This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
more perlfunc tweaks
[perl5.git] / pod / perlre.pod
CommitLineData
a0d0e21e 1=head1 NAME
d74e8afc 2X<regular expression> X<regex> X<regexp>
a0d0e21e
LW
3
4perlre - Perl regular expressions
5
6=head1 DESCRIPTION
7
5d458dd8 8This page describes the syntax of regular expressions in Perl.
91e0c79e 9
cc46d5f2 10If you haven't used regular expressions before, a quick-start
91e0c79e
MJD
11introduction is available in L<perlrequick>, and a longer tutorial
12introduction is available in L<perlretut>.
13
14For reference on how regular expressions are used in matching
15operations, plus various examples of the same, see discussions of
16C<m//>, C<s///>, C<qr//> and C<??> in L<perlop/"Regexp Quote-Like
17Operators">.
cb1a09d0 18
0d017f4d
WL
19
20=head2 Modifiers
21
19799a22 22Matching operations can have various modifiers. Modifiers
5a964f20 23that relate to the interpretation of the regular expression inside
19799a22 24are listed below. Modifiers that alter the way a regular expression
5d458dd8 25is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
1e66bd83 26L<perlop/"Gory details of parsing quoted constructs">.
a0d0e21e 27
55497cff
PP
28=over 4
29
54310121 30=item m
d74e8afc 31X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff
PP
32
33Treat string as multiple lines. That is, change "^" and "$" from matching
14218588 34the start or end of the string to matching the start or end of any
7f761169 35line anywhere within the string.
55497cff 36
54310121 37=item s
d74e8afc
ITB
38X</s> X<regex, single-line> X<regexp, single-line>
39X<regular expression, single-line>
55497cff
PP
40
41Treat string as single line. That is, change "." to match any character
19799a22 42whatsoever, even a newline, which normally it would not match.
55497cff 43
34d67d80 44Used together, as C</ms>, they let the "." match any character whatsoever,
fb55449c 45while still allowing "^" and "$" to match, respectively, just after
19799a22 46and just before newlines within the string.
7b8d334a 47
87e95b7f
YO
48=item i
49X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
50X<regular expression, case-insensitive>
51
52Do case-insensitive pattern matching.
53
54If C<use locale> is in effect, the case map is taken from the current
17580e7a
KW
55locale for code points less than 255, and from Unicode rules for larger
56code points. See L<perllocale>.
87e95b7f 57
54310121 58=item x
d74e8afc 59X</x>
55497cff
PP
60
61Extend your pattern's legibility by permitting whitespace and comments.
62
87e95b7f
YO
63=item p
64X</p> X<regex, preserve> X<regexp, preserve>
65
632a1772 66Preserve the string matched such that ${^PREMATCH}, ${^MATCH}, and
87e95b7f
YO
67${^POSTMATCH} are available for use after matching.
68
e2e6bec7
DN
69=item g and c
70X</g> X</c>
71
72Global matching, and keep the Current position after failed matching.
73Unlike i, m, s and x, these two flags affect the way the regex is used
74rather than the regex itself. See
75L<perlretut/"Using regular expressions in Perl"> for further explanation
76of the g and c modifiers.
77
55497cff 78=back
a0d0e21e
LW
79
80These are usually written as "the C</x> modifier", even though the delimiter
ab7bb42d
A
81in question might not really be a slash. The modifiers C</imsx>
82may also be embedded within the regular expression itself using
da392a17
KW
83the C<(?...)> construct. Also are new (in 5.14) character set semantics
84modifiers B<C<<"a">>, B<C<"d">>, B<C<"l">> and B<C<"u">>, which, in 5.14
85only, must be used embedded in the regular expression, and not after the
86trailing delimiter. All this is discussed below in
87L</Extended Patterns>.
88X</a> X</d> X</l> X</u>
a0d0e21e 89
4633a7c4 90The C</x> modifier itself needs a little more explanation. It tells
7b059540 91the regular expression parser to ignore most whitespace that is neither
55497cff 92backslashed nor within a character class. You can use this to break up
4633a7c4 93your regular expression into (slightly) more readable parts. The C<#>
54310121 94character is also treated as a metacharacter introducing a comment,
55497cff 95just as in ordinary Perl code. This also means that if you want real
14218588 96whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a 97class, where they are unaffected by C</x>), then you'll either have to
7b059540
KW
98escape them (using backslashes or C<\Q...\E>) or encode them using octal,
99hex, or C<\N{}> escapes. Taken together, these features go a long way towards
8933a740
RGS
100making Perl's regular expressions more readable. Note that you have to
101be careful not to include the pattern delimiter in the comment--perl has
102no way of knowing you did not intend to close the pattern early. See
103the C-comment deletion code in L<perlop>. Also note that anything inside
7651b971
KW
104a C<\Q...\E> stays unaffected by C</x>. And note that C</x> doesn't affect
105whether space interpretation within a single multi-character construct. For
106example in C<\x{...}>, regardless of the C</x> modifier, there can be no
9bb1f947 107spaces. Same for a L<quantifier|/Quantifiers> such as C<{3}> or
f9e949fd
KW
108C<{5,}>. Similarly, C<(?:...)> can't have a space between the C<?> and C<:>,
109but can between the C<(> and C<?>. Within any delimiters for such a
110construct, allowed spaces are not affected by C</x>, and depend on the
111construct. For example, C<\x{...}> can't have spaces because hexadecimal
112numbers don't have spaces in them. But, Unicode properties can have spaces, so
113in C<\p{...}> there can be spaces that follow the Unicode rules, for which see
9bb1f947 114L<perluniprops/Properties accessible through \p{} and \P{}>.
d74e8afc 115X</x>
a0d0e21e
LW
116
117=head2 Regular Expressions
118
04838cea
RGS
119=head3 Metacharacters
120
384f06ae 121The patterns used in Perl pattern matching evolved from those supplied in
14218588 122the Version 8 regex routines. (The routines are derived
19799a22
GS
123(distantly) from Henry Spencer's freely redistributable reimplementation
124of the V8 routines.) See L<Version 8 Regular Expressions> for
125details.
a0d0e21e
LW
126
127In particular the following metacharacters have their standard I<egrep>-ish
128meanings:
d74e8afc
ITB
129X<metacharacter>
130X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
131
a0d0e21e 132
f793d64a
KW
133 \ Quote the next metacharacter
134 ^ Match the beginning of the line
135 . Match any character (except newline)
136 $ Match the end of the line (or before newline at the end)
137 | Alternation
138 () Grouping
139 [] Bracketed Character class
a0d0e21e 140
14218588
GS
141By default, the "^" character is guaranteed to match only the
142beginning of the string, the "$" character only the end (or before the
143newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
144assumption that the string contains only one line. Embedded newlines
145will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 146string as a multi-line buffer, such that the "^" will match after any
0d520e8e
YO
147newline within the string (except if the newline is the last character in
148the string), and "$" will match before any newline. At the
a0d0e21e
LW
149cost of a little more overhead, you can do this by using the /m modifier
150on the pattern match operator. (Older programs did this by setting C<$*>,
f02c194e 151but this practice has been removed in perl 5.9.)
d74e8afc 152X<^> X<$> X</m>
a0d0e21e 153
14218588 154To simplify multi-line substitutions, the "." character never matches a
55497cff 155newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 156the string is a single line--even if it isn't.
d74e8afc 157X<.> X</s>
a0d0e21e 158
04838cea
RGS
159=head3 Quantifiers
160
a0d0e21e 161The following standard quantifiers are recognized:
d74e8afc 162X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e 163
f793d64a
KW
164 * Match 0 or more times
165 + Match 1 or more times
166 ? Match 1 or 0 times
167 {n} Match exactly n times
168 {n,} Match at least n times
169 {n,m} Match at least n but not more than m times
a0d0e21e
LW
170
171(If a curly bracket occurs in any other context, it is treated
b975c076 172as a regular character. In particular, the lower bound
527e91da
BB
173is not optional.) The "*" quantifier is equivalent to C<{0,}>, the "+"
174quantifier to C<{1,}>, and the "?" quantifier to C<{0,1}>. n and m are limited
d0b16107 175to non-negative integral values less than a preset limit defined when perl is built.
9c79236d
GS
176This is usually 32766 on the most common platforms. The actual limit can
177be seen in the error message generated by code such as this:
178
820475bd 179 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 180
54310121
PP
181By default, a quantified subpattern is "greedy", that is, it will match as
182many times as possible (given a particular starting location) while still
183allowing the rest of the pattern to match. If you want it to match the
184minimum number of times possible, follow the quantifier with a "?". Note
185that the meanings don't change, just the "greediness":
0d017f4d 186X<metacharacter> X<greedy> X<greediness>
d74e8afc 187X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e 188
f793d64a
KW
189 *? Match 0 or more times, not greedily
190 +? Match 1 or more times, not greedily
191 ?? Match 0 or 1 time, not greedily
192 {n}? Match exactly n times, not greedily
193 {n,}? Match at least n times, not greedily
194 {n,m}? Match at least n but not more than m times, not greedily
a0d0e21e 195
b9b4dddf
YO
196By default, when a quantified subpattern does not allow the rest of the
197overall pattern to match, Perl will backtrack. However, this behaviour is
0d017f4d 198sometimes undesirable. Thus Perl provides the "possessive" quantifier form
b9b4dddf
YO
199as well.
200
f793d64a
KW
201 *+ Match 0 or more times and give nothing back
202 ++ Match 1 or more times and give nothing back
203 ?+ Match 0 or 1 time and give nothing back
204 {n}+ Match exactly n times and give nothing back (redundant)
205 {n,}+ Match at least n times and give nothing back
206 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
207
208For instance,
209
210 'aaaa' =~ /a++a/
211
212will never match, as the C<a++> will gobble up all the C<a>'s in the
213string and won't leave any for the remaining part of the pattern. This
214feature can be extremely useful to give perl hints about where it
215shouldn't backtrack. For instance, the typical "match a double-quoted
216string" problem can be most efficiently performed when written as:
217
218 /"(?:[^"\\]++|\\.)*+"/
219
0d017f4d 220as we know that if the final quote does not match, backtracking will not
b9b4dddf
YO
221help. See the independent subexpression C<< (?>...) >> for more details;
222possessive quantifiers are just syntactic sugar for that construct. For
223instance the above example could also be written as follows:
224
225 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
226
04838cea
RGS
227=head3 Escape sequences
228
5f05dabc 229Because patterns are processed as double quoted strings, the following
a0d0e21e
LW
230also work:
231
f793d64a
KW
232 \t tab (HT, TAB)
233 \n newline (LF, NL)
234 \r return (CR)
235 \f form feed (FF)
236 \a alarm (bell) (BEL)
237 \e escape (think troff) (ESC)
f793d64a 238 \cK control char (example: VT)
dc0d9c48 239 \x{}, \x00 character whose ordinal is the given hexadecimal number
fb121860 240 \N{name} named Unicode character or character sequence
f793d64a 241 \N{U+263D} Unicode character (example: FIRST QUARTER MOON)
f0a2b745 242 \o{}, \000 character whose ordinal is the given octal number
f793d64a
KW
243 \l lowercase next char (think vi)
244 \u uppercase next char (think vi)
245 \L lowercase till \E (think vi)
246 \U uppercase till \E (think vi)
247 \Q quote (disable) pattern metacharacters till \E
248 \E end either case modification or quoted section, think vi
a0d0e21e 249
9bb1f947 250Details are in L<perlop/Quote and Quote-like Operators>.
1d2dff63 251
e1d1eefb 252=head3 Character Classes and other Special Escapes
04838cea 253
a0d0e21e 254In addition, Perl defines the following:
d0b16107 255X<\g> X<\k> X<\K> X<backreference>
a0d0e21e 256
f793d64a
KW
257 Sequence Note Description
258 [...] [1] Match a character according to the rules of the
259 bracketed character class defined by the "...".
