of the same, see discussions of
16C<m//>, C<s///>, C<qr//> and C<??> in L<perlop/"Regexp Quote-Like
17Operators">.
cb1a09d0 18
0d017f4d
WL
19
20=head2 Modifiers
21
19799a22 22Matching operations can have various modifiers. Modifiers
5a964f20 23that relate to the interpretation of the regular expression inside
19799a22 24are listed below. Modifiers that alter the way a regular expression
5d458dd8 25is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
1e66bd83 26L<perlop/"Gory details of parsing quoted constructs">.
a0d0e21e 27
55497cff 28=over 4
29
54310121 30=item m
d74e8afc 31X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff 32
33Treat string as multiple lines. That is, change "^" and "$" from matching
14218588 34the start or end of the string to matching the start or end of any
7f761169 35line anywhere within the string.
55497cff 36
54310121 37=item s
d74e8afc
ITB
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
55497cff 40
41Treat string as single line. That is, change "." to match any character
19799a22 42whatsoever, even a newline, which normally it would not match.
55497cff 43
34d67d80 44Used together, as C</ms>, they let the "." match any character whatsoever,
fb55449c 45while still allowing "^" and "$" to match, respectively, just after
19799a22 46and just before newlines within the string.
7b8d334a 47
87e95b7f
YO
48=item i
49X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
50X<regular expression, case-insensitive>
51
52Do case-insensitive pattern matching.
53
54If C<use locale> is in effect, the case map is taken from the current
55locale. See L<perllocale>.
56
54310121 57=item x
d74e8afc 58X</x>
55497cff 59
60Extend your pattern's legibility by permitting whitespace and comments.
61
87e95b7f
YO
62=item p
63X</p> X<regex, preserve> X<regexp, preserve>
64
632a1772 65Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
87e95b7f
YO
66${^POSTMATCH} are available for use after matching.
67
e2e6bec7
DN
68=item g and c
69X</g> X</c>
70
71Global matching, and keep the Current position after failed matching.
72Unlike i, m, s and x, these two flags affect the way the regex is used
73rather than the regex itself. See
74L<perlretut/"Using regular expressions in Perl"> for further explanation
75of the g and c modifiers.
76
55497cff 77=back
a0d0e21e
LW
78
79These are usually written as "the C</x> modifier", even though the delimiter
14218588 80in question might not really be a slash. Any of these
a0d0e21e 81modifiers may also be embedded within the regular expression itself using
14218588 82the C<(?...)> construct. See below.
a0d0e21e 83
4633a7c4 84The C</x> modifier itself needs a little more explanation. It tells
7b059540 85the regular expression parser to ignore most whitespace that is neither
55497cff 86backslashed nor within a character class. You can use this to break up
4633a7c4 87your regular expression into (slightly) more readable parts. The C<#>
54310121 88character is also treated as a metacharacter introducing a comment,
55497cff 89just as in ordinary Perl code. This also means that if you want real
14218588 90whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a 91class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
92escape them (using backslashes or C<\Q...\E>) or encode them using octal,
93hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
94making Perl's regular expressions more readable. Note that you have to
95be careful not to include the pattern delimiter in the comment--perl has
96no way of knowing you did not intend to close the pattern early. See
97the C-comment deletion code in L<perlop>. Also note that anything inside
7651b971
KW
98a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
99whether space interpretation within a single multi-character construct. For
100example in C<\x{...}>, regardless of the C</x> modifier, there can be no
9bb1f947 101spaces. Same for a L<quantifier|/Quantifiers> such as C<{3}> or
f9e949fd
KW
102C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<?> and C<:>,
103but can between the C<(> and C<?>. Within any delimiters for such a
104construct, allowed spaces are not affected by C</x>, and depend on the
105construct. For example, C<\x{...}> can't have spaces because hexadecimal
106numbers don't have spaces in them. But, Unicode properties can have spaces, so
107in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
9bb1f947 108L<perluniprops/Properties accessible through \p{} and \P{}>.
d74e8afc 109X</x>
a0d0e21e
LW
110
111=head2 Regular Expressions
112
04838cea
RGS
113=head3 Metacharacters
114
384f06ae 115The patterns used in Perl pattern matching evolved from those supplied in
14218588 116the Version 8 regex routines. (The routines are derived
19799a22
GS
117(distantly) from Henry Spencer's freely redistributable reimplementation
118of the V8 routines.) See L<Version 8 Regular Expressions> for
119details.
a0d0e21e
LW
120
121In particular the following metacharacters have their standard I<egrep>-ish
122meanings:
d74e8afc
ITB
123X<metacharacter>
124X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
125
a0d0e21e 126
f793d64a
KW
127 \ Quote the next metacharacter
128 ^ Match the beginning of the line
129 . Match any character (except newline)
130 $ Match the end of the line (or before newline at the end)
131 | Alternation
132 () Grouping
133 [] Bracketed Character class
a0d0e21e 134
14218588
GS
135By default, the "^" character is guaranteed to match only the
136beginning of the string, the "$" character only the end (or before the
137newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
138assumption that the string contains only one line. Embedded newlines
139will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 140string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
141newline within the string (except if the newline is the last character in
142the string), and "$" will match before any newline. At the
a0d0e21e
LW
143cost of a little more overhead, you can do this by using the /m modifier
144on the pattern match operator. (Older programs did this by setting C<$*>,
f02c194e 145but this practice has been removed in perl 5.9.)
d74e8afc 146X<^> X<$> X</m>
a0d0e21e 147
14218588 148To simplify multi-line substitutions, the "." character never matches a
55497cff 149newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 150the string is a single line--even if it isn't.
d74e8afc 151X<.> X</s>
a0d0e21e 152
04838cea
RGS
153=head3 Quantifiers
154
a0d0e21e 155The following standard quantifiers are recognized:
d74e8afc 156X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 157
f793d64a
KW
158 * Match 0 or more times
159 + Match 1 or more times
160 ? Match 1 or 0 times
161 {n} Match exactly n times
162 {n,} Match at least n times
163 {n,m} Match at least n but not more than m times
a0d0e21e
LW
164
165(If a curly bracket occurs in any other context, it is treated
b975c076 166as a regular character. In particular, the lower bound
527e91da
BB
167is not optional.) The "*" quantifier is equivalent to C<{0,}>, the "+"
168quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 169to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
170This is usually 32766 on the most common platforms. The actual limit can
171be seen in the error message generated by code such as this:
172
820475bd 173 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 174
54310121 175By default, a quantified subpattern is "greedy", that is, it will match as
176many times as possible (given a particular starting location) while still
177allowing the rest of the pattern to match. If you want it to match the
178minimum number of times possible, follow the quantifier with a "?". Note
179that the meanings don't change, just the "greediness":
0d017f4d 180X<metacharacter> X<greedy> X<greediness>
d74e8afc 181X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 182
f793d64a
KW
183 *? Match 0 or more times, not greedily
184 +? Match 1 or more times, not greedily
185 ?? Match 0 or 1 time, not greedily
186 {n}? Match exactly n times, not greedily
187 {n,}? Match at least n times, not greedily
188 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 189
b9b4dddf
YO
190By default, when a quantified subpattern does not allow the rest of the
191overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 192sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
193as well.
194
f793d64a
KW
195 *+ Match 0 or more times and give nothing back
196 ++ Match 1 or more times and give nothing back
197 ?+ Match 0 or 1 time and give nothing back
198 {n}+ Match exactly n times and give nothing back (redundant)
199 {n,}+ Match at least n times and give nothing back
200 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
201
202For instance,
203
204 'aaaa' =~ /a++a/
205
206will never match, as the C<a++> will gobble up all the C<a>'s in the
207string and won't leave any for the remaining part of the pattern. This
208feature can be extremely useful to give perl hints about where it
209shouldn't backtrack. For instance, the typical "match a double-quoted
210string" problem can be most efficiently performed when written as:
211
212 /"(?:[^"\\]++|\\.)*+"/
213
0d017f4d 214as we know that if the final quote does not match, backtracking will not
b9b4dddf
YO
215help. See the independent subexpression C<< (?>...) >> for more details;
216possessive quantifiers are just syntactic sugar for that construct. For
217instance the above example could also be written as follows:
218
219 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
220
04838cea
RGS
221=head3 Escape sequences
222
5f05dabc 223Because patterns are processed as double quoted strings, the following
a0d0e21e
LW
224also work:
225
f793d64a
KW
226 \t tab (HT, TAB)
227 \n newline (LF, NL)
228 \r return (CR)
229 \f form feed (FF)
230 \a alarm (bell) (BEL)
231 \e escape (think troff) (ESC)
f793d64a 232 \cK control char (example: VT)
dc0d9c48 233 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 234 \N{name} named Unicode character or character sequence
f793d64a 235 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 236 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
237 \l lowercase next char (think vi)
238 \u uppercase next char (think vi)
239 \L lowercase till \E (think vi)
240 \U uppercase till \E (think vi)
241 \Q quote (disable) pattern metacharacters till \E
242 \E end either case modification or quoted section, think vi
a0d0e21e 243
9bb1f947 244Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 245
e1d1eefb 246=head3 Character Classes and other Special Escapes
04838cea 247
a0d0e21e 248In addition, Perl defines the following:
d0b16107 249X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 250
f793d64a
KW
251 Sequence Note Description
252 [...] [1] Match a character according to the rules of the
253 bracketed character class defined by the "...".
254 Example: [a-z] matches "a" or "b" or "c" ... or "z"
255 [[:...:]] [2] Match a character according to the rules of the POSIX
256 character class "..." within the outer bracketed
257 character class. Example: [[:upper:]] matches any
258 uppercase character.
d35dd6c6
KW
259 \w [3] Match a "word" character (alphanumeric plus "_", plus
260 other connector punctuation chars plus Unicode
261 marks
f793d64a
KW
262 \W [3] Match a non-"word" character
263 \s [3] Match a whitespace character
264 \S [3] Match a non-whitespace character
265 \d [3] Match a decimal digit character
266 \D [3] Match a non-digit character
267 \pP [3] Match P, named property. Use \p{Prop} for longer names
268 \PP [3] Match non-P
269 \X [4] Match Unicode "eXtended grapheme cluster"
270 \C Match a single C-language char (octet) even if that is
271 part of a larger UTF-8 character. Thus it breaks up
272 characters into their UTF-8 bytes, so you may end up
273 with malformed pieces of UTF-8. Unsupported in
274 lookbehind.
c27a5cfe 275 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
276 '1' may actually be any positive integer.
