This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
assert() that SvPV_set() isn't used on arrays.
[perl5.git] / pod / perlre.pod
CommitLineData
a0d0e21e 1=head1 NAME
d74e8afc 2X<regular expression> X<regex> X<regexp>
a0d0e21e
LW
3
4perlre - Perl regular expressions
5
6=head1 DESCRIPTION
7
91e0c79e
MJD
8This page describes the syntax of regular expressions in Perl.
9
cc46d5f2 10If you haven't used regular expressions before, a quick-start
91e0c79e
MJD
11introduction is available in L<perlrequick>, and a longer tutorial
12introduction is available in L<perlretut>.
13
14For reference on how regular expressions are used in matching
15operations, plus various examples of the same, see discussions of
16C<m//>, C<s///>, C<qr//> and C<??> in L<perlop/"Regexp Quote-Like
17Operators">.
cb1a09d0 18
19799a22 19Matching operations can have various modifiers. Modifiers
5a964f20 20that relate to the interpretation of the regular expression inside
19799a22
GS
21are listed below. Modifiers that alter the way a regular expression
22is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
1e66bd83 23L<perlop/"Gory details of parsing quoted constructs">.
a0d0e21e 24
55497cff
PP
25=over 4
26
27=item i
d74e8afc
ITB
28X</i> X<regex, case-insensitive> X<regexp, case-insensitive>
29X<regular expression, case-insensitive>
55497cff
PP
30
31Do case-insensitive pattern matching.
32
a034a98d
DD
33If C<use locale> is in effect, the case map is taken from the current
34locale. See L<perllocale>.
35
54310121 36=item m
d74e8afc 37X</m> X<regex, multiline> X<regexp, multiline> X<regular expression, multiline>
55497cff
PP
38
39Treat string as multiple lines. That is, change "^" and "$" from matching
14218588 40the start or end of the string to matching the start or end of any
7f761169 41line anywhere within the string.
55497cff 42
54310121 43=item s
d74e8afc
ITB
44X</s> X<regex, single-line> X<regexp, single-line>
45X<regular expression, single-line>
55497cff
PP
46
47Treat string as single line. That is, change "." to match any character
19799a22 48whatsoever, even a newline, which normally it would not match.
55497cff 49
f02c194e 50Used together, as /ms, they let the "." match any character whatsoever,
fb55449c 51while still allowing "^" and "$" to match, respectively, just after
19799a22 52and just before newlines within the string.
7b8d334a 53
54310121 54=item x
d74e8afc 55X</x>
55497cff
PP
56
57Extend your pattern's legibility by permitting whitespace and comments.
58
59=back
a0d0e21e
LW
60
61These are usually written as "the C</x> modifier", even though the delimiter
14218588 62in question might not really be a slash. Any of these
a0d0e21e 63modifiers may also be embedded within the regular expression itself using
14218588 64the C<(?...)> construct. See below.
a0d0e21e 65
4633a7c4 66The C</x> modifier itself needs a little more explanation. It tells
55497cff
PP
67the regular expression parser to ignore whitespace that is neither
68backslashed nor within a character class. You can use this to break up
4633a7c4 69your regular expression into (slightly) more readable parts. The C<#>
54310121 70character is also treated as a metacharacter introducing a comment,
55497cff 71just as in ordinary Perl code. This also means that if you want real
14218588 72whitespace or C<#> characters in the pattern (outside a character
f9a3ff1a
RGS
73class, where they are unaffected by C</x>), then you'll either have to
74escape them (using backslashes or C<\Q...\E>) or encode them using octal
8933a740
RGS
75or hex escapes. Taken together, these features go a long way towards
76making Perl's regular expressions more readable. Note that you have to
77be careful not to include the pattern delimiter in the comment--perl has
78no way of knowing you did not intend to close the pattern early. See
79the C-comment deletion code in L<perlop>. Also note that anything inside
1031e5db 80a C<\Q...\E> stays unaffected by C</x>.
d74e8afc 81X</x>
a0d0e21e
LW
82
83=head2 Regular Expressions
84
04838cea
RGS
85=head3 Metacharacters
86
19799a22 87The patterns used in Perl pattern matching derive from supplied in
14218588 88the Version 8 regex routines. (The routines are derived
19799a22
GS
89(distantly) from Henry Spencer's freely redistributable reimplementation
90of the V8 routines.) See L<Version 8 Regular Expressions> for
91details.