260 Example: [a-z] matches "a" or "b" or "c" ... or "z"
261 [[:...:]] [2] Match a character according to the rules of the POSIX
262 character class "..." within the outer bracketed
263 character class. Example: [[:upper:]] matches any
264 uppercase character.
d35dd6c6
KW
265 \w [3] Match a "word" character (alphanumeric plus "_", plus
266 other connector punctuation chars plus Unicode
267 marks
f793d64a
KW
268 \W [3] Match a non-"word" character
269 \s [3] Match a whitespace character
270 \S [3] Match a non-whitespace character
271 \d [3] Match a decimal digit character
272 \D [3] Match a non-digit character
273 \pP [3] Match P, named property. Use \p{Prop} for longer names
274 \PP [3] Match non-P
275 \X [4] Match Unicode "eXtended grapheme cluster"
276 \C Match a single C-language char (octet) even if that is
277 part of a larger UTF-8 character. Thus it breaks up
278 characters into their UTF-8 bytes, so you may end up
279 with malformed pieces of UTF-8. Unsupported in
280 lookbehind.
c27a5cfe 281 \1 [5] Backreference to a specific capture group or buffer.
f793d64a
KW
282 '1' may actually be any positive integer.
283 \g1 [5] Backreference to a specific or previous group,
284 \g{-1} [5] The number may be negative indicating a relative
c27a5cfe 285 previous group and may optionally be wrapped in
f793d64a
KW
286 curly brackets for safer parsing.
287 \g{name} [5] Named backreference
288 \k<name> [5] Named backreference
289 \K [6] Keep the stuff left of the \K, don't include it in $&
290 \N [7] Any character but \n (experimental). Not affected by
291 /s modifier
292 \v [3] Vertical whitespace
293 \V [3] Not vertical whitespace
294 \h [3] Horizontal whitespace
295 \H [3] Not horizontal whitespace
296 \R [4] Linebreak
e1d1eefb 297
9bb1f947
KW
298=over 4
299
300=item [1]
301
302See L<perlrecharclass/Bracketed Character Classes> for details.
df225385 303
9bb1f947 304=item [2]
b8c5462f 305
9bb1f947 306See L<perlrecharclass/POSIX Character Classes> for details.
b8c5462f 307
9bb1f947 308=item [3]
5496314a 309
9bb1f947 310See L<perlrecharclass/Backslash sequences> for details.
5496314a 311
9bb1f947 312=item [4]
5496314a 313
9bb1f947 314See L<perlrebackslash/Misc> for details.
d0b16107 315
9bb1f947 316=item [5]
b8c5462f 317
c27a5cfe 318See L</Capture groups> below for details.
93733859 319
9bb1f947 320=item [6]
b8c5462f 321
9bb1f947
KW
322See L</Extended Patterns> below for details.
323
324=item [7]
325
326Note that C<\N> has two meanings. When of the form C<\N{NAME}>, it matches the
fb121860
KW
327character or character sequence whose name is C<NAME>; and similarly
328when of the form C<\N{U+I<hex>}>, it matches the character whose Unicode
329code point is I<hex>. Otherwise it matches any character but C<\n>.
9bb1f947
KW
330
331=back
d0b16107 332
04838cea
RGS
333=head3 Assertions
334
a0d0e21e 335Perl defines the following zero-width assertions:
d74e8afc
ITB
336X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
337X<regexp, zero-width assertion>
338X<regular expression, zero-width assertion>
339X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e 340
9bb1f947
KW
341 \b Match a word boundary
342 \B Match except at a word boundary
343 \A Match only at beginning of string
344 \Z Match only at end of string, or before newline at the end
345 \z Match only at end of string
346 \G Match only at pos() (e.g. at the end-of-match position
9da458fc 347 of prior m//g)
a0d0e21e 348
14218588 349A word boundary (C<\b>) is a spot between two characters
19799a22
GS
350that has a C<\w> on one side of it and a C<\W> on the other side
351of it (in either order), counting the imaginary characters off the
352beginning and end of the string as matching a C<\W>. (Within
353character classes C<\b> represents backspace rather than a word
354boundary, just as it normally does in any double-quoted string.)
355The C<\A> and C<\Z> are just like "^" and "$", except that they
356won't match multiple times when the C</m> modifier is used, while
357"^" and "$" will match at every internal line boundary. To match
358the actual end of the string and not ignore an optional trailing
359newline, use C<\z>.
d74e8afc 360X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
361
362The C<\G> assertion can be used to chain global matches (using
363C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
364It is also useful when writing C<lex>-like scanners, when you have
365several patterns that you want to match against consequent substrings
366of your string, see the previous reference. The actual location
367where C<\G> will match can also be influenced by using C<pos()> as
58e23c8d
YO
368an lvalue: see L<perlfunc/pos>. Note that the rule for zero-length
369matches is modified somewhat, in that contents to the left of C<\G> is
370not counted when determining the length of the match. Thus the following
371will not match forever:
d74e8afc 372X<\G>
c47ff5f1 373
e761bb84
CO
374 my $string = 'ABC';
375 pos($string) = 1;
376 while ($string =~ /(.\G)/g) {
377 print $1;
378 }
58e23c8d
YO
379
380It will print 'A' and then terminate, as it considers the match to
381be zero-width, and thus will not match at the same position twice in a
382row.
383
384It is worth noting that C<\G> improperly used can result in an infinite
385loop. Take care when using patterns that include C<\G> in an alternation.
386
c27a5cfe 387=head3 Capture groups
04838cea 388
c27a5cfe
KW
389The bracketing construct C<( ... )> creates capture groups (also referred to as
390capture buffers). To refer to the current contents of a group later on, within
d8b950dc
KW
391the same pattern, use C<\g1> (or C<\g{1}>) for the first, C<\g2> (or C<\g{2}>)
392for the second, and so on.
393This is called a I<backreference>.
d74e8afc 394X<regex, capture buffer> X<regexp, capture buffer>
c27a5cfe 395X<regex, capture group> X<regexp, capture group>
d74e8afc 396X<regular expression, capture buffer> X<backreference>
c27a5cfe 397X<regular expression, capture group> X<backreference>
1f1031fe 398X<\g{1}> X<\g{-1}> X<\g{name}> X<relative backreference> X<named backreference>
d8b950dc
KW
399X<named capture buffer> X<regular expression, named capture buffer>
400X<named capture group> X<regular expression, named capture group>
401X<%+> X<$+{name}> X<< \k<name> >>
402There is no limit to the number of captured substrings that you may use.
403Groups are numbered with the leftmost open parenthesis being number 1, etc. If
404a group did not match, the associated backreference won't match either. (This
405can happen if the group is optional, or in a different branch of an
406alternation.)
407You can omit the C<"g">, and write C<"\1">, etc, but there are some issues with
408this form, described below.
409
410You can also refer to capture groups relatively, by using a negative number, so
411that C<\g-1> and C<\g{-1}> both refer to the immediately preceding capture
412group, and C<\g-2> and C<\g{-2}> both refer to the group before it. For
413example:
5624f11d
YO
414
415 /
c27a5cfe
KW
416 (Y) # group 1
417 ( # group 2
418 (X) # group 3
419 \g{-1} # backref to group 3
420 \g{-3} # backref to group 1
5624f11d
YO
421 )
422 /x
423
d8b950dc
KW
424would match the same as C</(Y) ( (X) \g3 \g1 )/x>. This allows you to
425interpolate regexes into larger regexes and not have to worry about the
426capture groups being renumbered.
427
428You can dispense with numbers altogether and create named capture groups.
429The notation is C<(?E<lt>I<name>E<gt>...)> to declare and C<\g{I<name>}> to
430reference. (To be compatible with .Net regular expressions, C<\g{I<name>}> may
431also be written as C<\k{I<name>}>, C<\kE<lt>I<name>E<gt>> or C<\k'I<name>'>.)
432I<name> must not begin with a number, nor contain hyphens.
433When different groups within the same pattern have the same name, any reference
434to that name assumes the leftmost defined group. Named groups count in
435absolute and relative numbering, and so can also be referred to by those
436numbers.
437(It's possible to do things with named capture groups that would otherwise
438require C<(??{})>.)
439
440Capture group contents are dynamically scoped and available to you outside the
441pattern until the end of the enclosing block or until the next successful
442match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
443You can refer to them by absolute number (using C<"$1"> instead of C<"\g1">,
444etc); or by name via the C<%+> hash, using C<"$+{I<name>}">.
445
446Braces are required in referring to named capture groups, but are optional for
447absolute or relative numbered ones. Braces are safer when creating a regex by
448concatenating smaller strings. For example if you have C<qr/$a$b/>, and C<$a>
449contained C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which
450is probably not what you intended.
451
452The C<\g> and C<\k> notations were introduced in Perl 5.10.0. Prior to that
453there were no named nor relative numbered capture groups. Absolute numbered
454groups were referred to using C<\1>, C<\2>, etc, and this notation is still
455accepted (and likely always will be). But it leads to some ambiguities if
456there are more than 9 capture groups, as C<\10> could mean either the tenth
457capture group, or the character whose ordinal in octal is 010 (a backspace in
458ASCII). Perl resolves this ambiguity by interpreting C<\10> as a backreference
459only if at least 10 left parentheses have opened before it. Likewise C<\11> is
460a backreference only if at least 11 left parentheses have opened before it.
e1f120a9
KW
461And so on. C<\1> through C<\9> are always interpreted as backreferences.
462There are several examples below that illustrate these perils. You can avoid
463the ambiguity by always using C<\g{}> or C<\g> if you mean capturing groups;
464and for octal constants always using C<\o{}>, or for C<\077> and below, using 3
465digits padded with leading zeros, since a leading zero implies an octal
466constant.
d8b950dc
KW
467
468The C<\I<digit>> notation also works in certain circumstances outside
469the pattern. See L</Warning on \1 Instead of $1> below for details.)
81714fb9 470
14218588 471Examples:
a0d0e21e
LW
472
473 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
474
d8b950dc 475 /(.)\g1/ # find first doubled char
81714fb9
YO
476 and print "'$1' is the first doubled character\n";
477
478 /(?<char>.)\k<char>/ # ... a different way
479 and print "'$+{char}' is the first doubled character\n";
480
d8b950dc 481 /(?'char'.)\g1/ # ... mix and match
81714fb9 482 and print "'$1' is the first doubled character\n";
c47ff5f1 483
14218588 484 if (/Time: (..):(..):(..)/) { # parse out values
f793d64a
KW
485 $hours = $1;
486 $minutes = $2;
487 $seconds = $3;
a0d0e21e 488 }
c47ff5f1 489
9d860678
KW
490 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\g10/ # \g10 is a backreference
491 /(.)(.)(.)(.)(.)(.)(.)(.)(.)\10/ # \10 is octal
492 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\10/ # \10 is a backreference
493 /((.)(.)(.)(.)(.)(.)(.)(.)(.))\010/ # \010 is octal
494
495 $a = '(.)\1'; # Creates problems when concatenated.
496 $b = '(.)\g{1}'; # Avoids the problems.
497 "aa" =~ /${a}/; # True
498 "aa" =~ /${b}/; # True
499 "aa0" =~ /${a}0/; # False!
500 "aa0" =~ /${b}0/; # True
dc0d9c48
KW
501 "aa\x08" =~ /${a}0/; # True!
502 "aa\x08" =~ /${b}0/; # False
9d860678 503
14218588
GS
504Several special variables also refer back to portions of the previous
505match. C<$+> returns whatever the last bracket match matched.
506C<$&> returns the entire matched string. (At one point C<$0> did
507also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
508everything before the matched string. C<$'> returns everything
509after the matched string. And C<$^N> contains whatever was matched by
510the most-recently closed group (submatch). C<$^N> can be used in
511extended patterns (see below), for example to assign a submatch to a
81714fb9 512variable.
d74e8afc 513X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 514
d8b950dc
KW
515These special variables, like the C<%+> hash and the numbered match variables
516(C<$1>, C<$2>, C<$3>, etc.) are dynamically scoped
14218588
GS
517until the end of the enclosing block or until the next successful
518match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
519X<$+> X<$^N> X<$&> X<$`> X<$'>
520X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
521
0d017f4d 522B<NOTE>: Failed matches in Perl do not reset the match variables,
5146ce24 523which makes it easier to write code that tests for a series of more
665e98b9
JH
524specific cases and remembers the best match.