277 \g1 [5] Backreference to a specific or previous group,
278 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 279 previous group and may optionally be wrapped in
f793d64a
KW
280 curly brackets for safer parsing.
281 \g{name} [5] Named backreference
282 \k<name> [5] Named backreference
283 \K [6] Keep the stuff left of the \K, don't include it in $&
284 \N [7] Any character but \n (experimental). Not affected by
285 /s modifier
286 \v [3] Vertical whitespace
287 \V [3] Not vertical whitespace
288 \h [3] Horizontal whitespace
289 \H [3] Not horizontal whitespace
290 \R [4] Linebreak
e1d1eefb 291
9bb1f947
KW
292=over 4
293
294=item [1]
295
296See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 297
9bb1f947 298=item [2]
b8c5462f 299
9bb1f947 300See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 301
9bb1f947 302=item [3]
5496314a 303
9bb1f947 304See L<perlrecharclass/Backslash sequences> for details.
5496314a 305
9bb1f947 306=item [4]
5496314a 307
9bb1f947 308See L<perlrebackslash/Misc> for details.
d0b16107 309
9bb1f947 310=item [5]
b8c5462f 311
c27a5cfe 312See L</Capture groups> below for details.
93733859 313
9bb1f947 314=item [6]
b8c5462f 315
9bb1f947
KW
316See L</Extended Patterns> below for details.
317
318=item [7]
319
320Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
321character or character sequence whose name is C<NAME>; and similarly
322when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
323code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947
KW
324
325=back
d0b16107 326
04838cea
RGS
327=head3 Assertions
328
a0d0e21e 329Perl defines the following zero-width assertions:
d74e8afc
ITB
330X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
331X<regexp, zero-width assertion>
332X<regular expression, zero-width assertion>
333X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 334
9bb1f947
KW
335 \b Match a word boundary
336 \B Match except at a word boundary
337 \A Match only at beginning of string
338 \Z Match only at end of string, or before newline at the end
339 \z Match only at end of string
340 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 341 of prior m//g)
a0d0e21e 342
14218588 343A word boundary (C<\b>) is a spot between two characters
19799a22
GS
344that has a C<\w> on one side of it and a C<\W> on the other side
345of it (in either order), counting the imaginary characters off the
346beginning and end of the string as matching a C<\W>. (Within
347character classes C<\b> represents backspace rather than a word
348boundary, just as it normally does in any double-quoted string.)
349The C<\A> and C<\Z> are just like "^" and "$", except that they
350won't match multiple times when the C</m> modifier is used, while
351"^" and "$" will match at every internal line boundary. To match
352the actual end of the string and not ignore an optional trailing
353newline, use C<\z>.
d74e8afc 354X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
355
356The C<\G> assertion can be used to chain global matches (using
357C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
358It is also useful when writing C<lex>-like scanners, when you have
359several patterns that you want to match against consequent substrings
360of your string, see the previous reference. The actual location
361where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d
YO
362an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
363matches is modified somewhat, in that contents to the left of C<\G> is
364not counted when determining the length of the match. Thus the following
365will not match forever:
d74e8afc 366X<\G>
c47ff5f1 367
e761bb84
CO
368 my $string = 'ABC';
369 pos($string) = 1;
370 while ($string =~ /(.\G)/g) {
371 print $1;
372 }
58e23c8d
YO
373
374It will print 'A' and then terminate, as it considers the match to
375be zero-width, and thus will not match at the same position twice in a
376row.
377
378It is worth noting that C<\G> improperly used can result in an infinite
379loop. Take care when using patterns that include C<\G> in an alternation.
380
c27a5cfe 381=head3 Capture groups
04838cea 382
c27a5cfe
KW
383The bracketing construct C<( ... )> creates capture groups (also referred to as
384capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
385the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
386for the second, and so on.
387This is called a I<backreference>.
d74e8afc 388X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 389X<regex, capture group> X<regexp, capture group>
d74e8afc 390X<regular expression, capture buffer> X<backreference>
c27a5cfe 391X<regular expression, capture group> X<backreference>
1f1031fe 392X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
393X<named capture buffer> X<regular expression, named capture buffer>
394X<named capture group> X<regular expression, named capture group>
395X<%+> X<$+{name}> X<< \k<name> >>
396There is no limit to the number of captured substrings that you may use.
397Groups are numbered with the leftmost open parenthesis being number 1, etc. If
398a group did not match, the associated backreference won't match either. (This
399can happen if the group is optional, or in a different branch of an
400alternation.)
401You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
402this form, described below.
403
404You can also refer to capture groups relatively, by using a negative number, so
405that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
406group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
407example:
5624f11d
YO
408
409 /
c27a5cfe
KW
410 (Y) # group 1
411 ( # group 2
412 (X) # group 3
413 \g{-1} # backref to group 3
414 \g{-3} # backref to group 1
5624f11d
YO
415 )
416 /x
417
d8b950dc
KW
418would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
419interpolate regexes into larger regexes and not have to worry about the
420capture groups being renumbered.
421
422You can dispense with numbers altogether and create named capture groups.
423The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
424reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
425also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
426I<name> must not begin with a number, nor contain hyphens.
427When different groups within the same pattern have the same name, any reference
428to that name assumes the leftmost defined group. Named groups count in
429absolute and relative numbering, and so can also be referred to by those
430numbers.
431(It's possible to do things with named capture groups that would otherwise
432require C<(??{})>.)
433
434Capture group contents are dynamically scoped and available to you outside the
435pattern until the end of the enclosing block or until the next successful
436match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
437You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
438etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
439
440Braces are required in referring to named capture groups, but are optional for
441absolute or relative numbered ones. Braces are safer when creating a regex by
442concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
443contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
444is probably not what you intended.
445
446The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
447there were no named nor relative numbered capture groups. Absolute numbered
448groups were referred to using C<\1>, C<\2>, etc, and this notation is still
449accepted (and likely always will be). But it leads to some ambiguities if
450there are more than 9 capture groups, as C<\10> could mean either the tenth
451capture group, or the character whose ordinal in octal is 010 (a backspace in
452ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
453only if at least 10 left parentheses have opened before it. Likewise C<\11> is
454a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
455And so on. C<\1> through C<\9> are always interpreted as backreferences.
456There are several examples below that illustrate these perils. You can avoid
457the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
458and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
459digits padded with leading zeros, since a leading zero implies an octal
460constant.
d8b950dc
KW
461
462The C<\I<digit>> notation also works in certain circumstances outside
463the pattern. See L</Warning on \1 Instead of $1> below for details.)
81714fb9 464
14218588 465Examples:
a0d0e21e
LW
466
467 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
468
d8b950dc 469 /(.)\g1/ # find first doubled char
81714fb9
YO
470 and print "'$1' is the first doubled character\n";
471
472 /(?<char>.)\k<char>/ # ... a different way
473 and print "'$+{char}' is the first doubled character\n";
474
d8b950dc 475 /(?'char'.)\g1/ # ... mix and match
81714fb9 476 and print "'$1' is the first doubled character\n";
c47ff5f1 477
14218588 478 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
479 $hours = $1;
480 $minutes = $2;
481 $seconds = $3;
a0d0e21e 482 }
c47ff5f1 483
9d860678
KW
484 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
485 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
486 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
487 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
488
489 $a = '(.)\1'; # Creates problems when concatenated.
490 $b = '(.)\g{1}'; # Avoids the problems.
491 "aa" =~ /${a}/; # True
492 "aa" =~ /${b}/; # True
493 "aa0" =~ /${a}0/; # False!
494 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
495 "aa\x08" =~ /${a}0/; # True!
496 "aa\x08" =~ /${b}0/; # False
9d860678 497
14218588
GS
498Several special variables also refer back to portions of the previous
499match. C<$+> returns whatever the last bracket match matched.
500C<$&> returns the entire matched string. (At one point C<$0> did
501also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
502everything before the matched string. C<$'> returns everything
503after the matched string. And C<$^N> contains whatever was matched by
504the most-recently closed group (submatch). C<$^N> can be used in
505extended patterns (see below), for example to assign a submatch to a
81714fb9 506variable.
d74e8afc 507X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 508
d8b950dc
KW
509These special variables, like the C<%+> hash and the numbered match variables
510(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
511until the end of the enclosing block or until the next successful
512match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
513X<$+> X<$^N> X<$&> X<$`> X<$'>
514X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
515
0d017f4d 516B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 517which makes it easier to write code that tests for a series of more
665e98b9
JH
518specific cases and remembers the best match.
519
14218588
GS
520B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
521C<$'> anywhere in the program, it has to provide them for every
522pattern match. This may substantially slow your program. Perl
d8b950dc 523uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
524price for each pattern that contains capturing parentheses. (To
525avoid this cost while retaining the grouping behaviour, use the
526extended regular expression C<(?: ... )> instead.) But if you never
527use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
528parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
529if you can, but if you can't (and some algorithms really appreciate
530them), once you've used them once, use them at will, because you've
531already paid the price. As of 5.005, C<$&> is not so costly as the
532other two.
d74e8afc 533X<$&> X<$`> X<$'>
68dc0745 534
99d59c4d 535As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
536C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
537and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 538successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
539The use of these variables incurs no global performance penalty, unlike
540their punctuation char equivalents, however at the trade-off that you
541have to tell perl when you want to use them.