525
14218588
GS
526B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
527C<$'> anywhere in the program, it has to provide them for every
528pattern match. This may substantially slow your program. Perl
d8b950dc 529uses the same mechanism to produce C<$1>, C<$2>, etc, so you also pay a
14218588
GS
530price for each pattern that contains capturing parentheses. (To
531avoid this cost while retaining the grouping behaviour, use the
532extended regular expression C<(?: ... )> instead.) But if you never
533use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
534parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
535if you can, but if you can't (and some algorithms really appreciate
536them), once you've used them once, use them at will, because you've
537already paid the price. As of 5.005, C<$&> is not so costly as the
538other two.
d74e8afc 539X<$&> X<$`> X<$'>
68dc0745 540
99d59c4d 541As a workaround for this problem, Perl 5.10.0 introduces C<${^PREMATCH}>,
cde0cee5
YO
542C<${^MATCH}> and C<${^POSTMATCH}>, which are equivalent to C<$`>, C<$&>
543and C<$'>, B<except> that they are only guaranteed to be defined after a
87e95b7f 544successful match that was executed with the C</p> (preserve) modifier.
cde0cee5
YO
545The use of these variables incurs no global performance penalty, unlike
546their punctuation char equivalents, however at the trade-off that you
547have to tell perl when you want to use them.
87e95b7f 548X</p> X<p modifier>
cde0cee5 549
9d727203
KW
550=head2 Quoting metacharacters
551
19799a22
GS
552Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
553C<\w>, C<\n>. Unlike some other regular expression languages, there
554are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 555that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
556interpreted as a literal character, not a metacharacter. This was
557once used in a common idiom to disable or quote the special meanings
558of regular expression metacharacters in a string that you want to
36bbe248 559use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
560
561 $pattern =~ s/(\W)/\\$1/g;
562
f1cbbd6e 563(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
564Today it is more common to use the quotemeta() function or the C<\Q>
565metaquoting escape sequence to disable all metacharacters' special
566meanings like this:
a0d0e21e
LW
567
568 /$unquoted\Q$quoted\E$unquoted/
569
9da458fc
IZ
570Beware that if you put literal backslashes (those not inside
571interpolated variables) between C<\Q> and C<\E>, double-quotish
572backslash interpolation may lead to confusing results. If you
573I<need> to use literal backslashes within C<\Q...\E>,
574consult L<perlop/"Gory details of parsing quoted constructs">.
575
19799a22
GS
576=head2 Extended Patterns
577
14218588
GS
578Perl also defines a consistent extension syntax for features not
579found in standard tools like B<awk> and B<lex>. The syntax is a
580pair of parentheses with a question mark as the first thing within
581the parentheses. The character after the question mark indicates
582the extension.
19799a22 583
14218588
GS
584The stability of these extensions varies widely. Some have been
585part of the core language for many years. Others are experimental
586and may change without warning or be completely removed. Check
587the documentation on an individual feature to verify its current
588status.
19799a22 589
14218588
GS
590A question mark was chosen for this and for the minimal-matching
591construct because 1) question marks are rare in older regular
592expressions, and 2) whenever you see one, you should stop and
593"question" exactly what is going on. That's psychology...
a0d0e21e
LW
594
595=over 10
596
cc6b7395 597=item C<(?#text)>
d74e8afc 598X<(?#)>
a0d0e21e 599
14218588 600A comment. The text is ignored. If the C</x> modifier enables
19799a22 601whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
602the comment as soon as it sees a C<)>, so there is no way to put a literal
603C<)> in the comment.
a0d0e21e 604
cfaf538b 605=item C<(?adlupimsx-imsx)>
fb85c044 606
cfaf538b 607=item C<(?^alupimsx)>
fb85c044 608X<(?)> X<(?^)>
19799a22 609
0b6d1084
JH
610One or more embedded pattern-match modifiers, to be turned on (or
611turned off, if preceded by C<->) for the remainder of the pattern or
fb85c044
KW
612the remainder of the enclosing pattern group (if any).
613
fb85c044 614This is particularly useful for dynamic patterns, such as those read in from a
0d017f4d
WL
615configuration file, taken from an argument, or specified in a table
616somewhere. Consider the case where some patterns want to be case
617sensitive and some do not: The case insensitive ones merely need to
618include C<(?i)> at the front of the pattern. For example:
19799a22
GS
619
620 $pattern = "foobar";
5d458dd8 621 if ( /$pattern/i ) { }
19799a22
GS
622
623 # more flexible:
624
625 $pattern = "(?i)foobar";
5d458dd8 626 if ( /$pattern/ ) { }
19799a22 627
0b6d1084 628These modifiers are restored at the end of the enclosing group. For example,
19799a22 629
d8b950dc 630 ( (?i) blah ) \s+ \g1
19799a22 631
0d017f4d
WL
632will match C<blah> in any case, some spaces, and an exact (I<including the case>!)
633repetition of the previous word, assuming the C</x> modifier, and no C</i>
634modifier outside this group.
19799a22 635
8eb5594e
DR
636These modifiers do not carry over into named subpatterns called in the
637enclosing group. In other words, a pattern such as C<((?i)(&NAME))> does not
638change the case-sensitivity of the "NAME" pattern.
639
dc925305
KW
640Any of these modifiers can be set to apply globally to all regular
641expressions compiled within the scope of a C<use re>. See
a0bbd6ff 642L<re/"'/flags' mode">.
dc925305 643
9de15fec
KW
644Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
645after the C<"?"> is a shorthand equivalent to C<d-imsx>. Flags (except
646C<"d">) may follow the caret to override it.
647But a minus sign is not legal with it.
648
cfaf538b
KW
649Also, starting in Perl 5.14, are modifiers C<"a">, C<"d">, C<"l">, and
650C<"u">, which for 5.14 may not be used as suffix modifiers.
9de15fec
KW
651
652C<"l"> means to use a locale (see L<perllocale>) when pattern matching.
653The locale used will be the one in effect at the time of execution of
654the pattern match. This may not be the same as the compilation-time
655locale, and can differ from one match to another if there is an
656intervening call of the
657L<setlocale() function|perllocale/The setlocale function>.
658This modifier is automatically set if the regular expression is compiled
17580e7a
KW
659within the scope of a C<"use locale"> pragma.
660Perl only allows single-byte locales. This means that code points above
661255 are treated as Unicode no matter what locale is in effect.
662Under Unicode rules, there are a few case-insensitive matches that cross the
663boundary 255/256 boundary. These are disallowed. For example,
6640xFF does not caselessly match the character at 0x178, LATIN CAPITAL
665LETTER Y WITH DIAERESIS, because 0xFF may not be LATIN SMALL LETTER Y
666in the current locale, and Perl has no way of knowing if that character
667even exists in the locale, much less what code point it is.
9de15fec 668
a12cf05f 669C<"u"> means to use Unicode semantics when pattern matching. It is
20db7501
KW
670automatically set if the regular expression is encoded in utf8, or is
671compiled within the scope of a
672L<C<"use feature 'unicode_strings">|feature> pragma (and isn't also in
673the scope of L<C<"use locale">|locale> nor L<C<"use bytes">|bytes>
674pragmas. On ASCII platforms, the code points between 128 and 255 take on their
675Latin-1 (ISO-8859-1) meanings (which are the same as Unicode's), whereas
676in strict ASCII their meanings are undefined. Thus the platform
677effectively becomes a Unicode platform. The ASCII characters remain as
678ASCII characters (since ASCII is a subset of Latin-1 and Unicode). For
dc925305 679example, when this option is not on, on a non-utf8 string, C<"\w">
20db7501
KW
680matches precisely C<[A-Za-z0-9_]>. When the option is on, it matches
681not just those, but all the Latin-1 word characters (such as an "n" with
682a tilde). On EBCDIC platforms, which already are equivalent to Latin-1,
683this modifier changes behavior only when the C<"/i"> modifier is also
684specified, and affects only two characters, giving them full Unicode
685semantics: the C<MICRO SIGN> will match the Greek capital and
686small letters C<MU>; otherwise not; and the C<LATIN CAPITAL LETTER SHARP
687S> will match any of C<SS>, C<Ss>, C<sS>, and C<ss>, otherwise not.
688(This last case is buggy, however.)
9de15fec 689
dc925305
KW
690C<"a"> is the same as C<"u">, except that C<\d>, C<\s>, C<\w>, and the
691Posix character classes are restricted to matching in the ASCII range
692only. That is, with this modifier, C<\d> always means precisely the
693digits C<"0"> to C<"9">; C<\s> means the five characters C<[ \f\n\r\t]>;
694C<\w> means the 63 characters C<[A-Za-z0-9_]>; and likewise, all the
cfaf538b
KW
695Posix classes such as C<[[:print:]]> match only the appropriate
696ASCII-range characters. As you would expect, this modifier causes, for
dc925305
KW
697example, C<\D> to mean the same thing as C<[^0-9]>; in fact, all
698non-ASCII characters match C<\D>, C<\S>, and C<\W>. C<\b> still means
699to match at the boundary between C<\w> and C<\W>, using the C<"a">
700definitions of them (similarly for C<\B>). Otherwise, C<"a"> behaves
701like the C<"u"> modifier, in that case-insensitive matching uses Unicode
702semantics; for example, "k" will match the Unicode C<\N{KELVIN SIGN}>
703under C</i> matching, and code points in the Latin1 range, above ASCII
704will have Unicode semantics when it comes to case-insensitive matching.
2f7f8cb1
KW
705But writing two in "a"'s in a row will increase its effect, causing the
706Kelvin sign and all other non-ASCII characters to not match any ASCII
707character under C</i> matching.
cfaf538b 708
9de15fec
KW
709C<"d"> means to use the traditional Perl pattern matching behavior.
710This is dualistic (hence the name C<"d">, which also could stand for
e40e74fe
KW
711"depends"). When this is in effect, Perl matches according to the
712platform's native character set rules unless there is something that
713indicates to use Unicode rules. If either the target string or the
714pattern itself is encoded in UTF-8, Unicode rules are used. Also, if
715the pattern contains Unicode-only features, such as code points above
716255, C<\p()> Unicode properties or C<\N{}> Unicode names, Unicode rules
717will be used. It is automatically selected by default if the regular
718expression is compiled neither within the scope of a C<"use locale">
719pragma nor a <C<"use feature 'unicode_strings"> pragma.
720This behavior causes a number of glitches, see
721L<perlunicode/The "Unicode Bug">.
9de15fec 722
dc925305 723Note that the C<a>, C<d>, C<l>, C<p>, and C<u> modifiers are special in
e1d8d8ac 724that they can only be enabled, not disabled, and the C<a>, C<d>, C<l>, and
dc925305 725C<u> modifiers are mutually exclusive: specifying one de-specifies the
bdc22dd5 726others, and a maximum of one may appear in the construct. Thus, for
315f11aa 727example, C<(?-p)>, C<(?-d:...)>, and C<(?dl:...)> will warn when
bdc22dd5 728compiled under C<use warnings>.
9de15fec
KW
729
730Note also that the C<p> modifier is special in that its presence
731anywhere in a pattern has a global effect.
cde0cee5 732
5a964f20 733=item C<(?:pattern)>
d74e8afc 734X<(?:)>
a0d0e21e 735
cfaf538b 736=item C<(?adluimsx-imsx:pattern)>
ca9dfc88 737
cfaf538b 738=item C<(?^aluimsx:pattern)>
fb85c044
KW
739X<(?^:)>
740
5a964f20
TC
741This is for clustering, not capturing; it groups subexpressions like
742"()", but doesn't make backreferences as "()" does. So
a0d0e21e 743
5a964f20 744 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
745
746is like
747
5a964f20 748 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 749
19799a22
GS
750but doesn't spit out extra fields. It's also cheaper not to capture
751characters if you don't need to.
a0d0e21e 752
19799a22 753Any letters between C<?> and C<:> act as flags modifiers as with
cfaf538b 754C<(?adluimsx-imsx)>. For example,
ca9dfc88
IZ
755
756 /(?s-i:more.*than).*million/i
757
14218588 758is equivalent to the more verbose
ca9dfc88
IZ
759
760 /(?:(?s-i)more.*than).*million/i
761
fb85c044 762Starting in Perl 5.14, a C<"^"> (caret or circumflex accent) immediately
9de15fec
KW
763after the C<"?"> is a shorthand equivalent to C<d-imsx>. Any positive
764flags (except C<"d">) may follow the caret, so
fb85c044
KW
765
766 (?^x:foo)
767
768is equivalent to
769
770 (?x-ims:foo)
771
772The caret tells Perl that this cluster doesn't inherit the flags of any
9de15fec 773surrounding pattern, but to go back to the system defaults (C<d-imsx>),
fb85c044
KW
774modified by any flags specified.