87e95b7f 542X</p> X<p modifier>
cde0cee5 543
9d727203
KW
544=head2 Quoting metacharacters
545
19799a22
GS
546Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
547C<\w>, C<\n>. Unlike some other regular expression languages, there
548are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 549that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
550interpreted as a literal character, not a metacharacter. This was
551once used in a common idiom to disable or quote the special meanings
552of regular expression metacharacters in a string that you want to
36bbe248 553use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
554
555 $pattern =~ s/(\W)/\\$1/g;
556
f1cbbd6e 557(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
558Today it is more common to use the quotemeta() function or the C<\Q>
559metaquoting escape sequence to disable all metacharacters' special
560meanings like this:
a0d0e21e
LW
561
562 /$unquoted\Q$quoted\E$unquoted/
563
9da458fc
IZ
564Beware that if you put literal backslashes (those not inside
565interpolated variables) between C<\Q> and C<\E>, double-quotish
566backslash interpolation may lead to confusing results. If you
567I<need> to use literal backslashes within C<\Q...\E>,
568consult L<perlop/"Gory details of parsing quoted constructs">.
569
19799a22
GS
570=head2 Extended Patterns
571
14218588
GS
572Perl also defines a consistent extension syntax for features not
573found in standard tools like B<awk> and B<lex>. The syntax is a
574pair of parentheses with a question mark as the first thing within
575the parentheses. The character after the question mark indicates
576the extension.
19799a22 577
14218588
GS
578The stability of these extensions varies widely. Some have been
579part of the core language for many years. Others are experimental
580and may change without warning or be completely removed. Check
581the documentation on an individual feature to verify its current
582status.
19799a22 583
14218588
GS
584A question mark was chosen for this and for the minimal-matching
585construct because 1) question marks are rare in older regular
586expressions, and 2) whenever you see one, you should stop and
587"question" exactly what is going on. That's psychology...
a0d0e21e
LW
588
589=over 10
590
cc6b7395 591=item C<(?#text)>
d74e8afc 592X<(?#)>
a0d0e21e 593
14218588 594A comment. The text is ignored. If the C</x> modifier enables
19799a22 595whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
596the comment as soon as it sees a C<)>, so there is no way to put a literal
597C<)> in the comment.
a0d0e21e 598
cfaf538b 599=item C<(?adlupimsx-imsx)>
fb85c044 600
cfaf538b 601=item C<(?^alupimsx)>
fb85c044 602X<(?)> X<(?^)>
19799a22 603
0b6d1084
JH
604One or more embedded pattern-match modifiers, to be turned on (or
605turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
606the remainder of the enclosing pattern group (if any).
607
fb85c044 608This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d
WL
609configuration file, taken from an argument, or specified in a table
610somewhere. Consider the case where some patterns want to be case
611sensitive and some do not: The case insensitive ones merely need to
612include C<(?i)> at the front of the pattern. For example:
19799a22
GS
613
614 $pattern = "foobar";
5d458dd8 615 if ( /$pattern/i ) { }
19799a22
GS
616
617 # more flexible:
618
619 $pattern = "(?i)foobar";
5d458dd8 620 if ( /$pattern/ ) { }
19799a22 621
0b6d1084 622These modifiers are restored at the end of the enclosing group. For example,
19799a22 623
d8b950dc 624 ( (?i) blah ) \s+ \g1
19799a22 625
0d017f4d
WL
626will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
627repetition of the previous word, assuming the C</x> modifier, and no C</i>
628modifier outside this group.
19799a22 629
8eb5594e
DR
630These modifiers do not carry over into named subpatterns called in the
631enclosing group. In other words, a pattern such as C<((?i)(&NAME))> does not
632change the case-sensitivity of the "NAME" pattern.
633
dc925305
KW
634Any of these modifiers can be set to apply globally to all regular
635expressions compiled within the scope of a C<use re>. See
636L<re/'/flags' mode>.
637
9de15fec
KW
638Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
639after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
640C<"d">) may follow the caret to override it.
641But a minus sign is not legal with it.
642
cfaf538b
KW
643Also, starting in Perl 5.14, are modifiers C<"a">, C<"d">, C<"l">, and
644C<"u">, which for 5.14 may not be used as suffix modifiers.
9de15fec
KW
645
646C<"l"> means to use a locale (see L<perllocale>) when pattern matching.
647The locale used will be the one in effect at the time of execution of
648the pattern match. This may not be the same as the compilation-time
649locale, and can differ from one match to another if there is an
650intervening call of the
651L<setlocale() function|perllocale/The setlocale function>.
652This modifier is automatically set if the regular expression is compiled
20db7501
KW
653within the scope of a C<"use locale"> pragma. Results are not
654well-defined when using this and matching against a utf8-encoded string.
9de15fec 655
a12cf05f 656C<"u"> means to use Unicode semantics when pattern matching. It is
20db7501
KW
657automatically set if the regular expression is encoded in utf8, or is
658compiled within the scope of a
659L<C<"use feature 'unicode_strings">|feature> pragma (and isn't also in
660the scope of L<C<"use locale">|locale> nor L<C<"use bytes">|bytes>
661pragmas. On ASCII platforms, the code points between 128 and 255 take on their
662Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's), whereas
663in strict ASCII their meanings are undefined. Thus the platform
664effectively becomes a Unicode platform. The ASCII characters remain as
665ASCII characters (since ASCII is a subset of Latin-1 and Unicode). For
dc925305 666example, when this option is not on, on a non-utf8 string, C<"\w">
20db7501
KW
667matches precisely C<[A-Za-z0-9_]>. When the option is on, it matches
668not just those, but all the Latin-1 word characters (such as an "n" with
669a tilde). On EBCDIC platforms, which already are equivalent to Latin-1,
670this modifier changes behavior only when the C<"/i"> modifier is also
671specified, and affects only two characters, giving them full Unicode
672semantics: the C<MICRO SIGN> will match the Greek capital and
673small letters C<MU>; otherwise not; and the C<LATIN CAPITAL LETTER SHARP
674S> will match any of C<SS>, C<Ss>, C<sS>, and C<ss>, otherwise not.
675(This last case is buggy, however.)
9de15fec 676
dc925305
KW
677C<"a"> is the same as C<"u">, except that C<\d>, C<\s>, C<\w>, and the
678Posix character classes are restricted to matching in the ASCII range
679only. That is, with this modifier, C<\d> always means precisely the
680digits C<"0"> to C<"9">; C<\s> means the five characters C<[ \f\n\r\t]>;
681C<\w> means the 63 characters C<[A-Za-z0-9_]>; and likewise, all the
cfaf538b
KW
682Posix classes such as C<[[:print:]]> match only the appropriate
683ASCII-range characters. As you would expect, this modifier causes, for
dc925305
KW
684example, C<\D> to mean the same thing as C<[^0-9]>; in fact, all
685non-ASCII characters match C<\D>, C<\S>, and C<\W>. C<\b> still means
686to match at the boundary between C<\w> and C<\W>, using the C<"a">
687definitions of them (similarly for C<\B>). Otherwise, C<"a"> behaves
688like the C<"u"> modifier, in that case-insensitive matching uses Unicode
689semantics; for example, "k" will match the Unicode C<\N{KELVIN SIGN}>
690under C</i> matching, and code points in the Latin1 range, above ASCII
691will have Unicode semantics when it comes to case-insensitive matching.
cfaf538b 692
9de15fec
KW
693C<"d"> means to use the traditional Perl pattern matching behavior.
694This is dualistic (hence the name C<"d">, which also could stand for
20db7501 695"depends"). When this is in effect, Perl matches utf8-encoded strings
9de15fec 696using Unicode rules, and matches non-utf8-encoded strings using the
20db7501
KW
697platform's native character set rules. (If the regular expression
698itself is encoded in utf8, Unicode rules are used regardless of the
699target string's encoding.)
9de15fec
KW
700See L<perlunicode/The "Unicode Bug">. It is automatically selected by
701default if the regular expression is compiled neither within the scope
702of a C<"use locale"> pragma nor a <C<"use feature 'unicode_strings">
703pragma.
704
dc925305
KW
705Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
706that they can only be enabled, not disabled, and the C<d>, C<l>, and
707C<u> modifiers are mutually exclusive: specifying one de-specifies the
bdc22dd5 708others, and a maximum of one may appear in the construct. Thus, for
315f11aa 709example, C<(?-p)>, C<(?-d:...)>, and C<(?dl:...)> will warn when
bdc22dd5 710compiled under C<use warnings>.
9de15fec
KW
711
712Note also that the C<p> modifier is special in that its presence
713anywhere in a pattern has a global effect.
cde0cee5 714
5a964f20 715=item C<(?:pattern)>
d74e8afc 716X<(?:)>
a0d0e21e 717
cfaf538b 718=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 719
cfaf538b 720=item C<(?^aluimsx:pattern)>
fb85c044
KW
721X<(?^:)>
722
5a964f20
TC
723This is for clustering, not capturing; it groups subexpressions like
724"()", but doesn't make backreferences as "()" does. So
a0d0e21e 725
5a964f20 726 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
727
728is like
729
5a964f20 730 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 731
19799a22
GS
732but doesn't spit out extra fields. It's also cheaper not to capture
733characters if you don't need to.
a0d0e21e 734
19799a22 735Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 736C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
737
738 /(?s-i:more.*than).*million/i
739
14218588 740is equivalent to the more verbose
ca9dfc88
IZ
741
742 /(?:(?s-i)more.*than).*million/i
743
fb85c044 744Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
745after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
746flags (except C<"d">) may follow the caret, so
fb85c044
KW
747
748 (?^x:foo)
749
750is equivalent to
751
752 (?x-ims:foo)
753
754The caret tells Perl that this cluster doesn't inherit the flags of any
9de15fec 755surrounding pattern, but to go back to the system defaults (C<d-imsx>),
fb85c044
KW
756modified by any flags specified.
757
758The caret allows for simpler stringification of compiled regular
759expressions. These look like
760
761 (?^:pattern)
762
763with any non-default flags appearing between the caret and the colon.
764A test that looks at such stringification thus doesn't need to have the
765system default flags hard-coded in it, just the caret. If new flags are
766added to Perl, the meaning of the caret's expansion will change to include
767the default for those flags, so the test will still work, unchanged.
768
769Specifying a negative flag after the caret is an error, as the flag is
770redundant.
771
772Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
773to match at the beginning.
774
594d7033
YO
775=item C<(?|pattern)>
776X<(?|)> X<Branch reset>
777
778This is the "branch reset" pattern, which has the special property
c27a5cfe 779that the capture groups are numbered from the same starting point
99d59c4d 780in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 781
c27a5cfe 782Capture groups are numbered from left to right, but inside this
693596a8 783construct the numbering is restarted for each branch.
4deaaa80 784
c27a5cfe 785The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
786following this construct will be numbered as though the construct
787contained only one branch, that being the one with the most capture
c27a5cfe 788groups in it.
4deaaa80
PJ
789
790This construct will be useful when you want to capture one of a
791number of alternative matches.
792
793Consider the following pattern. The numbers underneath show in
c27a5cfe 794which group the captured content will be stored.
594d7033
YO
795
796
797 # before ---------------branch-reset----------- after
798 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
799 # 1 2 2 3 2 3 4
800
ab106183
A
801Be careful when using the branch reset pattern in combination with
802named captures. Named captures are implemented as being aliases to
c27a5cfe 803numbered groups holding the captures, and that interferes with the
ab106183
A
804implementation of the branch reset pattern. If you are using named
805captures in a branch reset pattern, it's best to use the same names,
806in the same order, in each of the alternations:
807
808 /(?| (?<a> x ) (?<b> y )
809 | (?<a> z ) (?<b> w )) /x
810
811Not doing so may lead to surprises:
812
813 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
814 say $+ {a}; # Prints '12'
815 say $+ {b}; # *Also* prints '12'.
816
c27a5cfe
KW
817The problem here is that both the group named C<< a >> and the group
818named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 819
ee9b8eae
YO
820=item Look-Around Assertions
821X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
822
823Look-around assertions are zero width patterns which match a specific
824pattern without including it in C<$&>. Positive assertions match when
825their subpattern matches, negative assertions match when their subpattern
826fails. Look-behind matches text up to the current match position,
827look-ahead matches text following the current match position.
828
829=over 4
830
5a964f20 831=item C<(?=pattern)>
d74e8afc 832X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 833
19799a22 834A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
835matches a word followed by a tab, without including the tab in C<$&>.
836
5a964f20 837=item C<(?!pattern)>
d74e8afc 838X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 839
19799a22 840A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 841matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
842however that look-ahead and look-behind are NOT the same thing. You cannot
843use this for look-behind.
7b8d334a 844
5a964f20 845If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
846will not do what you want. That's because the C<(?!foo)> is just saying that
847the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
848match. You would have to do something like C</(?!foo)...bar/> for that. We
849say "like" because there's the case of your "bar" not having three characters
850before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
851Sometimes it's still easier just to say:
a0d0e21e 852
a3cb178b 853 if (/bar/ && $` !~ /foo$/)
a0d0e21e 854
19799a22 855For look-behind see below.
c277df42 856
ee9b8eae
YO
857=item C<(?<=pattern)> C<\K>
858X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 859
c47ff5f1 860A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
861matches a word that follows a tab, without including the tab in C<$&>.
862Works only for fixed-width look-behind.
c277df42 863
ee9b8eae
YO
864There is a special form of this construct, called C<\K>, which causes the
865regex engine to "keep" everything it had matched prior to the C<\K> and
866not include it in C<$&>. This effectively provides variable length
867look-behind. The use of C<\K> inside of another look-around assertion
868is allowed, but the behaviour is currently not well defined.
869
c62285ac 870For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
871equivalent C<< (?<=...) >> construct, and it is especially useful in
872situations where you want to efficiently remove something following
873something else in a string. For instance
874
875 s/(foo)bar/$1/g;
876
877can be rewritten as the much more efficient
878
879 s/foo\Kbar//g;
880
5a964f20 881=item C<(?<!pattern)>
d74e8afc 882X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 883
19799a22
GS
884A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
885matches any occurrence of "foo" that does not follow "bar". Works
886only for fixed-width look-behind.
c277df42 887
ee9b8eae
YO
888=back
889
81714fb9
YO
890=item C<(?'NAME'pattern)>
891
892=item C<< (?<NAME>pattern) >>
893X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
894
c27a5cfe 895A named capture group. Identical in every respect to normal capturing
90a18110 896parentheses C<()> but for the additional fact that C<%+> or C<%-> may be
c27a5cfe 897used after a successful match to refer to a named group. See C<perlvar>
90a18110 898for more details on the C<%+> and C<%-> hashes.
81714fb9 899
c27a5cfe
KW
900If multiple distinct capture groups have the same name then the
901$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 902
0d017f4d 903The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
904
905B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 906function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
907numbered sequentially regardless of being named or not. Thus in the
908pattern
909
910 /(x)(?<foo>y)(z)/
911
912$+{foo} will be the same as $2, and $3 will contain 'z' instead of
913the opposite which is what a .NET regex hacker might expect.
914
1f1031fe
YO
915Currently NAME is restricted to simple identifiers only.
916In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
917its Unicode extension (see L<utf8>),
918though it isn't extended by the locale (see L<perllocale>).
81714fb9 919
1f1031fe 920B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 921with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 922may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 923support the use of single quotes as a delimiter for the name.
81714fb9 924
1f1031fe
YO
925=item C<< \k<NAME> >>
926
927=item C<< \k'NAME' >>
81714fb9
YO
928
929Named backreference. Similar to numeric backreferences, except that
930the group is designated by name and not number. If multiple groups
931have the same name then it refers to the leftmost defined group in
932the current match.
933
0d017f4d 934It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
935earlier in the pattern.
936
937Both forms are equivalent.
938
1f1031fe 939B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 940with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 941may be used instead of C<< \k<NAME> >>.
1f1031fe 942
cc6b7395 943=item C<(?{ code })>
d74e8afc 944X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 945
19799a22 946B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
947experimental, and may be changed without notice. Code executed that
948has side effects may not perform identically from version to version
949due to the effect of future optimisations in the regex engine.
c277df42 950
cc46d5f2 951This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
952always succeeds, and its C<code> is not interpolated. Currently,
953the rules to determine where the C<code> ends are somewhat convoluted.
954
77ea4f6d
JV
955This feature can be used together with the special variable C<$^N> to
956capture the results of submatches in variables without having to keep
957track of the number of nested parentheses. For example:
958
959 $_ = "The brown fox jumps over the lazy dog";
960 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
961 print "color = $color, animal = $animal\n";
962
754091cb
RGS
963Inside the C<(?{...})> block, C<$_> refers to the string the regular
964expression is matching against. You can also use C<pos()> to know what is
fa11829f 965the current position of matching within this string.
754091cb 966
19799a22
GS
967The C<code> is properly scoped in the following sense: If the assertion
968is backtracked (compare L<"Backtracking">), all changes introduced after
969C<local>ization are undone, so that
b9ac3b5b
GS
970
971 $_ = 'a' x 8;
5d458dd8 972 m<
f793d64a 973 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 974 (
5d458dd8 975 a
b9ac3b5b 976 (?{
f793d64a 977 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
b9ac3b5b 978 })
5d458dd8 979 )*
b9ac3b5b 980 aaaa
f793d64a
KW
981 (?{ $res = $cnt }) # On success copy to
982 # non-localized location.
b9ac3b5b
GS
983 >x;
984
0d017f4d 985will set C<$res = 4>. Note that after the match, C<$cnt> returns to the globally
14218588 986introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
987are unwound.
988
19799a22
GS
989This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
990switch. If I<not> used in this way, the result of evaluation of
991C<code> is put into the special variable C<$^R>. This happens
992immediately, so C<$^R> can be used from other C<(?{ code })> assertions
993inside the same regular expression.
b9ac3b5b 994
19799a22
GS
995The assignment to C<$^R> above is properly localized, so the old
996value of C<$^R> is restored if the assertion is backtracked; compare
997L<"Backtracking">.
b9ac3b5b 998
19799a22
GS
999For reasons of security, this construct is forbidden if the regular
1000expression involves run-time interpolation of variables, unless the
1001perilous C<use re 'eval'> pragma has been used (see L<re>), or the
1002variables contain results of C<qr//> operator (see
9bb1f947 1003L<perlop/"qr/STRINGE<sol>msixpo">).
871b0233 1004
0d017f4d 1005This restriction is due to the wide-spread and remarkably convenient
19799a22 1006custom of using run-time determined strings as patterns. For example:
871b0233
IZ
1007
1008 $re = <>;
1009 chomp $re;
1010 $string =~ /$re/;
1011
14218588
GS
1012Before Perl knew how to execute interpolated code within a pattern,
1013this operation was completely safe from a security point of view,
1014although it could raise an exception from an illegal pattern. If
1015you turn on the C<use re 'eval'>, though, it is no longer secure,
1016so you should only do so if you are also using taint checking.
1017Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 1018compartment. See L<perlsec> for details about both these mechanisms.
871b0233 1019
e95d7314
GG
1020B<WARNING>: Use of lexical (C<my>) variables in these blocks is
1021broken. The result is unpredictable and will make perl unstable. The
1022workaround is to use global (C<our>) variables.
1023
8525cfae
FC
1024B<WARNING>: In perl 5.12.x and earlier, the regex engine
1025was not re-entrant, so interpolated code could not
1026safely invoke the regex engine either directly with
e95d7314 1027C<m//> or C<s///>), or indirectly with functions such as
8525cfae 1028C<split>. Invoking the regex engine in these blocks would make perl
e95d7314 1029unstable.
8988a1bb 1030
14455d6c 1031=item C<(??{ code })>
d74e8afc
ITB
1032X<(??{})>
1033X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1034
19799a22 1035B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1036experimental, and may be changed without notice. Code executed that
1037has side effects may not perform identically from version to version
1038due to the effect of future optimisations in the regex engine.
0f5d15d6 1039
19799a22
GS
1040This is a "postponed" regular subexpression. The C<code> is evaluated
1041at run time, at the moment this subexpression may match. The result
1042of evaluation is considered as a regular expression and matched as
61528107 1043if it were inserted instead of this construct. Note that this means
c27a5cfe 1044that the contents of capture groups defined inside an eval'ed pattern
6bda09f9 1045are not available outside of the pattern, and vice versa, there is no
c27a5cfe 1046way for the inner pattern to refer to a capture group defined outside.
6bda09f9
YO
1047Thus,
1048
1049 ('a' x 100)=~/(??{'(.)' x 100})/
1050
81714fb9 1051B<will> match, it will B<not> set $1.