775
776The caret allows for simpler stringification of compiled regular
777expressions. These look like
778
779 (?^:pattern)
780
781with any non-default flags appearing between the caret and the colon.
782A test that looks at such stringification thus doesn't need to have the
783system default flags hard-coded in it, just the caret. If new flags are
784added to Perl, the meaning of the caret's expansion will change to include
785the default for those flags, so the test will still work, unchanged.
786
787Specifying a negative flag after the caret is an error, as the flag is
788redundant.
789
790Mnemonic for C<(?^...)>: A fresh beginning since the usual use of a caret is
791to match at the beginning.
792
594d7033
YO
793=item C<(?|pattern)>
794X<(?|)> X<Branch reset>
795
796This is the "branch reset" pattern, which has the special property
c27a5cfe 797that the capture groups are numbered from the same starting point
99d59c4d 798in each alternation branch. It is available starting from perl 5.10.0.
4deaaa80 799
c27a5cfe 800Capture groups are numbered from left to right, but inside this
693596a8 801construct the numbering is restarted for each branch.
4deaaa80 802
c27a5cfe 803The numbering within each branch will be as normal, and any groups
4deaaa80
PJ
804following this construct will be numbered as though the construct
805contained only one branch, that being the one with the most capture
c27a5cfe 806groups in it.
4deaaa80
PJ
807
808This construct will be useful when you want to capture one of a
809number of alternative matches.
810
811Consider the following pattern. The numbers underneath show in
c27a5cfe 812which group the captured content will be stored.
594d7033
YO
813
814
815 # before ---------------branch-reset----------- after
816 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
817 # 1 2 2 3 2 3 4
818
ab106183
A
819Be careful when using the branch reset pattern in combination with
820named captures. Named captures are implemented as being aliases to
c27a5cfe 821numbered groups holding the captures, and that interferes with the
ab106183
A
822implementation of the branch reset pattern. If you are using named
823captures in a branch reset pattern, it's best to use the same names,
824in the same order, in each of the alternations:
825
826 /(?| (?<a> x ) (?<b> y )
827 | (?<a> z ) (?<b> w )) /x
828
829Not doing so may lead to surprises:
830
831 "12" =~ /(?| (?<a> \d+ ) | (?<b> \D+))/x;
832 say $+ {a}; # Prints '12'
833 say $+ {b}; # *Also* prints '12'.
834
c27a5cfe
KW
835The problem here is that both the group named C<< a >> and the group
836named C<< b >> are aliases for the group belonging to C<< $1 >>.
90a18110 837
ee9b8eae
YO
838=item Look-Around Assertions
839X<look-around assertion> X<lookaround assertion> X<look-around> X<lookaround>
840
841Look-around assertions are zero width patterns which match a specific
842pattern without including it in C<$&>. Positive assertions match when
843their subpattern matches, negative assertions match when their subpattern
844fails. Look-behind matches text up to the current match position,
845look-ahead matches text following the current match position.
846
847=over 4
848
5a964f20 849=item C<(?=pattern)>
d74e8afc 850X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 851
19799a22 852A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
853matches a word followed by a tab, without including the tab in C<$&>.
854
5a964f20 855=item C<(?!pattern)>
d74e8afc 856X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 857
19799a22 858A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 859matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
860however that look-ahead and look-behind are NOT the same thing. You cannot
861use this for look-behind.
7b8d334a 862
5a964f20 863If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
864will not do what you want. That's because the C<(?!foo)> is just saying that
865the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
866match. You would have to do something like C</(?!foo)...bar/> for that. We
867say "like" because there's the case of your "bar" not having three characters
868before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
869Sometimes it's still easier just to say:
a0d0e21e 870
a3cb178b 871 if (/bar/ && $` !~ /foo$/)
a0d0e21e 872
19799a22 873For look-behind see below.
c277df42 874
ee9b8eae
YO
875=item C<(?<=pattern)> C<\K>
876X<(?<=)> X<look-behind, positive> X<lookbehind, positive> X<\K>
c277df42 877
c47ff5f1 878A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
879matches a word that follows a tab, without including the tab in C<$&>.
880Works only for fixed-width look-behind.
c277df42 881
ee9b8eae
YO
882There is a special form of this construct, called C<\K>, which causes the
883regex engine to "keep" everything it had matched prior to the C<\K> and
884not include it in C<$&>. This effectively provides variable length
885look-behind. The use of C<\K> inside of another look-around assertion
886is allowed, but the behaviour is currently not well defined.
887
c62285ac 888For various reasons C<\K> may be significantly more efficient than the
ee9b8eae
YO
889equivalent C<< (?<=...) >> construct, and it is especially useful in
890situations where you want to efficiently remove something following
891something else in a string. For instance
892
893 s/(foo)bar/$1/g;
894
895can be rewritten as the much more efficient
896
897 s/foo\Kbar//g;
898
5a964f20 899=item C<(?<!pattern)>
d74e8afc 900X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 901
19799a22
GS
902A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
903matches any occurrence of "foo" that does not follow "bar". Works
904only for fixed-width look-behind.
c277df42 905
ee9b8eae
YO
906=back
907
81714fb9
YO
908=item C<(?'NAME'pattern)>
909
910=item C<< (?<NAME>pattern) >>
911X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
912
c27a5cfe 913A named capture group. Identical in every respect to normal capturing
90a18110 914parentheses C<()> but for the additional fact that C<%+> or C<%-> may be
1b2010d5 915used after a successful match to refer to a named group. See L<perlvar>
90a18110 916for more details on the C<%+> and C<%-> hashes.
81714fb9 917
c27a5cfe
KW
918If multiple distinct capture groups have the same name then the
919$+{NAME} will refer to the leftmost defined group in the match.
81714fb9 920
0d017f4d 921The forms C<(?'NAME'pattern)> and C<< (?<NAME>pattern) >> are equivalent.
81714fb9
YO
922
923B<NOTE:> While the notation of this construct is the same as the similar
c27a5cfe 924function in .NET regexes, the behavior is not. In Perl the groups are
81714fb9
YO
925numbered sequentially regardless of being named or not. Thus in the
926pattern
927
928 /(x)(?<foo>y)(z)/
929
930$+{foo} will be the same as $2, and $3 will contain 'z' instead of
931the opposite which is what a .NET regex hacker might expect.
932
1f1031fe
YO
933Currently NAME is restricted to simple identifiers only.
934In other words, it must match C</^[_A-Za-z][_A-Za-z0-9]*\z/> or
935its Unicode extension (see L<utf8>),
936though it isn't extended by the locale (see L<perllocale>).
81714fb9 937
1f1031fe 938B<NOTE:> In order to make things easier for programmers with experience
ae5648b3 939with the Python or PCRE regex engines, the pattern C<< (?PE<lt>NAMEE<gt>pattern) >>
0d017f4d 940may be used instead of C<< (?<NAME>pattern) >>; however this form does not
64c5a566 941support the use of single quotes as a delimiter for the name.
81714fb9 942
1f1031fe
YO
943=item C<< \k<NAME> >>
944
945=item C<< \k'NAME' >>
81714fb9
YO
946
947Named backreference. Similar to numeric backreferences, except that
948the group is designated by name and not number. If multiple groups
949have the same name then it refers to the leftmost defined group in
950the current match.
951
0d017f4d 952It is an error to refer to a name not defined by a C<< (?<NAME>) >>
81714fb9
YO
953earlier in the pattern.
954
955Both forms are equivalent.
956
1f1031fe 957B<NOTE:> In order to make things easier for programmers with experience
0d017f4d 958with the Python or PCRE regex engines, the pattern C<< (?P=NAME) >>
64c5a566 959may be used instead of C<< \k<NAME> >>.
1f1031fe 960
cc6b7395 961=item C<(?{ code })>
d74e8afc 962X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 963
19799a22 964B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
965experimental, and may be changed without notice. Code executed that
966has side effects may not perform identically from version to version
967due to the effect of future optimisations in the regex engine.
c277df42 968
cc46d5f2 969This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
970always succeeds, and its C<code> is not interpolated. Currently,
971the rules to determine where the C<code> ends are somewhat convoluted.
972
77ea4f6d
JV
973This feature can be used together with the special variable C<$^N> to
974capture the results of submatches in variables without having to keep
975track of the number of nested parentheses. For example:
976
977 $_ = "The brown fox jumps over the lazy dog";
978 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
979 print "color = $color, animal = $animal\n";
980
754091cb
RGS
981Inside the C<(?{...})> block, C<$_> refers to the string the regular
982expression is matching against. You can also use C<pos()> to know what is
fa11829f 983the current position of matching within this string.
754091cb 984
19799a22
GS
985The C<code> is properly scoped in the following sense: If the assertion
986is backtracked (compare L<"Backtracking">), all changes introduced after
987C<local>ization are undone, so that
b9ac3b5b
GS
988
989 $_ = 'a' x 8;
5d458dd8 990 m<
f793d64a 991 (?{ $cnt = 0 }) # Initialize $cnt.
b9ac3b5b 992 (
5d458dd8 993 a
b9ac3b5b 994 (?{
f793d64a 995 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
b9ac3b5b 996 })
5d458dd8 997 )*
b9ac3b5b 998 aaaa
f793d64a
KW
999 (?{ $res = $cnt }) # On success copy to
1000 # non-localized location.
b9ac3b5b
GS
1001 >x;
1002
0d017f4d 1003will set C<$res = 4>. Note that after the match, C<$cnt> returns to the globally
14218588 1004introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
1005are unwound.
1006
19799a22
GS
1007This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
1008switch. If I<not> used in this way, the result of evaluation of
1009C<code> is put into the special variable C<$^R>. This happens
1010immediately, so C<$^R> can be used from other C<(?{ code })> assertions
1011inside the same regular expression.
b9ac3b5b 1012
19799a22
GS
1013The assignment to C<$^R> above is properly localized, so the old
1014value of C<$^R> is restored if the assertion is backtracked; compare
1015L<"Backtracking">.
b9ac3b5b 1016
19799a22
GS
1017For reasons of security, this construct is forbidden if the regular
1018expression involves run-time interpolation of variables, unless the
1019perilous C<use re 'eval'> pragma has been used (see L<re>), or the
1020variables contain results of C<qr//> operator (see
9bb1f947 1021L<perlop/"qr/STRINGE<sol>msixpo">).
871b0233 1022
0d017f4d 1023This restriction is due to the wide-spread and remarkably convenient
19799a22 1024custom of using run-time determined strings as patterns. For example:
871b0233
IZ
1025
1026 $re = <>;
1027 chomp $re;
1028 $string =~ /$re/;
1029
14218588
GS
1030Before Perl knew how to execute interpolated code within a pattern,
1031this operation was completely safe from a security point of view,
1032although it could raise an exception from an illegal pattern. If
1033you turn on the C<use re 'eval'>, though, it is no longer secure,
1034so you should only do so if you are also using taint checking.
1035Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 1036compartment. See L<perlsec> for details about both these mechanisms.
871b0233 1037
e95d7314
GG
1038B<WARNING>: Use of lexical (C<my>) variables in these blocks is
1039broken. The result is unpredictable and will make perl unstable. The
1040workaround is to use global (C<our>) variables.
1041
8525cfae
FC
1042B<WARNING>: In perl 5.12.x and earlier, the regex engine
1043was not re-entrant, so interpolated code could not
1044safely invoke the regex engine either directly with
e95d7314 1045C<m//> or C<s///>), or indirectly with functions such as
8525cfae 1046C<split>. Invoking the regex engine in these blocks would make perl
e95d7314 1047unstable.
8988a1bb 1048
14455d6c 1049=item C<(??{ code })>
d74e8afc
ITB
1050X<(??{})>
1051X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 1052
19799a22 1053B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
1054experimental, and may be changed without notice. Code executed that
1055has side effects may not perform identically from version to version
1056due to the effect of future optimisations in the regex engine.
0f5d15d6 1057
19799a22
GS
1058This is a "postponed" regular subexpression. The C<code> is evaluated
1059at run time, at the moment this subexpression may match. The result
1060of evaluation is considered as a regular expression and matched as
61528107 1061if it were inserted instead of this construct. Note that this means
c27a5cfe 1062that the contents of capture groups defined inside an eval'ed pattern
6bda09f9 1063are not available outside of the pattern, and vice versa, there is no
c27a5cfe 1064way for the inner pattern to refer to a capture group defined outside.