0f5d15d6 1052
428594d9 1053The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
1054where the C<code> ends are currently somewhat convoluted.
1055
1056The following pattern matches a parenthesized group:
0f5d15d6
IZ
1057
1058 $re = qr{
f793d64a
KW
1059 \(
1060 (?:
1061 (?> [^()]+ ) # Non-parens without backtracking
1062 |
1063 (??{ $re }) # Group with matching parens
1064 )*
1065 \)
1066 }x;
0f5d15d6 1067
6bda09f9
YO
1068See also C<(?PARNO)> for a different, more efficient way to accomplish
1069the same task.
1070
0b370c0a
A
1071For reasons of security, this construct is forbidden if the regular
1072expression involves run-time interpolation of variables, unless the
1073perilous C<use re 'eval'> pragma has been used (see L<re>), or the
1074variables contain results of C<qr//> operator (see
9bb1f947 1075L<perlop/"qrE<sol>STRINGE<sol>msixpo">).
0b370c0a 1076
8525cfae
FC
1077In perl 5.12.x and earlier, because the regex engine was not re-entrant,
1078delayed code could not safely invoke the regex engine either directly with
1079C<m//> or C<s///>), or indirectly with functions such as C<split>.
8988a1bb 1080
5d458dd8
YO
1081Recursing deeper than 50 times without consuming any input string will
1082result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
1083changing it requires a custom build.
1084
542fa716
YO
1085=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1086X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1087X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1088X<regex, relative recursion>
6bda09f9 1089
81714fb9 1090Similar to C<(??{ code })> except it does not involve compiling any code,
c27a5cfe
KW
1091instead it treats the contents of a capture group as an independent
1092pattern that must match at the current position. Capture groups
81714fb9 1093contained by the pattern will have the value as determined by the
6bda09f9
YO
1094outermost recursion.
1095
894be9b7 1096PARNO is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1097the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1098the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716 1099C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
c27a5cfe 1100to be relative, with negative numbers indicating preceding capture groups
542fa716 1101and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1102declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1103Note that the counting for relative recursion differs from that of
c27a5cfe 1104relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1105included.
6bda09f9 1106
81714fb9 1107The following pattern matches a function foo() which may contain
f145b7e9 1108balanced parentheses as the argument.
6bda09f9
YO
1109
1110 $re = qr{ ( # paren group 1 (full function)
81714fb9 1111 foo
6bda09f9
YO
1112 ( # paren group 2 (parens)
1113 \(
1114 ( # paren group 3 (contents of parens)
1115 (?:
1116 (?> [^()]+ ) # Non-parens without backtracking
1117 |
1118 (?2) # Recurse to start of paren group 2
1119 )*
1120 )
1121 \)
1122 )
1123 )
1124 }x;
1125
1126If the pattern was used as follows
1127
1128 'foo(bar(baz)+baz(bop))'=~/$re/
1129 and print "\$1 = $1\n",
1130 "\$2 = $2\n",
1131 "\$3 = $3\n";
1132
1133the output produced should be the following:
1134
1135 $1 = foo(bar(baz)+baz(bop))
1136 $2 = (bar(baz)+baz(bop))
81714fb9 1137 $3 = bar(baz)+baz(bop)
6bda09f9 1138
c27a5cfe 1139If there is no corresponding capture group defined, then it is a
61528107 1140fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1141string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1142into perl, so changing it requires a custom build.
1143
542fa716
YO
1144The following shows how using negative indexing can make it
1145easier to embed recursive patterns inside of a C<qr//> construct
1146for later use:
1147
1148 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1149 if (/foo $parens \s+ + \s+ bar $parens/x) {
1150 # do something here...
1151 }
1152
81714fb9 1153B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1154PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1155a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1156as atomic. Also, modifiers are resolved at compile time, so constructs
1157like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1158be processed.
6bda09f9 1159
894be9b7
YO
1160=item C<(?&NAME)>
1161X<(?&NAME)>
1162
0d017f4d
WL
1163Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1164parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1165the same name, then it recurses to the leftmost.
1166
1167It is an error to refer to a name that is not declared somewhere in the
1168pattern.
1169
1f1031fe
YO
1170B<NOTE:> In order to make things easier for programmers with experience
1171with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1172may be used instead of C<< (?&NAME) >>.
1f1031fe 1173
e2e6a0f1
YO
1174=item C<(?(condition)yes-pattern|no-pattern)>
1175X<(?()>
286f584a 1176
e2e6a0f1 1177=item C<(?(condition)yes-pattern)>
286f584a 1178
e2e6a0f1
YO
1179Conditional expression. C<(condition)> should be either an integer in
1180parentheses (which is valid if the corresponding pair of parentheses
1181matched), a look-ahead/look-behind/evaluate zero-width assertion, a
c27a5cfe 1182name in angle brackets or single quotes (which is valid if a group
e2e6a0f1
YO
1183with the given name matched), or the special symbol (R) (true when
1184evaluated inside of recursion or eval). Additionally the R may be
1185followed by a number, (which will be true when evaluated when recursing
1186inside of the appropriate group), or by C<&NAME>, in which case it will
1187be true only when evaluated during recursion in the named group.
1188
1189Here's a summary of the possible predicates:
1190
1191=over 4
1192
1193=item (1) (2) ...
1194
c27a5cfe 1195Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1196
1197=item (<NAME>) ('NAME')
1198
c27a5cfe 1199Checks if a group with the given name has matched something.
e2e6a0f1
YO
1200
1201=item (?{ CODE })
1202
1203Treats the code block as the condition.
1204
1205=item (R)
1206
1207Checks if the expression has been evaluated inside of recursion.
1208
1209=item (R1) (R2) ...
1210
1211Checks if the expression has been evaluated while executing directly
1212inside of the n-th capture group. This check is the regex equivalent of
1213
1214 if ((caller(0))[3] eq 'subname') { ... }
1215
1216In other words, it does not check the full recursion stack.
1217
1218=item (R&NAME)
1219
1220Similar to C<(R1)>, this predicate checks to see if we're executing
1221directly inside of the leftmost group with a given name (this is the same
1222logic used by C<(?&NAME)> to disambiguate). It does not check the full
1223stack, but only the name of the innermost active recursion.
1224
1225=item (DEFINE)
1226
1227In this case, the yes-pattern is never directly executed, and no
1228no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1229See below for details.
1230
1231=back
1232
1233For example:
1234
1235 m{ ( \( )?
1236 [^()]+
1237 (?(1) \) )
1238 }x
1239
1240matches a chunk of non-parentheses, possibly included in parentheses
1241themselves.
1242
1243A special form is the C<(DEFINE)> predicate, which never executes directly
1244its yes-pattern, and does not allow a no-pattern. This allows to define
1245subpatterns which will be executed only by using the recursion mechanism.
1246This way, you can define a set of regular expression rules that can be
1247bundled into any pattern you choose.
1248
1249It is recommended that for this usage you put the DEFINE block at the
1250end of the pattern, and that you name any subpatterns defined within it.
1251
1252Also, it's worth noting that patterns defined this way probably will
1253not be as efficient, as the optimiser is not very clever about
1254handling them.
1255
1256An example of how this might be used is as follows:
1257
2bf803e2 1258 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1259 (?(DEFINE)
2bf803e2
YO
1260 (?<NAME_PAT>....)
1261 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1262 )/x
1263
c27a5cfe
KW
1264Note that capture groups matched inside of recursion are not accessible
1265after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1266necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1267C<$+{NAME}> would be.
286f584a 1268
c47ff5f1 1269=item C<< (?>pattern) >>
6bda09f9 1270X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1271
19799a22
GS
1272An "independent" subexpression, one which matches the substring
1273that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1274position, and it matches I<nothing other than this substring>. This
19799a22
GS
1275construct is useful for optimizations of what would otherwise be
1276"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1277It may also be useful in places where the "grab all you can, and do not
1278give anything back" semantic is desirable.
19799a22 1279
c47ff5f1 1280For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1281(anchored at the beginning of string, as above) will match I<all>
1282characters C<a> at the beginning of string, leaving no C<a> for
1283C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1284since the match of the subgroup C<a*> is influenced by the following
1285group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1286C<a*ab> will match fewer characters than a standalone C<a*>, since
1287this makes the tail match.
1288
c47ff5f1 1289An effect similar to C<< (?>pattern) >> may be achieved by writing
d8b950dc
KW
1290C<(?=(pattern))\g1>. This matches the same substring as a standalone
1291C<a+>, and the following C<\g1> eats the matched string; it therefore
c47ff5f1 1292makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1293(The difference between these two constructs is that the second one
1294uses a capturing group, thus shifting ordinals of backreferences
1295in the rest of a regular expression.)
1296
1297Consider this pattern:
c277df42 1298
871b0233 1299 m{ \(
e2e6a0f1 1300 (
f793d64a 1301 [^()]+ # x+
e2e6a0f1 1302 |
871b0233
IZ
1303 \( [^()]* \)
1304 )+
e2e6a0f1 1305 \)
871b0233 1306 }x
5a964f20 1307
19799a22
GS
1308That will efficiently match a nonempty group with matching parentheses
1309two levels deep or less. However, if there is no such group, it
1310will take virtually forever on a long string. That's because there
1311are so many different ways to split a long string into several
1312substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1313to a subpattern of the above pattern. Consider how the pattern
1314above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1315seconds, but that each extra letter doubles this time. This
1316exponential performance will make it appear that your program has
14218588 1317hung. However, a tiny change to this pattern
5a964f20 1318
e2e6a0f1
YO
1319 m{ \(
1320 (
f793d64a 1321 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1322 |
871b0233
IZ
1323 \( [^()]* \)
1324 )+
e2e6a0f1 1325 \)
871b0233 1326 }x
c277df42 1327
c47ff5f1 1328which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1329this yourself would be a productive exercise), but finishes in a fourth
1330the time when used on a similar string with 1000000 C<a>s. Be aware,
1331however, that this pattern currently triggers a warning message under
9f1b1f2d 1332the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1333C<"matches null string many times in regex">.
c277df42 1334
c47ff5f1 1335On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1336effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1337This was only 4 times slower on a string with 1000000 C<a>s.