6bda09f9
YO
1065Thus,
1066
1067 ('a' x 100)=~/(??{'(.)' x 100})/
1068
81714fb9 1069B<will> match, it will B<not> set $1.
0f5d15d6 1070
428594d9 1071The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
1072where the C<code> ends are currently somewhat convoluted.
1073
1074The following pattern matches a parenthesized group:
0f5d15d6
IZ
1075
1076 $re = qr{
f793d64a
KW
1077 \(
1078 (?:
1079 (?> [^()]+ ) # Non-parens without backtracking
1080 |
1081 (??{ $re }) # Group with matching parens
1082 )*
1083 \)
1084 }x;
0f5d15d6 1085
6bda09f9
YO
1086See also C<(?PARNO)> for a different, more efficient way to accomplish
1087the same task.
1088
0b370c0a
A
1089For reasons of security, this construct is forbidden if the regular
1090expression involves run-time interpolation of variables, unless the
1091perilous C<use re 'eval'> pragma has been used (see L<re>), or the
1092variables contain results of C<qr//> operator (see
9bb1f947 1093L<perlop/"qrE<sol>STRINGE<sol>msixpo">).
0b370c0a 1094
8525cfae
FC
1095In perl 5.12.x and earlier, because the regex engine was not re-entrant,
1096delayed code could not safely invoke the regex engine either directly with
1097C<m//> or C<s///>), or indirectly with functions such as C<split>.
8988a1bb 1098
5d458dd8
YO
1099Recursing deeper than 50 times without consuming any input string will
1100result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
1101changing it requires a custom build.
1102
542fa716
YO
1103=item C<(?PARNO)> C<(?-PARNO)> C<(?+PARNO)> C<(?R)> C<(?0)>
1104X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)> X<(?-1)> X<(?+1)> X<(?-PARNO)> X<(?+PARNO)>
6bda09f9 1105X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
542fa716 1106X<regex, relative recursion>
6bda09f9 1107
81714fb9 1108Similar to C<(??{ code })> except it does not involve compiling any code,
c27a5cfe
KW
1109instead it treats the contents of a capture group as an independent
1110pattern that must match at the current position. Capture groups
81714fb9 1111contained by the pattern will have the value as determined by the
6bda09f9
YO
1112outermost recursion.
1113
894be9b7 1114PARNO is a sequence of digits (not starting with 0) whose value reflects
c27a5cfe 1115the paren-number of the capture group to recurse to. C<(?R)> recurses to
894be9b7 1116the beginning of the whole pattern. C<(?0)> is an alternate syntax for
542fa716 1117C<(?R)>. If PARNO is preceded by a plus or minus sign then it is assumed
c27a5cfe 1118to be relative, with negative numbers indicating preceding capture groups
542fa716 1119and positive ones following. Thus C<(?-1)> refers to the most recently
c27a5cfe 1120declared group, and C<(?+1)> indicates the next group to be declared.
c74340f9 1121Note that the counting for relative recursion differs from that of
c27a5cfe 1122relative backreferences, in that with recursion unclosed groups B<are>
c74340f9 1123included.
6bda09f9 1124
81714fb9 1125The following pattern matches a function foo() which may contain
f145b7e9 1126balanced parentheses as the argument.
6bda09f9
YO
1127
1128 $re = qr{ ( # paren group 1 (full function)
81714fb9 1129 foo
6bda09f9
YO
1130 ( # paren group 2 (parens)
1131 \(
1132 ( # paren group 3 (contents of parens)
1133 (?:
1134 (?> [^()]+ ) # Non-parens without backtracking
1135 |
1136 (?2) # Recurse to start of paren group 2
1137 )*
1138 )
1139 \)
1140 )
1141 )
1142 }x;
1143
1144If the pattern was used as follows
1145
1146 'foo(bar(baz)+baz(bop))'=~/$re/
1147 and print "\$1 = $1\n",
1148 "\$2 = $2\n",
1149 "\$3 = $3\n";
1150
1151the output produced should be the following:
1152
1153 $1 = foo(bar(baz)+baz(bop))
1154 $2 = (bar(baz)+baz(bop))
81714fb9 1155 $3 = bar(baz)+baz(bop)
6bda09f9 1156
c27a5cfe 1157If there is no corresponding capture group defined, then it is a
61528107 1158fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 1159string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
1160into perl, so changing it requires a custom build.
1161
542fa716
YO
1162The following shows how using negative indexing can make it
1163easier to embed recursive patterns inside of a C<qr//> construct
1164for later use:
1165
1166 my $parens = qr/(\((?:[^()]++|(?-1))*+\))/;
1167 if (/foo $parens \s+ + \s+ bar $parens/x) {
1168 # do something here...
1169 }
1170
81714fb9 1171B<Note> that this pattern does not behave the same way as the equivalent
0d017f4d 1172PCRE or Python construct of the same form. In Perl you can backtrack into
6bda09f9 1173a recursed group, in PCRE and Python the recursed into group is treated
542fa716
YO
1174as atomic. Also, modifiers are resolved at compile time, so constructs
1175like (?i:(?1)) or (?:(?i)(?1)) do not affect how the sub-pattern will
1176be processed.
6bda09f9 1177
894be9b7
YO
1178=item C<(?&NAME)>
1179X<(?&NAME)>
1180
0d017f4d
WL
1181Recurse to a named subpattern. Identical to C<(?PARNO)> except that the
1182parenthesis to recurse to is determined by name. If multiple parentheses have
894be9b7
YO
1183the same name, then it recurses to the leftmost.
1184
1185It is an error to refer to a name that is not declared somewhere in the
1186pattern.
1187
1f1031fe
YO
1188B<NOTE:> In order to make things easier for programmers with experience
1189with the Python or PCRE regex engines the pattern C<< (?P>NAME) >>
64c5a566 1190may be used instead of C<< (?&NAME) >>.
1f1031fe 1191
e2e6a0f1
YO
1192=item C<(?(condition)yes-pattern|no-pattern)>
1193X<(?()>
286f584a 1194
e2e6a0f1 1195=item C<(?(condition)yes-pattern)>
286f584a 1196
e2e6a0f1
YO
1197Conditional expression. C<(condition)> should be either an integer in
1198parentheses (which is valid if the corresponding pair of parentheses
1199matched), a look-ahead/look-behind/evaluate zero-width assertion, a
c27a5cfe 1200name in angle brackets or single quotes (which is valid if a group
e2e6a0f1
YO
1201with the given name matched), or the special symbol (R) (true when
1202evaluated inside of recursion or eval). Additionally the R may be
1203followed by a number, (which will be true when evaluated when recursing
1204inside of the appropriate group), or by C<&NAME>, in which case it will
1205be true only when evaluated during recursion in the named group.
1206
1207Here's a summary of the possible predicates:
1208
1209=over 4
1210
1211=item (1) (2) ...
1212
c27a5cfe 1213Checks if the numbered capturing group has matched something.
e2e6a0f1
YO
1214
1215=item (<NAME>) ('NAME')
1216
c27a5cfe 1217Checks if a group with the given name has matched something.
e2e6a0f1 1218
f01cd190
FC
1219=item (?=...) (?!...) (?<=...) (?<!...)
1220
1221Checks whether the pattern matches (or does not match, for the '!'
1222variants).
1223
e2e6a0f1
YO
1224=item (?{ CODE })
1225
f01cd190 1226Treats the return value of the code block as the condition.
e2e6a0f1
YO
1227
1228=item (R)
1229
1230Checks if the expression has been evaluated inside of recursion.
1231
1232=item (R1) (R2) ...
1233
1234Checks if the expression has been evaluated while executing directly
1235inside of the n-th capture group. This check is the regex equivalent of
1236
1237 if ((caller(0))[3] eq 'subname') { ... }
1238
1239In other words, it does not check the full recursion stack.
1240
1241=item (R&NAME)
1242
1243Similar to C<(R1)>, this predicate checks to see if we're executing
1244directly inside of the leftmost group with a given name (this is the same
1245logic used by C<(?&NAME)> to disambiguate). It does not check the full
1246stack, but only the name of the innermost active recursion.
1247
1248=item (DEFINE)
1249
1250In this case, the yes-pattern is never directly executed, and no
1251no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1252See below for details.
1253
1254=back
1255
1256For example:
1257
1258 m{ ( \( )?
1259 [^()]+
1260 (?(1) \) )
1261 }x
1262
1263matches a chunk of non-parentheses, possibly included in parentheses
1264themselves.
1265
1266A special form is the C<(DEFINE)> predicate, which never executes directly
1267its yes-pattern, and does not allow a no-pattern. This allows to define
1268subpatterns which will be executed only by using the recursion mechanism.
1269This way, you can define a set of regular expression rules that can be
1270bundled into any pattern you choose.
1271
1272It is recommended that for this usage you put the DEFINE block at the
1273end of the pattern, and that you name any subpatterns defined within it.
1274
1275Also, it's worth noting that patterns defined this way probably will
1276not be as efficient, as the optimiser is not very clever about
1277handling them.
1278
1279An example of how this might be used is as follows:
1280
2bf803e2 1281 /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
e2e6a0f1 1282 (?(DEFINE)
2bf803e2
YO
1283 (?<NAME_PAT>....)
1284 (?<ADRESS_PAT>....)
e2e6a0f1
YO
1285 )/x
1286
c27a5cfe
KW
1287Note that capture groups matched inside of recursion are not accessible
1288after the recursion returns, so the extra layer of capturing groups is
e2e6a0f1
YO
1289necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1290C<$+{NAME}> would be.
286f584a 1291
c47ff5f1 1292=item C<< (?>pattern) >>
6bda09f9 1293X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 1294
19799a22
GS
1295An "independent" subexpression, one which matches the substring
1296that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 1297position, and it matches I<nothing other than this substring>. This
19799a22
GS
1298construct is useful for optimizations of what would otherwise be
1299"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
1300It may also be useful in places where the "grab all you can, and do not
1301give anything back" semantic is desirable.
19799a22 1302
c47ff5f1 1303For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
1304(anchored at the beginning of string, as above) will match I<all>
1305characters C<a> at the beginning of string, leaving no C<a> for
1306C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
1307since the match of the subgroup C<a*> is influenced by the following
1308group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
1309C<a*ab> will match fewer characters than a standalone C<a*>, since
1310this makes the tail match.
1311
c47ff5f1 1312An effect similar to C<< (?>pattern) >> may be achieved by writing
d8b950dc
KW
1313C<(?=(pattern))\g1>. This matches the same substring as a standalone
1314C<a+>, and the following C<\g1> eats the matched string; it therefore
c47ff5f1 1315makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
1316(The difference between these two constructs is that the second one
1317uses a capturing group, thus shifting ordinals of backreferences
1318in the rest of a regular expression.)
1319
1320Consider this pattern:
c277df42 1321
871b0233 1322 m{ \(
e2e6a0f1 1323 (
f793d64a 1324 [^()]+ # x+
e2e6a0f1 1325 |
871b0233
IZ
1326 \( [^()]* \)
1327 )+
e2e6a0f1 1328 \)
871b0233 1329 }x
5a964f20 1330
19799a22
GS
1331That will efficiently match a nonempty group with matching parentheses
1332two levels deep or less. However, if there is no such group, it
1333will take virtually forever on a long string. That's because there
1334are so many different ways to split a long string into several
1335substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
1336to a subpattern of the above pattern. Consider how the pattern
1337above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
1338seconds, but that each extra letter doubles this time. This
1339exponential performance will make it appear that your program has
14218588 1340hung. However, a tiny change to this pattern
5a964f20 1341
e2e6a0f1
YO
1342 m{ \(
1343 (
f793d64a 1344 (?> [^()]+ ) # change x+ above to (?> x+ )
e2e6a0f1 1345 |
871b0233
IZ
1346 \( [^()]* \)
1347 )+
e2e6a0f1 1348 \)
871b0233 1349 }x
c277df42 1350
c47ff5f1 1351which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1352this yourself would be a productive exercise), but finishes in a fourth
1353the time when used on a similar string with 1000000 C<a>s. Be aware,
1354however, that this pattern currently triggers a warning message under
9f1b1f2d 1355the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1356C<"matches null string many times in regex">.
c277df42 1357
c47ff5f1 1358On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1359effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1360This was only 4 times slower on a string with 1000000 C<a>s.
1361
9da458fc
IZ
1362The "grab all you can, and do not give anything back" semantic is desirable
1363in many situations where on the first sight a simple C<()*> looks like
1364the correct solution. Suppose we parse text with comments being delimited
1365by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1366its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1367the comment delimiter, because it may "give up" some whitespace if
1368the remainder of the pattern can be made to match that way. The correct
1369answer is either one of these:
1370
1371 (?>#[ \t]*)
1372 #[ \t]*(?![ \t])
1373
1374For example, to grab non-empty comments into $1, one should use either
1375one of these:
1376
1377 / (?> \# [ \t]* ) ( .+ ) /x;
1378 / \# [ \t]* ( [^ \t] .* ) /x;
1379
1380Which one you pick depends on which of these expressions better reflects
1381the above specification of comments.
1382
6bda09f9
YO
1383In some literature this construct is called "atomic matching" or
1384"possessive matching".
1385
b9b4dddf
YO
1386Possessive quantifiers are equivalent to putting the item they are applied
1387to inside of one of these constructs. The following equivalences apply:
1388
1389 Quantifier Form Bracketing Form
1390 --------------- ---------------
1391 PAT*+ (?>PAT*)
1392 PAT++ (?>PAT+)
1393 PAT?+ (?>PAT?)
1394 PAT{min,max}+ (?>PAT{min,max})
1395
e2e6a0f1
YO
1396=back
1397
1398=head2 Special Backtracking Control Verbs
1399
1400B<WARNING:> These patterns are experimental and subject to change or
0d017f4d 1401removal in a future version of Perl. Their usage in production code should
e2e6a0f1
YO
1402be noted to avoid problems during upgrades.
1403
1404These special patterns are generally of the form C<(*VERB:ARG)>. Unless
1405otherwise stated the ARG argument is optional; in some cases, it is
1406forbidden.
1407
1408Any pattern containing a special backtracking verb that allows an argument
e1020413 1409has the special behaviour that when executed it sets the current package's
5d458dd8
YO
1410C<$REGERROR> and C<$REGMARK> variables. When doing so the following
1411rules apply:
e2e6a0f1 1412
5d458dd8
YO
1413On failure, the C<$REGERROR> variable will be set to the ARG value of the
1414verb pattern, if the verb was involved in the failure of the match. If the
1415ARG part of the pattern was omitted, then C<$REGERROR> will be set to the
1416name of the last C<(*MARK:NAME)> pattern executed, or to TRUE if there was
1417none. Also, the C<$REGMARK> variable will be set to FALSE.
e2e6a0f1 1418
5d458dd8
YO
1419On a successful match, the C<$REGERROR> variable will be set to FALSE, and
1420the C<$REGMARK> variable will be set to the name of the last
1421C<(*MARK:NAME)> pattern executed. See the explanation for the
1422C<(*MARK:NAME)> verb below for more details.
e2e6a0f1 1423
5d458dd8
YO
1424B<NOTE:> C<$REGERROR> and C<$REGMARK> are not magic variables like C<$1>
1425and most other regex related variables. They are not local to a scope, nor
1426readonly, but instead are volatile package variables similar to C<$AUTOLOAD>.
1427Use C<local> to localize changes to them to a specific scope if necessary.
e2e6a0f1
YO
1428
1429If a pattern does not contain a special backtracking verb that allows an
5d458dd8 1430argument, then C<$REGERROR> and C<$REGMARK> are not touched at all.
e2e6a0f1
YO
1431
1432=over 4
1433
1434=item Verbs that take an argument
1435
1436=over 4
1437
5d458dd8 1438=item C<(*PRUNE)> C<(*PRUNE:NAME)>
f7819f85 1439X<(*PRUNE)> X<(*PRUNE:NAME)>
54612592 1440
5d458dd8
YO
1441This zero-width pattern prunes the backtracking tree at the current point
1442when backtracked into on failure. Consider the pattern C<A (*PRUNE) B>,
1443where A and B are complex patterns. Until the C<(*PRUNE)> verb is reached,
1444A may backtrack as necessary to match. Once it is reached, matching
1445continues in B, which may also backtrack as necessary; however, should B
1446not match, then no further backtracking will take place, and the pattern
1447will fail outright at the current starting position.
54612592
YO
1448
1449The following example counts all the possible matching strings in a
1450pattern (without actually matching any of them).
1451
e2e6a0f1 1452 'aaab' =~ /a+b?(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1453 print "Count=$count\n";
1454
1455which produces:
1456
1457 aaab
1458 aaa
1459 aa
1460 a
1461 aab
1462 aa
1463 a
1464 ab
1465 a
1466 Count=9
1467
5d458dd8 1468If we add a C<(*PRUNE)> before the count like the following
54612592 1469
5d458dd8 1470 'aaab' =~ /a+b?(*PRUNE)(?{print "$&\n"; $count++})(*FAIL)/;
54612592
YO
1471 print "Count=$count\n";
1472
1473we prevent backtracking and find the count of the longest matching
353c6505 1474at each matching starting point like so:
54612592
YO
1475
1476 aaab
1477 aab
1478 ab
1479 Count=3
1480
5d458dd8 1481Any number of C<(*PRUNE)> assertions may be used in a pattern.
54612592 1482
5d458dd8
YO
1483See also C<< (?>pattern) >> and possessive quantifiers for other ways to
1484control backtracking. In some cases, the use of C<(*PRUNE)> can be
1485replaced with a C<< (?>pattern) >> with no functional difference; however,
1486C<(*PRUNE)> can be used to handle cases that cannot be expressed using a
1487C<< (?>pattern) >> alone.
54612592 1488
e2e6a0f1 1489
5d458dd8
YO
1490=item C<(*SKIP)> C<(*SKIP:NAME)>
1491X<(*SKIP)>
e2e6a0f1 1492
5d458dd8 1493This zero-width pattern is similar to C<(*PRUNE)>, except that on
e2e6a0f1 1494failure it also signifies that whatever text that was matched leading up
5d458dd8
YO
1495to the C<(*SKIP)> pattern being executed cannot be part of I<any> match
1496of this pattern. This effectively means that the regex engine "skips" forward
1497to this position on failure and tries to match again, (assuming that
1498there is sufficient room to match).
1499
1500The name of the C<(*SKIP:NAME)> pattern has special significance. If a
1501C<(*MARK:NAME)> was encountered while matching, then it is that position
1502which is used as the "skip point". If no C<(*MARK)> of that name was
1503encountered, then the C<(*SKIP)> operator has no effect. When used
1504without a name the "skip point" is where the match point was when
1505executing the (*SKIP) pattern.
1506
1507Compare the following to the examples in C<(*PRUNE)>, note the string
24b23f37
YO
1508is twice as long:
1509
5d458dd8 1510 'aaabaaab' =~ /a+b?(*SKIP)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1511 print "Count=$count\n";
1512
1513outputs
1514
1515 aaab
1516 aaab
1517 Count=2
1518
5d458dd8 1519Once the 'aaab' at the start of the string has matched, and the C<(*SKIP)>
353c6505 1520executed, the next starting point will be where the cursor was when the
5d458dd8
YO
1521C<(*SKIP)> was executed.
1522
5d458dd8
YO
1523=item C<(*MARK:NAME)> C<(*:NAME)>
1524X<(*MARK)> C<(*MARK:NAME)> C<(*:NAME)>
1525
1526This zero-width pattern can be used to mark the point reached in a string
1527when a certain part of the pattern has been successfully matched. This
1528mark may be given a name. A later C<(*SKIP)> pattern will then skip
1529forward to that point if backtracked into on failure. Any number of
b4222fa9 1530C<(*MARK)> patterns are allowed, and the NAME portion may be duplicated.
5d458dd8
YO
1531
1532In addition to interacting with the C<(*SKIP)> pattern, C<(*MARK:NAME)>
1533can be used to "label" a pattern branch, so that after matching, the
1534program can determine which branches of the pattern were involved in the
1535match.
1536
1537When a match is successful, the C<$REGMARK> variable will be set to the
1538name of the most recently executed C<(*MARK:NAME)> that was involved
1539in the match.
1540
1541This can be used to determine which branch of a pattern was matched
c27a5cfe 1542without using a separate capture group for each branch, which in turn
5d458dd8
YO
1543can result in a performance improvement, as perl cannot optimize
1544C</(?:(x)|(y)|(z))/> as efficiently as something like
1545C</(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/>.
1546
1547When a match has failed, and unless another verb has been involved in
1548failing the match and has provided its own name to use, the C<$REGERROR>
1549variable will be set to the name of the most recently executed
1550C<(*MARK:NAME)>.
1551
1552See C<(*SKIP)> for more details.
1553
b62d2d15
YO
1554As a shortcut C<(*MARK:NAME)> can be written C<(*:NAME)>.
1555
5d458dd8
YO
1556=item C<(*THEN)> C<(*THEN:NAME)>
1557
241e7389 1558This is similar to the "cut group" operator C<::> from Perl 6. Like
5d458dd8
YO
1559C<(*PRUNE)>, this verb always matches, and when backtracked into on
1560failure, it causes the regex engine to try the next alternation in the
1561innermost enclosing group (capturing or otherwise).
1562
1563Its name comes from the observation that this operation combined with the
1564alternation operator (C<|>) can be used to create what is essentially a
1565pattern-based if/then/else block:
1566
1567 ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
1568
1569Note that if this operator is used and NOT inside of an alternation then
1570it acts exactly like the C<(*PRUNE)> operator.
1571
1572 / A (*PRUNE) B /
1573
1574is the same as
1575
1576 / A (*THEN) B /
1577
1578but
1579
1580 / ( A (*THEN) B | C (*THEN) D ) /
1581
1582is not the same as
1583
1584 / ( A (*PRUNE) B | C (*PRUNE) D ) /
1585
1586as after matching the A but failing on the B the C<(*THEN)> verb will
1587backtrack and try C; but the C<(*PRUNE)> verb will simply fail.
24b23f37 1588
e2e6a0f1
YO
1589=item C<(*COMMIT)>
1590X<(*COMMIT)>
24b23f37 1591
241e7389 1592This is the Perl 6 "commit pattern" C<< <commit> >> or C<:::>. It's a
5d458dd8
YO
1593zero-width pattern similar to C<(*SKIP)>, except that when backtracked
1594into on failure it causes the match to fail outright. No further attempts
1595to find a valid match by advancing the start pointer will occur again.
1596For example,
24b23f37 1597
e2e6a0f1 1598 'aaabaaab' =~ /a+b?(*COMMIT)(?{print "$&\n"; $count++})(*FAIL)/;
24b23f37
YO
1599 print "Count=$count\n";
1600
1601outputs
1602
1603 aaab
1604 Count=1
1605
e2e6a0f1
YO
1606In other words, once the C<(*COMMIT)> has been entered, and if the pattern
1607does not match, the regex engine will not try any further matching on the
1608rest of the string.
c277df42 1609
e2e6a0f1 1610=back
9af228c6 1611
e2e6a0f1 1612=item Verbs without an argument
9af228c6
YO
1613
1614=over 4
1615
e2e6a0f1
YO
1616=item C<(*FAIL)> C<(*F)>
1617X<(*FAIL)> X<(*F)>
9af228c6 1618
e2e6a0f1
YO
1619This pattern matches nothing and always fails. It can be used to force the
1620engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
1621fact, C<(?!)> gets optimised into C<(*FAIL)> internally.
9af228c6 1622
e2e6a0f1 1623It is probably useful only when combined with C<(?{})> or C<(??{})>.
9af228c6 1624
e2e6a0f1
YO
1625=item C<(*ACCEPT)>
1626X<(*ACCEPT)>
9af228c6 1627
e2e6a0f1
YO
1628B<WARNING:> This feature is highly experimental. It is not recommended
1629for production code.
9af228c6 1630
e2e6a0f1
YO
1631This pattern matches nothing and causes the end of successful matching at
1632the point at which the C<(*ACCEPT)> pattern was encountered, regardless of
1633whether there is actually more to match in the string. When inside of a
0d017f4d 1634nested pattern, such as recursion, or in a subpattern dynamically generated
e2e6a0f1 1635via C<(??{})>, only the innermost pattern is ended immediately.