1338
9da458fc
IZ
1339The "grab all you can, and do not give anything back" semantic is desirable
1340in many situations where on the first sight a simple C<()*> looks like
1341the correct solution. Suppose we parse text with comments being delimited
1342by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1343its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1344the comment delimiter, because it may "give up" some whitespace if
1345the remainder of the pattern can be made to match that way. The correct
1346answer is either one of these:
1347
1348 (?>#[ \t]*)
1349 #[ \t]*(?![ \t])
1350
1351For example, to grab non-empty comments into $1, one should use either
1352one of these:
1353
1354 / (?> \# [ \t]* ) ( .+ ) /x;
1355 / \# [ \t]* ( [^ \t] .* ) /x;
1356
1357Which one you pick depends on which of these expressions better reflects
1358the above specification of comments.
1359
6bda09f9
YO
1360In some literature this construct is called "atomic matching" or
1361"possessive matching".
1362
b9b4dddf
YO
1363Possessive quantifiers are equivalent to putting the item they are applied
1364to inside of one of these constructs. The following equivalences apply:
1365
1366 Quantifier Form Bracketing Form
1367 --------------- ---------------
1368 PAT*+ (?>PAT*)
1369 PAT++ (?>PAT+)
1370 PAT?+ (?>PAT?)
1371 PAT{min,max}+ (?>PAT{min,max})
1372
e2e6a0f1
YO
1373=back
1374
1375=head2 Special Backtracking Control Verbs
1376
1377B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1378removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1379be noted to avoid problems during upgrades.
1380
1381These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1382otherwise stated the ARG argument is optional; in some cases, it is
1383forbidden.
1384
1385Any pattern containing a special backtracking verb that allows an argument
e1020413 1386has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1387C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1388rules apply:
e2e6a0f1 1389
5d458dd8
YO
1390On failure, the C<$REGERROR> variable will be set to the ARG value of the
1391verb pattern, if the verb was involved in the failure of the match. If the
1392ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1393name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1394none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1395
5d458dd8
YO
1396On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1397the C<$REGMARK> variable will be set to the name of the last
1398C<(*MARK:NAME)> pattern executed. See the explanation for the
1399C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1400
5d458dd8
YO
1401B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1402and most other regex related variables. They are not local to a scope, nor
1403readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1404Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1405
1406If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1407argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1
YO
1408
1409=over 4
1410
1411=item Verbs that take an argument
1412
1413=over 4
1414
5d458dd8 1415=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1416X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1417
5d458dd8
YO
1418This zero-width pattern prunes the backtracking tree at the current point
1419when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1420where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1421A may backtrack as necessary to match. Once it is reached, matching
1422continues in B, which may also backtrack as necessary; however, should B
1423not match, then no further backtracking will take place, and the pattern
1424will fail outright at the current starting position.
54612592
YO
1425
1426The following example counts all the possible matching strings in a
1427pattern (without actually matching any of them).
1428
e2e6a0f1 1429 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1430 print "Count=$count\n";
1431
1432which produces:
1433
1434 aaab
1435 aaa
1436 aa
1437 a
1438 aab
1439 aa
1440 a
1441 ab
1442 a
1443 Count=9
1444
5d458dd8 1445If we add a C<(*PRUNE)> before the count like the following
54612592 1446
5d458dd8 1447 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1448 print "Count=$count\n";
1449
1450we prevent backtracking and find the count of the longest matching
353c6505 1451at each matching starting point like so:
54612592
YO
1452
1453 aaab
1454 aab
1455 ab
1456 Count=3
1457
5d458dd8 1458Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1459
5d458dd8
YO
1460See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1461control backtracking. In some cases, the use of C<(*PRUNE)> can be
1462replaced with a C<< (?>pattern) >> with no functional difference; however,
1463C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1464C<< (?>pattern) >> alone.
54612592 1465
e2e6a0f1 1466
5d458dd8
YO
1467=item C<(*SKIP)> C<(*SKIP:NAME)>
1468X<(*SKIP)>
e2e6a0f1 1469
5d458dd8 1470This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1471failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1472to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1473of this pattern. This effectively means that the regex engine "skips" forward
1474to this position on failure and tries to match again, (assuming that
1475there is sufficient room to match).
1476
1477The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1478C<(*MARK:NAME)> was encountered while matching, then it is that position
1479which is used as the "skip point". If no C<(*MARK)> of that name was
1480encountered, then the C<(*SKIP)> operator has no effect. When used
1481without a name the "skip point" is where the match point was when
1482executing the (*SKIP) pattern.
1483
1484Compare the following to the examples in C<(*PRUNE)>, note the string
24b23f37
YO
1485is twice as long:
1486
5d458dd8 1487 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1488 print "Count=$count\n";
1489
1490outputs
1491
1492 aaab
1493 aaab
1494 Count=2
1495
5d458dd8 1496Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1497executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1498C<(*SKIP)> was executed.
1499
5d458dd8
YO
1500=item C<(*MARK:NAME)> C<(*:NAME)>
1501X<(*MARK)> C<(*MARK:NAME)> C<(*:NAME)>
1502
1503This zero-width pattern can be used to mark the point reached in a string
1504when a certain part of the pattern has been successfully matched. This
1505mark may be given a name. A later C<(*SKIP)> pattern will then skip
1506forward to that point if backtracked into on failure. Any number of
b4222fa9 1507C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1508
1509In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1510can be used to "label" a pattern branch, so that after matching, the
1511program can determine which branches of the pattern were involved in the
1512match.
1513
1514When a match is successful, the C<$REGMARK> variable will be set to the
1515name of the most recently executed C<(*MARK:NAME)> that was involved
1516in the match.
1517
1518This can be used to determine which branch of a pattern was matched
c27a5cfe 1519without using a separate capture group for each branch, which in turn
5d458dd8
YO
1520can result in a performance improvement, as perl cannot optimize
1521C</(?:(x)|(y)|(z))/> as efficiently as something like
1522C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1523
1524When a match has failed, and unless another verb has been involved in
1525failing the match and has provided its own name to use, the C<$REGERROR>
1526variable will be set to the name of the most recently executed
1527C<(*MARK:NAME)>.
1528
1529See C<(*SKIP)> for more details.
1530
b62d2d15
YO
1531As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1532
5d458dd8
YO
1533=item C<(*THEN)> C<(*THEN:NAME)>
1534
241e7389 1535This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1536C<(*PRUNE)>, this verb always matches, and when backtracked into on
1537failure, it causes the regex engine to try the next alternation in the
1538innermost enclosing group (capturing or otherwise).
1539
1540Its name comes from the observation that this operation combined with the
1541alternation operator (C<|>) can be used to create what is essentially a
1542pattern-based if/then/else block:
1543
1544 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1545
1546Note that if this operator is used and NOT inside of an alternation then
1547it acts exactly like the C<(*PRUNE)> operator.
1548
1549 / A (*PRUNE) B /
1550
1551is the same as
1552
1553 / A (*THEN) B /
1554
1555but
1556
1557 / ( A (*THEN) B | C (*THEN) D ) /
1558
1559is not the same as
1560
1561 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1562
1563as after matching the A but failing on the B the C<(*THEN)> verb will
1564backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1565
e2e6a0f1
YO
1566=item C<(*COMMIT)>
1567X<(*COMMIT)>
24b23f37 1568
241e7389 1569This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1570zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1571into on failure it causes the match to fail outright. No further attempts
1572to find a valid match by advancing the start pointer will occur again.
1573For example,
24b23f37 1574
e2e6a0f1 1575 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1576 print "Count=$count\n";
1577
1578outputs
1579
1580 aaab
1581 Count=1
1582
e2e6a0f1
YO
1583In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1584does not match, the regex engine will not try any further matching on the
1585rest of the string.
c277df42 1586
e2e6a0f1 1587=back
9af228c6 1588
e2e6a0f1 1589=item Verbs without an argument
9af228c6
YO
1590
1591=over 4
1592
e2e6a0f1
YO
1593=item C<(*FAIL)> C<(*F)>
1594X<(*FAIL)> X<(*F)>
9af228c6 1595
e2e6a0f1
YO
1596This pattern matches nothing and always fails. It can be used to force the
1597engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1598fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1599
e2e6a0f1 1600It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1601
e2e6a0f1
YO
1602=item C<(*ACCEPT)>
1603X<(*ACCEPT)>
9af228c6 1604
e2e6a0f1
YO
1605B<WARNING:> This feature is highly experimental. It is not recommended
1606for production code.
9af228c6 1607
e2e6a0f1
YO
1608This pattern matches nothing and causes the end of successful matching at
1609the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1610whether there is actually more to match in the string. When inside of a
0d017f4d 1611nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1612via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1613
c27a5cfe 1614If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1615marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1616For instance:
9af228c6 1617
e2e6a0f1 1618 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1619
e2e6a0f1 1620will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0d017f4d 1621be set. If another branch in the inner parentheses were matched, such as in the
e2e6a0f1 1622string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1623
1624=back
c277df42 1625
a0d0e21e
LW
1626=back
1627
c07a80fd 1628=head2 Backtracking
d74e8afc 1629X<backtrack> X<backtracking>
c07a80fd 1630
35a734be
IZ
1631NOTE: This section presents an abstract approximation of regular
1632expression behavior. For a more rigorous (and complicated) view of
1633the rules involved in selecting a match among possible alternatives,
0d017f4d 1634see L<Combining RE Pieces>.
35a734be 1635
c277df42 1636A fundamental feature of regular expression matching involves the
5a964f20 1637notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1638by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1639C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1640internally, but the general principle outlined here is valid.
c07a80fd 1641
1642For a regular expression to match, the I<entire> regular expression must
1643match, not just part of it. So if the beginning of a pattern containing a
1644quantifier succeeds in a way that causes later parts in the pattern to
1645fail, the matching engine backs up and recalculates the beginning
1646part--that's why it's called backtracking.