9af228c6 1636
c27a5cfe 1637If the C<(*ACCEPT)> is inside of capturing groups then the groups are
e2e6a0f1
YO
1638marked as ended at the point at which the C<(*ACCEPT)> was encountered.
1639For instance:
9af228c6 1640
e2e6a0f1 1641 'AB' =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
9af228c6 1642
e2e6a0f1 1643will match, and C<$1> will be C<AB> and C<$2> will be C<B>, C<$3> will not
0d017f4d 1644be set. If another branch in the inner parentheses were matched, such as in the
e2e6a0f1 1645string 'ACDE', then the C<D> and C<E> would have to be matched as well.
9af228c6
YO
1646
1647=back
c277df42 1648
a0d0e21e
LW
1649=back
1650
c07a80fd 1651=head2 Backtracking
d74e8afc 1652X<backtrack> X<backtracking>
c07a80fd 1653
35a734be
IZ
1654NOTE: This section presents an abstract approximation of regular
1655expression behavior. For a more rigorous (and complicated) view of
1656the rules involved in selecting a match among possible alternatives,
0d017f4d 1657see L<Combining RE Pieces>.
35a734be 1658
c277df42 1659A fundamental feature of regular expression matching involves the
5a964f20 1660notion called I<backtracking>, which is currently used (when needed)
0d017f4d 1661by all regular non-possessive expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1662C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1663internally, but the general principle outlined here is valid.
c07a80fd
PP
1664
1665For a regular expression to match, the I<entire> regular expression must
1666match, not just part of it. So if the beginning of a pattern containing a
1667quantifier succeeds in a way that causes later parts in the pattern to
1668fail, the matching engine backs up and recalculates the beginning
1669part--that's why it's called backtracking.
1670
1671Here is an example of backtracking: Let's say you want to find the
1672word following "foo" in the string "Food is on the foo table.":
1673
1674 $_ = "Food is on the foo table.";
1675 if ( /\b(foo)\s+(\w+)/i ) {
f793d64a 1676 print "$2 follows $1.\n";
c07a80fd
PP
1677 }
1678
1679When the match runs, the first part of the regular expression (C<\b(foo)>)
1680finds a possible match right at the beginning of the string, and loads up
1681$1 with "Foo". However, as soon as the matching engine sees that there's
1682no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1683mistake and starts over again one character after where it had the
c07a80fd
PP
1684tentative match. This time it goes all the way until the next occurrence
1685of "foo". The complete regular expression matches this time, and you get
1686the expected output of "table follows foo."
1687
1688Sometimes minimal matching can help a lot. Imagine you'd like to match
1689everything between "foo" and "bar". Initially, you write something
1690like this:
1691
1692 $_ = "The food is under the bar in the barn.";
1693 if ( /foo(.*)bar/ ) {
f793d64a 1694 print "got <$1>\n";
c07a80fd
PP
1695 }
1696
1697Which perhaps unexpectedly yields:
1698
1699 got <d is under the bar in the >
1700
1701That's because C<.*> was greedy, so you get everything between the
14218588 1702I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1703to use minimal matching to make sure you get the text between a "foo"
1704and the first "bar" thereafter.
1705
1706 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1707 got <d is under the >
1708
0d017f4d 1709Here's another example. Let's say you'd like to match a number at the end
b6e13d97 1710of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
1711So you write this:
1712
1713 $_ = "I have 2 numbers: 53147";
f793d64a
KW
1714 if ( /(.*)(\d*)/ ) { # Wrong!
1715 print "Beginning is <$1>, number is <$2>.\n";
c07a80fd
PP
1716 }
1717
1718That won't work at all, because C<.*> was greedy and gobbled up the
1719whole string. As C<\d*> can match on an empty string the complete
1720regular expression matched successfully.
1721
8e1088bc 1722 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
1723
1724Here are some variants, most of which don't work:
1725
1726 $_ = "I have 2 numbers: 53147";
1727 @pats = qw{
f793d64a
KW
1728 (.*)(\d*)
1729 (.*)(\d+)
1730 (.*?)(\d*)
1731 (.*?)(\d+)
1732 (.*)(\d+)$
1733 (.*?)(\d+)$
1734 (.*)\b(\d+)$
1735 (.*\D)(\d+)$
c07a80fd
PP
1736 };
1737
1738 for $pat (@pats) {
f793d64a
KW
1739 printf "%-12s ", $pat;
1740 if ( /$pat/ ) {
1741 print "<$1> <$2>\n";
1742 } else {
1743 print "FAIL\n";
1744 }
c07a80fd
PP
1745 }
1746
1747That will print out:
1748
1749 (.*)(\d*) <I have 2 numbers: 53147> <>
1750 (.*)(\d+) <I have 2 numbers: 5314> <7>
1751 (.*?)(\d*) <> <>
1752 (.*?)(\d+) <I have > <2>
1753 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1754 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1755 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1756 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1757
1758As you see, this can be a bit tricky. It's important to realize that a
1759regular expression is merely a set of assertions that gives a definition
1760of success. There may be 0, 1, or several different ways that the
1761definition might succeed against a particular string. And if there are
5a964f20
TC
1762multiple ways it might succeed, you need to understand backtracking to
1763know which variety of success you will achieve.
c07a80fd 1764
19799a22 1765When using look-ahead assertions and negations, this can all get even
8b19b778 1766trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
1767followed by "123". You might try to write that as
1768
871b0233 1769 $_ = "ABC123";
f793d64a
KW
1770 if ( /^\D*(?!123)/ ) { # Wrong!
1771 print "Yup, no 123 in $_\n";
871b0233 1772 }
c07a80fd
PP
1773
1774But that isn't going to match; at least, not the way you're hoping. It
1775claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1776why that pattern matches, contrary to popular expectations:
c07a80fd 1777
4358a253
SS
1778 $x = 'ABC123';
1779 $y = 'ABC445';
c07a80fd 1780
4358a253
SS
1781 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1782 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1783
4358a253
SS
1784 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1785 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
1786
1787This prints
1788
1789 2: got ABC
1790 3: got AB
1791 4: got ABC
1792
5f05dabc 1793You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
1794general purpose version of test 1. The important difference between
1795them is that test 3 contains a quantifier (C<\D*>) and so can use
1796backtracking, whereas test 1 will not. What's happening is
1797that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1798non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1799let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1800fail.
14218588 1801
c07a80fd 1802The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1803try to match C<(?!123> with "123", which fails. But because
c07a80fd
PP
1804a quantifier (C<\D*>) has been used in the regular expression, the
1805search engine can backtrack and retry the match differently
54310121 1806in the hope of matching the complete regular expression.
c07a80fd 1807
5a964f20
TC
1808The pattern really, I<really> wants to succeed, so it uses the
1809standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1810time. Now there's indeed something following "AB" that is not
14218588 1811"123". It's "C123", which suffices.
c07a80fd 1812
14218588
GS
1813We can deal with this by using both an assertion and a negation.
1814We'll say that the first part in $1 must be followed both by a digit
1815and by something that's not "123". Remember that the look-aheads
1816are zero-width expressions--they only look, but don't consume any
1817of the string in their match. So rewriting this way produces what
c07a80fd
PP
1818you'd expect; that is, case 5 will fail, but case 6 succeeds:
1819
4358a253
SS
1820 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1821 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
1822
1823 6: got ABC
1824
5a964f20 1825In other words, the two zero-width assertions next to each other work as though
19799a22 1826they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
1827matches only if you're at the beginning of the line AND the end of the
1828line simultaneously. The deeper underlying truth is that juxtaposition in
1829regular expressions always means AND, except when you write an explicit OR
1830using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1831although the attempted matches are made at different positions because "a"
1832is not a zero-width assertion, but a one-width assertion.
1833
0d017f4d 1834B<WARNING>: Particularly complicated regular expressions can take
14218588 1835exponential time to solve because of the immense number of possible
0d017f4d 1836ways they can use backtracking to try for a match. For example, without
9da458fc
IZ
1837internal optimizations done by the regular expression engine, this will
1838take a painfully long time to run:
c07a80fd 1839
e1901655
IZ
1840 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1841
1842And if you used C<*>'s in the internal groups instead of limiting them
1843to 0 through 5 matches, then it would take forever--or until you ran
1844out of stack space. Moreover, these internal optimizations are not
1845always applicable. For example, if you put C<{0,5}> instead of C<*>
1846on the external group, no current optimization is applicable, and the
1847match takes a long time to finish.
c07a80fd 1848
9da458fc
IZ
1849A powerful tool for optimizing such beasts is what is known as an
1850"independent group",
96090e4f 1851which does not backtrack (see L</C<< (?>pattern) >>>). Note also that
9da458fc 1852zero-length look-ahead/look-behind assertions will not backtrack to make
5d458dd8 1853the tail match, since they are in "logical" context: only
14218588 1854whether they match is considered relevant. For an example
9da458fc 1855where side-effects of look-ahead I<might> have influenced the
96090e4f 1856following match, see L</C<< (?>pattern) >>>.
c277df42 1857
a0d0e21e 1858=head2 Version 8 Regular Expressions
d74e8afc 1859X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1860
5a964f20 1861In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1862routines, here are the pattern-matching rules not described above.
1863
54310121 1864Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1865with a special meaning described here or above. You can cause
5a964f20 1866characters that normally function as metacharacters to be interpreted
5f05dabc 1867literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
0d017f4d
WL
1868character; "\\" matches a "\"). This escape mechanism is also required
1869for the character used as the pattern delimiter.
1870
1871A series of characters matches that series of characters in the target
1872string, so the pattern C<blurfl> would match "blurfl" in the target
1873string.
a0d0e21e
LW
1874
1875You can specify a character class, by enclosing a list of characters
5d458dd8 1876in C<[]>, which will match any character from the list. If the
a0d0e21e 1877first character after the "[" is "^", the class matches any character not
14218588 1878in the list. Within a list, the "-" character specifies a
5a964f20 1879range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1880inclusive. If you want either "-" or "]" itself to be a member of a
1881class, put it at the start of the list (possibly after a "^"), or
1882escape it with a backslash. "-" is also taken literally when it is
1883at the end of the list, just before the closing "]". (The
84850974
DD
1884following all specify the same class of three characters: C<[-az]>,
1885C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
5d458dd8
YO
1886specifies a class containing twenty-six characters, even on EBCDIC-based
1887character sets.) Also, if you try to use the character
1888classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1889a range, the "-" is understood literally.
a0d0e21e 1890
8ada0baa
JH
1891Note also that the whole range idea is rather unportable between
1892character sets--and even within character sets they may cause results
1893you probably didn't expect. A sound principle is to use only ranges
0d017f4d 1894that begin from and end at either alphabetics of equal case ([a-e],
8ada0baa
JH
1895[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1896spell out the character sets in full.
1897
54310121 1898Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1899used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1900"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
dc0d9c48 1901of three octal digits, matches the character whose coded character set value
5d458dd8 1902is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
dc0d9c48 1903matches the character whose ordinal is I<nn>. The expression \cI<x>
5d458dd8 1904matches the character control-I<x>. Finally, the "." metacharacter
fb55449c 1905matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1906
1907You can specify a series of alternatives for a pattern using "|" to
1908separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1909or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1910first alternative includes everything from the last pattern delimiter
1911("(", "[", or the beginning of the pattern) up to the first "|", and
1912the last alternative contains everything from the last "|" to the next
14218588
GS
1913pattern delimiter. That's why it's common practice to include
1914alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1915start and end.
1916
5a964f20 1917Alternatives are tried from left to right, so the first
a3cb178b
GS
1918alternative found for which the entire expression matches, is the one that
1919is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1920example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1921part will match, as that is the first alternative tried, and it successfully
1922matches the target string. (This might not seem important, but it is
1923important when you are capturing matched text using parentheses.)
1924
5a964f20 1925Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1926so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1927
14218588
GS
1928Within a pattern, you may designate subpatterns for later reference
1929by enclosing them in parentheses, and you may refer back to the
1930I<n>th subpattern later in the pattern using the metacharacter
1931\I<n>. Subpatterns are numbered based on the left to right order
1932of their opening parenthesis. A backreference matches whatever
1933actually matched the subpattern in the string being examined, not
d8b950dc 1934the rules for that subpattern. Therefore, C<(0|0x)\d*\s\g1\d*> will
14218588
GS
1935match "0x1234 0x4321", but not "0x1234 01234", because subpattern
19361 matched "0x", even though the rule C<0|0x> could potentially match
1937the leading 0 in the second number.
cb1a09d0 1938
0d017f4d 1939=head2 Warning on \1 Instead of $1
cb1a09d0 1940
5a964f20 1941Some people get too used to writing things like:
cb1a09d0
AD
1942
1943 $pattern =~ s/(\W)/\\\1/g;
1944
3ff1c45a
KW
1945This is grandfathered (for \1 to \9) for the RHS of a substitute to avoid
1946shocking the
cb1a09d0 1947B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1948PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1949the usual double-quoted string means a control-A. The customary Unix
1950meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1951of doing that, you get yourself into trouble if you then add an C</e>
1952modifier.
1953
f793d64a 1954 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1955
1956Or if you try to do
1957
1958 s/(\d+)/\1000/;
1959
1960You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1961C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1962with the operation of matching a backreference. Certainly they mean two
1963different things on the I<left> side of the C<s///>.
9fa51da4 1964
0d017f4d 1965=head2 Repeated Patterns Matching a Zero-length Substring
c84d73f1 1966
19799a22 1967B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1968
1969Regular expressions provide a terse and powerful programming language. As
1970with most other power tools, power comes together with the ability
1971to wreak havoc.
1972
1973A common abuse of this power stems from the ability to make infinite
628afcb5 1974loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1975
1976 'foo' =~ m{ ( o? )* }x;
1977
0d017f4d 1978The C<o?> matches at the beginning of C<'foo'>, and since the position
c84d73f1 1979in the string is not moved by the match, C<o?> would match again and again
527e91da 1980because of the C<*> quantifier. Another common way to create a similar cycle
c84d73f1
IZ
1981is with the looping modifier C<//g>:
1982
1983 @matches = ( 'foo' =~ m{ o? }xg );
1984
1985or
1986
1987 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1988
1989or the loop implied by split().
1990
1991However, long experience has shown that many programming tasks may
14218588
GS
1992be significantly simplified by using repeated subexpressions that
1993may match zero-length substrings. Here's a simple example being:
c84d73f1 1994
f793d64a 1995 @chars = split //, $string; # // is not magic in split
c84d73f1
IZ
1996 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1997
9da458fc 1998Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1 1999the infinite loop>. The rules for this are different for lower-level
527e91da 2000loops given by the greedy quantifiers C<*+{}>, and for higher-level
c84d73f1
IZ
2001ones like the C</g> modifier or split() operator.
2002
19799a22
GS
2003The lower-level loops are I<interrupted> (that is, the loop is
2004broken) when Perl detects that a repeated expression matched a
2005zero-length substring. Thus
c84d73f1
IZ
2006
2007 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
2008
5d458dd8 2009is made equivalent to
c84d73f1 2010
5d458dd8
YO
2011 m{ (?: NON_ZERO_LENGTH )*
2012 |
2013 (?: ZERO_LENGTH )?
c84d73f1
IZ
2014 }x;
2015
2016The higher level-loops preserve an additional state between iterations:
5d458dd8 2017whether the last match was zero-length. To break the loop, the following
c84d73f1 2018match after a zero-length match is prohibited to have a length of zero.
5d458dd8 2019This prohibition interacts with backtracking (see L<"Backtracking">),
c84d73f1
IZ
2020and so the I<second best> match is chosen if the I<best> match is of
2021zero length.
2022
19799a22 2023For example:
c84d73f1
IZ
2024
2025 $_ = 'bar';
2026 s/\w??/<$&>/g;
2027
20fb949f 2028results in C<< <><b><><a><><r><> >>. At each position of the string the best
5d458dd8 2029match given by non-greedy C<??> is the zero-length match, and the I<second
c84d73f1
IZ
2030best> match is what is matched by C<\w>. Thus zero-length matches
2031alternate with one-character-long matches.
2032
5d458dd8 2033Similarly, for repeated C<m/()/g> the second-best match is the match at the
c84d73f1
IZ
2034position one notch further in the string.
2035
19799a22 2036The additional state of being I<matched with zero-length> is associated with
c84d73f1 2037the matched string, and is reset by each assignment to pos().
9da458fc
IZ
2038Zero-length matches at the end of the previous match are ignored
2039during C<split>.
c84d73f1 2040
0d017f4d 2041=head2 Combining RE Pieces
35a734be
IZ
2042
2043Each of the elementary pieces of regular expressions which were described
2044before (such as C<ab> or C<\Z>) could match at most one substring
2045at the given position of the input string. However, in a typical regular
2046expression these elementary pieces are combined into more complicated
2047patterns using combining operators C<ST>, C<S|T>, C<S*> etc
2048(in these examples C<S> and C<T> are regular subexpressions).
2049
2050Such combinations can include alternatives, leading to a problem of choice:
2051if we match a regular expression C<a|ab> against C<"abc">, will it match
2052substring C<"a"> or C<"ab">? One way to describe which substring is
2053actually matched is the concept of backtracking (see L<"Backtracking">).
2054However, this description is too low-level and makes you think
2055in terms of a particular implementation.
2056
2057Another description starts with notions of "better"/"worse". All the
2058substrings which may be matched by the given regular expression can be
2059sorted from the "best" match to the "worst" match, and it is the "best"
2060match which is chosen. This substitutes the question of "what is chosen?"
2061by the question of "which matches are better, and which are worse?".
2062
2063Again, for elementary pieces there is no such question, since at most
2064one match at a given position is possible. This section describes the
2065notion of better/worse for combining operators. In the description
2066below C<S> and C<T> are regular subexpressions.
2067
13a2d996 2068=over 4
35a734be
IZ
2069
2070=item C<ST>
2071
2072Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
2073substrings which can be matched by C<S>, C<B> and C<B'> are substrings
5d458dd8 2074which can be matched by C<T>.
35a734be
IZ
2075
2076If C<A> is better match for C<S> than C<A'>, C<AB> is a better
2077match than C<A'B'>.
2078
2079If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
2080C<B> is better match for C<T> than C<B'>.
2081
2082=item C<S|T>
2083
2084When C<S> can match, it is a better match than when only C<T> can match.
2085
2086Ordering of two matches for C<S> is the same as for C<S>. Similar for
2087two matches for C<T>.
2088
2089=item C<S{REPEAT_COUNT}>
2090
2091Matches as C<SSS...S> (repeated as many times as necessary).
2092
2093=item C<S{min,max}>
2094
2095Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
2096
2097=item C<S{min,max}?>
2098
2099Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
2100
2101=item C<S?>, C<S*>, C<S+>
2102
2103Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
2104
2105=item C<S??>, C<S*?>, C<S+?>
2106
2107Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
2108
c47ff5f1 2109=item C<< (?>S) >>
35a734be
IZ
2110
2111Matches the best match for C<S> and only that.
2112
2113=item C<(?=S)>, C<(?<=S)>
2114
2115Only the best match for C<S> is considered. (This is important only if
2116C<S> has capturing parentheses, and backreferences are used somewhere
2117else in the whole regular expression.)
2118
2119=item C<(?!S)>, C<(?<!S)>
2120
2121For this grouping operator there is no need to describe the ordering, since
2122only whether or not C<S> can match is important.
2123
6bda09f9 2124=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
2125
2126The ordering is the same as for the regular expression which is
c27a5cfe 2127the result of EXPR, or the pattern contained by capture group PARNO.
35a734be
IZ
2128
2129=item C<(?(condition)yes-pattern|no-pattern)>
2130
2131Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
2132already determined. The ordering of the matches is the same as for the
2133chosen subexpression.
2134
2135=back
2136
2137The above recipes describe the ordering of matches I<at a given position>.
2138One more rule is needed to understand how a match is determined for the
2139whole regular expression: a match at an earlier position is always better
2140than a match at a later position.
2141
0d017f4d 2142=head2 Creating Custom RE Engines
c84d73f1
IZ
2143
2144Overloaded constants (see L<overload>) provide a simple way to extend
2145the functionality of the RE engine.
2146
2147Suppose that we want to enable a new RE escape-sequence C<\Y|> which
0d017f4d 2148matches at a boundary between whitespace characters and non-whitespace
c84d73f1
IZ
2149characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
2150at these positions, so we want to have each C<\Y|> in the place of the
2151more complicated version. We can create a module C<customre> to do
2152this:
2153
2154 package customre;
2155 use overload;
2156
2157 sub import {
2158 shift;
2159 die "No argument to customre::import allowed" if @_;
2160 overload::constant 'qr' => \&convert;
2161 }
2162
2163 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
2164
580a9fe1
RGS
2165 # We must also take care of not escaping the legitimate \\Y|
2166 # sequence, hence the presence of '\\' in the conversion rules.
5d458dd8 2167 my %rules = ( '\\' => '\\\\',
f793d64a 2168 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
c84d73f1
IZ
2169 sub convert {
2170 my $re = shift;
5d458dd8 2171 $re =~ s{
c84d73f1
IZ
2172 \\ ( \\ | Y . )
2173 }
5d458dd8 2174 { $rules{$1} or invalid($re,$1) }sgex;
c84d73f1
IZ
2175 return $re;
2176 }
2177
2178Now C<use customre> enables the new escape in constant regular
2179expressions, i.e., those without any runtime variable interpolations.
2180As documented in L<overload>, this conversion will work only over
2181literal parts of regular expressions. For C<\Y|$re\Y|> the variable
2182part of this regular expression needs to be converted explicitly
2183(but only if the special meaning of C<\Y|> should be enabled inside $re):
2184
2185 use customre;
2186 $re = <>;
2187 chomp $re;
2188 $re = customre::convert $re;
2189 /\Y|$re\Y|/;
2190
1f1031fe
YO
2191=head1 PCRE/Python Support
2192
99d59c4d 2193As of Perl 5.10.0, Perl supports several Python/PCRE specific extensions
1f1031fe 2194to the regex syntax. While Perl programmers are encouraged to use the
99d59c4d 2195Perl specific syntax, the following are also accepted:
1f1031fe
YO
2196
2197=over 4
2198
ae5648b3 2199=item C<< (?PE<lt>NAMEE<gt>pattern) >>
1f1031fe 2200
c27a5cfe 2201Define a named capture group. Equivalent to C<< (?<NAME>pattern) >>.
1f1031fe
YO
2202
2203=item C<< (?P=NAME) >>
2204
c27a5cfe 2205Backreference to a named capture group. Equivalent to C<< \g{NAME} >>.
1f1031fe
YO
2206
2207=item C<< (?P>NAME) >>
2208
c27a5cfe 2209Subroutine call to a named capture group. Equivalent to C<< (?&NAME) >>.
1f1031fe 2210
ee9b8eae 2211=back
1f1031fe 2212
19799a22
GS
2213=head1 BUGS
2214
78288b8e
KW
2215There are numerous problems with case insensitive matching of characters
2216outside the ASCII range, especially with those whose folds are multiple
2217characters, such as ligatures like C<LATIN SMALL LIGATURE FF>.
2218
f253210b
KW
2219In a bracketed character class with case insensitive matching, ranges only work
2220for ASCII characters. For example,
2221C<m/[\N{CYRILLIC CAPITAL LETTER A}-\N{CYRILLIC CAPITAL LETTER YA}]/i>
2222doesn't match all the Russian upper and lower case letters.
2223
88c9975e
KW
2224Many regular expression constructs don't work on EBCDIC platforms.
2225
9da458fc
IZ
2226This document varies from difficult to understand to completely
2227and utterly opaque. The wandering prose riddled with jargon is
2228hard to fathom in several places.
2229
2230This document needs a rewrite that separates the tutorial content
2231from the reference content.
19799a22
GS
2232
2233=head1 SEE ALSO
9fa51da4 2234
91e0c79e
MJD
2235L<perlrequick>.
2236
2237L<perlretut>.
2238
9b599b2a
GS
2239L<perlop/"Regexp Quote-Like Operators">.
2240
1e66bd83
PP
2241L<perlop/"Gory details of parsing quoted constructs">.
2242
14218588
GS
2243L<perlfaq6>.
2244
9b599b2a
GS
2245L<perlfunc/pos>.
2246
2247L<perllocale>.
2248
fb55449c
JH
2249L<perlebcdic>.
2250
14218588
GS
2251I<Mastering Regular Expressions> by Jeffrey Friedl, published
2252by O'Reilly and Associates.