1647
1648Here is an example of backtracking: Let's say you want to find the
1649word following "foo" in the string "Food is on the foo table.":
1650
1651 $_ = "Food is on the foo table.";
1652 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 1653 print "$2 follows $1.\n";
c07a80fd 1654 }
1655
1656When the match runs, the first part of the regular expression (C<\b(foo)>)
1657finds a possible match right at the beginning of the string, and loads up
1658$1 with "Foo". However, as soon as the matching engine sees that there's
1659no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1660mistake and starts over again one character after where it had the
c07a80fd 1661tentative match. This time it goes all the way until the next occurrence
1662of "foo". The complete regular expression matches this time, and you get
1663the expected output of "table follows foo."
1664
1665Sometimes minimal matching can help a lot. Imagine you'd like to match
1666everything between "foo" and "bar". Initially, you write something
1667like this:
1668
1669 $_ = "The food is under the bar in the barn.";
1670 if ( /foo(.*)bar/ ) {
f793d64a 1671 print "got <$1>\n";
c07a80fd 1672 }
1673
1674Which perhaps unexpectedly yields:
1675
1676 got <d is under the bar in the >
1677
1678That's because C<.*> was greedy, so you get everything between the
14218588 1679I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd 1680to use minimal matching to make sure you get the text between a "foo"
1681and the first "bar" thereafter.
1682
1683 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1684 got <d is under the >
1685
0d017f4d 1686Here's another example. Let's say you'd like to match a number at the end
b6e13d97 1687of a string, and you also want to keep the preceding part of the match.
c07a80fd 1688So you write this:
1689
1690 $_ = "I have 2 numbers: 53147";
f793d64a
KW
1691 if ( /(.*)(\d*)/ ) { # Wrong!
1692 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd 1693 }
1694
1695That won't work at all, because C<.*> was greedy and gobbled up the
1696whole string. As C<\d*> can match on an empty string the complete
1697regular expression matched successfully.
1698
8e1088bc 1699 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd 1700
1701Here are some variants, most of which don't work:
1702
1703 $_ = "I have 2 numbers: 53147";
1704 @pats = qw{
f793d64a
KW
1705 (.*)(\d*)
1706 (.*)(\d+)
1707 (.*?)(\d*)
1708 (.*?)(\d+)
1709 (.*)(\d+)$
1710 (.*?)(\d+)$
1711 (.*)\b(\d+)$
1712 (.*\D)(\d+)$
c07a80fd 1713 };
1714
1715 for $pat (@pats) {
f793d64a
KW
1716 printf "%-12s ", $pat;
1717 if ( /$pat/ ) {
1718 print "<$1> <$2>\n";
1719 } else {
1720 print "FAIL\n";
1721 }
c07a80fd 1722 }
1723
1724That will print out:
1725
1726 (.*)(\d*) <I have 2 numbers: 53147> <>
1727 (.*)(\d+) <I have 2 numbers: 5314> <7>
1728 (.*?)(\d*) <> <>
1729 (.*?)(\d+) <I have > <2>
1730 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1731 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1732 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1733 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1734
1735As you see, this can be a bit tricky. It's important to realize that a
1736regular expression is merely a set of assertions that gives a definition
1737of success. There may be 0, 1, or several different ways that the
1738definition might succeed against a particular string. And if there are
5a964f20
TC
1739multiple ways it might succeed, you need to understand backtracking to
1740know which variety of success you will achieve.
c07a80fd 1741
19799a22 1742When using look-ahead assertions and negations, this can all get even
8b19b778 1743trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd 1744followed by "123". You might try to write that as
1745
871b0233 1746 $_ = "ABC123";
f793d64a
KW
1747 if ( /^\D*(?!123)/ ) { # Wrong!
1748 print "Yup, no 123 in $_\n";
871b0233 1749 }
c07a80fd 1750
1751But that isn't going to match; at least, not the way you're hoping. It
1752claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1753why that pattern matches, contrary to popular expectations:
c07a80fd 1754
4358a253
SS
1755 $x = 'ABC123';
1756 $y = 'ABC445';
c07a80fd 1757
4358a253
SS
1758 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1759 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1760
4358a253
SS
1761 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1762 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd 1763
1764This prints
1765
1766 2: got ABC
1767 3: got AB
1768 4: got ABC
1769
5f05dabc 1770You might have expected test 3 to fail because it seems to a more
c07a80fd 1771general purpose version of test 1. The important difference between
1772them is that test 3 contains a quantifier (C<\D*>) and so can use
1773backtracking, whereas test 1 will not. What's happening is
1774that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1775non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1776let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1777fail.
14218588 1778
c07a80fd 1779The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1780try to match C<(?!123> with "123", which fails. But because
c07a80fd 1781a quantifier (C<\D*>) has been used in the regular expression, the
1782search engine can backtrack and retry the match differently
54310121 1783in the hope of matching the complete regular expression.
c07a80fd 1784
5a964f20
TC
1785The pattern really, I<really> wants to succeed, so it uses the
1786standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1787time. Now there's indeed something following "AB" that is not
14218588 1788"123". It's "C123", which suffices.
c07a80fd 1789
14218588
GS
1790We can deal with this by using both an assertion and a negation.
1791We'll say that the first part in $1 must be followed both by a digit
1792and by something that's not "123". Remember that the look-aheads
1793are zero-width expressions--they only look, but don't consume any
1794of the string in their match. So rewriting this way produces what
c07a80fd 1795you'd expect; that is, case 5 will fail, but case 6 succeeds:
1796
4358a253
SS
1797 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1798 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd 1799
1800 6: got ABC
1801
5a964f20 1802In other words, the two zero-width assertions next to each other work as though
19799a22 1803they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd 1804matches only if you're at the beginning of the line AND the end of the
1805line simultaneously. The deeper underlying truth is that juxtaposition in
1806regular expressions always means AND, except when you write an explicit OR
1807using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1808although the attempted matches are made at different positions because "a"
1809is not a zero-width assertion, but a one-width assertion.
1810
0d017f4d 1811B<WARNING>: Particularly complicated regular expressions can take
14218588 1812exponential time to solve because of the immense number of possible
0d017f4d 1813ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
1814internal optimizations done by the regular expression engine, this will
1815take a painfully long time to run:
c07a80fd 1816
e1901655
IZ
1817 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1818
1819And if you used C<*>'s in the internal groups instead of limiting them
1820to 0 through 5 matches, then it would take forever--or until you ran
1821out of stack space. Moreover, these internal optimizations are not
1822always applicable. For example, if you put C<{0,5}> instead of C<*>
1823on the external group, no current optimization is applicable, and the
1824match takes a long time to finish.
c07a80fd 1825
9da458fc
IZ
1826A powerful tool for optimizing such beasts is what is known as an
1827"independent group",
c47ff5f1 1828which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1829zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 1830the tail match, since they are in "logical" context: only
14218588 1831whether they match is considered relevant. For an example
9da458fc 1832where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1833following match, see L<C<< (?>pattern) >>>.
c277df42 1834
a0d0e21e 1835=head2 Version 8 Regular Expressions
d74e8afc 1836X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1837
5a964f20 1838In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1839routines, here are the pattern-matching rules not described above.
1840
54310121 1841Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1842with a special meaning described here or above. You can cause
5a964f20 1843characters that normally function as metacharacters to be interpreted
5f05dabc 1844literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
1845character; "\\" matches a "\"). This escape mechanism is also required
1846for the character used as the pattern delimiter.
1847
1848A series of characters matches that series of characters in the target
1849string, so the pattern C<blurfl> would match "blurfl" in the target
1850string.
a0d0e21e
LW
1851
1852You can specify a character class, by enclosing a list of characters
5d458dd8 1853in C<[]>, which will match any character from the list. If the
a0d0e21e 1854first character after the "[" is "^", the class matches any character not
14218588 1855in the list. Within a list, the "-" character specifies a
5a964f20 1856range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1857inclusive. If you want either "-" or "]" itself to be a member of a
1858class, put it at the start of the list (possibly after a "^"), or
1859escape it with a backslash. "-" is also taken literally when it is
1860at the end of the list, just before the closing "]". (The
84850974
DD
1861following all specify the same class of three characters: C<[-az]>,
1862C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
1863specifies a class containing twenty-six characters, even on EBCDIC-based
1864character sets.) Also, if you try to use the character
1865classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1866a range, the "-" is understood literally.
a0d0e21e 1867
8ada0baa
JH
1868Note also that the whole range idea is rather unportable between
1869character sets--and even within character sets they may cause results
1870you probably didn't expect. A sound principle is to use only ranges
0d017f4d 1871that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
1872[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1873spell out the character sets in full.
1874
54310121 1875Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1876used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1877"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 1878of three octal digits, matches the character whose coded character set value
5d458dd8 1879is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 1880matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 1881matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 1882matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1883
1884You can specify a series of alternatives for a pattern using "|" to
1885separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1886or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1887first alternative includes everything from the last pattern delimiter
1888("(", "[", or the beginning of the pattern) up to the first "|", and
1889the last alternative contains everything from the last "|" to the next
14218588
GS
1890pattern delimiter. That's why it's common practice to include
1891alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1892start and end.
1893
5a964f20 1894Alternatives are tried from left to right, so the first
a3cb178b
GS
1895alternative found for which the entire expression matches, is the one that
1896is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1897example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1898part will match, as that is the first alternative tried, and it successfully
1899matches the target string. (This might not seem important, but it is
1900important when you are capturing matched text using parentheses.)
1901
5a964f20 1902Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1903so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1904
14218588
GS
1905Within a pattern, you may designate subpatterns for later reference
1906by enclosing them in parentheses, and you may refer back to the
1907I<n>th subpattern later in the pattern using the metacharacter
1908\I<n>. Subpatterns are numbered based on the left to right order
1909of their opening parenthesis. A backreference matches whatever
1910actually matched the subpattern in the string being examined, not
d8b950dc 1911the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
1912match "0x1234 0x4321", but not "0x1234 01234", because subpattern
19131 matched "0x", even though the rule C<0|0x> could potentially match
1914the leading 0 in the second number.
cb1a09d0 1915
0d017f4d 1916=head2 Warning on \1 Instead of $1
cb1a09d0 1917
5a964f20 1918Some people get too used to writing things like:
cb1a09d0
AD
1919
1920 $pattern =~ s/(\W)/\\\1/g;
1921
3ff1c45a
KW
1922This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
1923shocking the
cb1a09d0 1924B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1925PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1926the usual double-quoted string means a control-A. The customary Unix
1927meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1928of doing that, you get yourself into trouble if you then add an C</e>
1929modifier.
1930
f793d64a 1931 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1932
1933Or if you try to do
1934
1935 s/(\d+)/\1000/;
1936
1937You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1938C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1939with the operation of matching a backreference. Certainly they mean two
1940different things on the I<left> side of the C<s///>.
9fa51da4 1941
0d017f4d 1942=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 1943
19799a22 1944B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1945
1946Regular expressions provide a terse and powerful programming language. As
1947with most other power tools, power comes together with the ability
1948to wreak havoc.
1949
1950A common abuse of this power stems from the ability to make infinite
628afcb5 1951loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1952
1953 'foo' =~ m{ ( o? )* }x;
1954
0d017f4d 1955The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 1956in the string is not moved by the match, C<o?> would match again and again
527e91da 1957because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
1958is with the looping modifier C<//g>:
1959
1960 @matches = ( 'foo' =~ m{ o? }xg );
1961
1962or
1963
1964 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1965
1966or the loop implied by split().
1967
1968However, long experience has shown that many programming tasks may
14218588
GS
1969be significantly simplified by using repeated subexpressions that
1970may match zero-length substrings. Here's a simple example being:
c84d73f1 1971
f793d64a 1972 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
1973 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1974
9da458fc 1975Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 1976the infinite loop>. The rules for this are different for lower-level
527e91da 1977loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
1978ones like the C</g> modifier or split() operator.
1979
19799a22
GS
1980The lower-level loops are I<interrupted> (that is, the loop is
1981broken) when Perl detects that a repeated expression matched a
1982zero-length substring. Thus
c84d73f1
IZ
1983
1984 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1985
5d458dd8 1986is made equivalent to
c84d73f1 1987
5d458dd8
YO
1988 m{ (?: NON_ZERO_LENGTH )*
1989 |
1990 (?: ZERO_LENGTH )?
c84d73f1
IZ
1991 }x;
1992
1993The higher level-loops preserve an additional state between iterations:
5d458dd8 1994whether the last match was zero-length. To break the loop, the following
c84d73f1 1995match after a zero-length match is prohibited to have a length of zero.
5d458dd8 1996This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
1997and so the I<second best> match is chosen if the I<best> match is of
1998zero length.
1999
19799a22 2000For example:
c84d73f1
IZ
2001
2002 $_ = 'bar';
2003 s/\w??/<$&>/g;
2004
20fb949f 2005results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2006match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2007best> match is what is matched by C<\w>. Thus zero-length matches
2008alternate with one-character-long matches.
2009
5d458dd8 2010Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2011position one notch further in the string.
2012
19799a22 2013The additional state of being I<matched with zero-length> is associated with
c84d73f1 2014the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2015Zero-length matches at the end of the previous match are ignored
2016during C<split>.
c84d73f1 2017
0d017f4d 2018=head2 Combining RE Pieces
35a734be
IZ
2019
2020Each of the elementary pieces of regular expressions which were described
2021before (such as C<ab> or C<\Z>) could match at most one substring
2022at the given position of the input string. However, in a typical regular
2023expression these elementary pieces are combined into more complicated
2024patterns using combining operators C<ST>, C<S|T>, C<S*> etc
2025(in these examples C<S> and C<T> are regular subexpressions).
2026
2027Such combinations can include alternatives, leading to a problem of choice:
2028if we match a regular expression C<a|ab> against C<"abc">, will it match
2029substring C<"a"> or C<"ab">? One way to describe which substring is
2030actually matched is the concept of backtracking (see L<"Backtracking">).
2031However, this description is too low-level and makes you think
2032in terms of a particular implementation.
2033
2034Another description starts with notions of "better"/"worse". All the
2035substrings which may be matched by the given regular expression can be
2036sorted from the "best" match to the "worst" match, and it is the "best"
2037match which is chosen. This substitutes the question of "what is chosen?"
2038by the question of "which matches are better, and which are worse?".
2039
2040Again, for elementary pieces there is no such question, since at most
2041one match at a given position is possible. This section describes the
2042notion of better/worse for combining operators. In the description
2043below C<S> and C<T> are regular subexpressions.
2044
13a2d996 2045=over 4
35a734be
IZ
2046
2047=item C<ST>
2048
2049Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2050substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2051which can be matched by C<T>.
35a734be
IZ
2052
2053If C<A> is better match for C<S> than C<A'>, C<AB> is a better
2054match than C<A'B'>.
2055
2056If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
2057C<B> is better match for C<T> than C<B'>.
2058
2059=item C<S|T>
2060
2061When C<S> can match, it is a better match than when only C<T> can match.
2062
2063Ordering of two matches for C<S> is the same as for C<S>. Similar for
2064two matches for C<T>.
2065
2066=item C<S{REPEAT_COUNT}>
2067
2068Matches as C<SSS...S> (repeated as many times as necessary).
2069
2070=item C<S{min,max}>
2071
2072Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2073
2074=item C<S{min,max}?>
2075
2076Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2077
2078=item C<S?>, C<S*>, C<S+>
2079
2080Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2081
2082=item C<S??>, C<S*?>, C<S+?>
2083
2084Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2085
c47ff5f1 2086=item C<< (?>S) >>
35a734be
IZ
2087
2088Matches the best match for C<S> and only that.
2089
2090=item C<(?=S)>, C<(?<=S)>
2091
2092Only the best match for C<S> is considered. (This is important only if
2093C<S> has capturing parentheses, and backreferences are used somewhere
2094else in the whole regular expression.)
2095
2096=item C<(?!S)>, C<(?<!S)>
2097
2098For this grouping operator there is no need to describe the ordering, since
2099only whether or not C<S> can match is important.
2100
6bda09f9 2101=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2102
2103The ordering is the same as for the regular expression which is
c27a5cfe 2104the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
2105
2106=item C<(?(condition)yes-pattern|no-pattern)>
2107
2108Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2109already determined. The ordering of the matches is the same as for the
2110chosen subexpression.
2111
2112=back
2113
2114The above recipes describe the ordering of matches I<at a given position>.
2115One more rule is needed to understand how a match is determined for the
2116whole regular expression: a match at an earlier position is always better
2117than a match at a later position.
2118
0d017f4d 2119=head2 Creating Custom RE Engines
c84d73f1
IZ
2120
2121Overloaded constants (see L<overload>) provide a simple way to extend
2122the functionality of the RE engine.
2123
2124Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2125matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2126characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2127at these positions, so we want to have each C<\Y|> in the place of the
2128more complicated version. We can create a module C<customre> to do
2129this:
2130
2131 package customre;
2132 use overload;
2133
2134 sub import {
2135 shift;
2136 die "No argument to customre::import allowed" if @_;
2137 overload::constant 'qr' => \&convert;
2138 }
2139
2140 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2141
580a9fe1
RGS
2142 # We must also take care of not escaping the legitimate \\Y|
2143 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2144 my %rules = ( '\\' => '\\\\',
f793d64a 2145 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2146 sub convert {
2147 my $re = shift;
5d458dd8 2148 $re =~ s{
c84d73f1
IZ
2149 \\ ( \\ | Y . )
2150 }
5d458dd8 2151 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2152 return $re;
2153 }
2154
2155Now C<use customre> enables the new escape in constant regular
2156expressions, i.e., those without any runtime variable interpolations.
2157As documented in L<overload>, this conversion will work only over
2158literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2159part of this regular expression needs to be converted explicitly
2160(but only if the special meaning of C<\Y|> should be enabled inside $re):
2161
2162 use customre;
2163 $re = <>;
2164 chomp $re;
2165 $re = customre::convert $re;
2166 /\Y|$re\Y|/;
2167
1f1031fe
YO
2168=head1 PCRE/Python Support
2169
99d59c4d 2170As of Perl 5.10.0, Perl supports several Python/PCRE specific extensions
1f1031fe 2171to the regex syntax. While Perl programmers are encouraged to use the
99d59c4d 2172Perl specific syntax, the following are also accepted:
1f1031fe
YO
2173
2174=over 4
2175
ae5648b3 2176=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2177
c27a5cfe 2178Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2179
2180=item C<< (?P=NAME) >>
2181
c27a5cfe 2182Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2183
2184=item C<< (?P>NAME) >>
2185
c27a5cfe 2186Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2187
ee9b8eae 2188=back
1f1031fe 2189
19799a22
GS
2190=head1 BUGS
2191
78288b8e
KW
2192There are numerous problems with case insensitive matching of characters
2193outside the ASCII range, especially with those whose folds are multiple
2194characters, such as ligatures like C<LATIN SMALL LIGATURE FF>.
2195
f253210b
KW
2196In a bracketed character class with case insensitive matching, ranges only work
2197for ASCII characters. For example,
2198C<m/[\N{CYRILLIC CAPITAL LETTER A}-\N{CYRILLIC CAPITAL LETTER YA}]/i>
2199doesn't match all the Russian upper and lower case letters.
2200
88c9975e
KW
2201Many regular expression constructs don't work on EBCDIC platforms.
2202
9da458fc
IZ
2203This document varies from difficult to understand to completely
2204and utterly opaque. The wandering prose riddled with jargon is
2205hard to fathom in several places.
2206
2207This document needs a rewrite that separates the tutorial content
2208from the reference content.
19799a22
GS
2209
2210=head1 SEE ALSO
9fa51da4 2211
91e0c79e
MJD
2212L<perlrequick>.
2213
2214L<perlretut>.
2215
9b599b2a
GS
2216L<perlop/"Regexp Quote-Like Operators">.
2217
1e66bd83
PP
2218L<perlop/"Gory details of parsing quoted constructs">.
2219
14218588
GS
2220L<perlfaq6>.
2221
9b599b2a
GS
2222L<perlfunc/pos>.
2223
2224L<perllocale>.
2225
fb55449c
JH
2226L<perlebcdic>.
2227
14218588
GS
2228I<Mastering Regular Expressions> by Jeffrey Friedl, published
2229by O'Reilly and Associates.