a0d0e21e
LW
92
93In particular the following metacharacters have their standard I<egrep>-ish
94meanings:
d74e8afc
ITB
95X<metacharacter>
96X<\> X<^> X<.> X<$> X<|> X<(> X<()> X<[> X<[]>
97
a0d0e21e 98
54310121 99 \ Quote the next metacharacter
a0d0e21e
LW
100 ^ Match the beginning of the line
101 . Match any character (except newline)
c07a80fd 102 $ Match the end of the line (or before newline at the end)
a0d0e21e
LW
103 | Alternation
104 () Grouping
105 [] Character class
106
14218588
GS
107By default, the "^" character is guaranteed to match only the
108beginning of the string, the "$" character only the end (or before the
109newline at the end), and Perl does certain optimizations with the
a0d0e21e
LW
110assumption that the string contains only one line. Embedded newlines
111will not be matched by "^" or "$". You may, however, wish to treat a
4a6725af 112string as a multi-line buffer, such that the "^" will match after any
a0d0e21e
LW
113newline within the string, and "$" will match before any newline. At the
114cost of a little more overhead, you can do this by using the /m modifier
115on the pattern match operator. (Older programs did this by setting C<$*>,
f02c194e 116but this practice has been removed in perl 5.9.)
d74e8afc 117X<^> X<$> X</m>
a0d0e21e 118
14218588 119To simplify multi-line substitutions, the "." character never matches a
55497cff 120newline unless you use the C</s> modifier, which in effect tells Perl to pretend
f02c194e 121the string is a single line--even if it isn't.
d74e8afc 122X<.> X</s>
a0d0e21e 123
04838cea
RGS
124=head3 Quantifiers
125
a0d0e21e 126The following standard quantifiers are recognized:
d74e8afc 127X<metacharacter> X<quantifier> X<*> X<+> X<?> X<{n}> X<{n,}> X<{n,m}>
a0d0e21e
LW
128
129 * Match 0 or more times
130 + Match 1 or more times
131 ? Match 1 or 0 times
132 {n} Match exactly n times
133 {n,} Match at least n times
134 {n,m} Match at least n but not more than m times
135
136(If a curly bracket occurs in any other context, it is treated
b975c076
JL
137as a regular character. In particular, the lower bound
138is not optional.) The "*" modifier is equivalent to C<{0,}>, the "+"
25f94b33 139modifier to C<{1,}>, and the "?" modifier to C<{0,1}>. n and m are limited
9c79236d
GS
140to integral values less than a preset limit defined when perl is built.
141This is usually 32766 on the most common platforms. The actual limit can
142be seen in the error message generated by code such as this:
143
820475bd 144 $_ **= $_ , / {$_} / for 2 .. 42;
a0d0e21e 145
54310121
PP
146By default, a quantified subpattern is "greedy", that is, it will match as
147many times as possible (given a particular starting location) while still
148allowing the rest of the pattern to match. If you want it to match the
149minimum number of times possible, follow the quantifier with a "?". Note
150that the meanings don't change, just the "greediness":
d74e8afc
ITB
151X<metacharacter> X<greedy> X<greedyness>
152X<?> X<*?> X<+?> X<??> X<{n}?> X<{n,}?> X<{n,m}?>
a0d0e21e
LW
153
154 *? Match 0 or more times
155 +? Match 1 or more times
156 ?? Match 0 or 1 time
157 {n}? Match exactly n times
158 {n,}? Match at least n times
159 {n,m}? Match at least n but not more than m times
160
b9b4dddf
YO
161By default, when a quantified subpattern does not allow the rest of the
162overall pattern to match, Perl will backtrack. However, this behaviour is
163sometimes undesirable. Thus Perl provides the "possesive" quantifier form
164as well.
165
166 *+ Match 0 or more times and give nothing back
04838cea 167 ++ Match 1 or more times and give nothing back
b9b4dddf
YO
168 ?+ Match 0 or 1 time and give nothing back
169 {n}+ Match exactly n times and give nothing back (redundant)
04838cea
RGS
170 {n,}+ Match at least n times and give nothing back
171 {n,m}+ Match at least n but not more than m times and give nothing back
b9b4dddf
YO
172
173For instance,
174
175 'aaaa' =~ /a++a/
176
177will never match, as the C<a++> will gobble up all the C<a>'s in the
178string and won't leave any for the remaining part of the pattern. This
179feature can be extremely useful to give perl hints about where it
180shouldn't backtrack. For instance, the typical "match a double-quoted
181string" problem can be most efficiently performed when written as:
182
183 /"(?:[^"\\]++|\\.)*+"/
184
185as we know that if the final quote does not match, bactracking will not
186help. See the independent subexpression C<< (?>...) >> for more details;
187possessive quantifiers are just syntactic sugar for that construct. For
188instance the above example could also be written as follows:
189
190 /"(?>(?:(?>[^"\\]+)|\\.)*)"/
191
04838cea
RGS
192=head3 Escape sequences
193
5f05dabc 194Because patterns are processed as double quoted strings, the following
a0d0e21e 195also work:
d74e8afc
ITB
196X<\t> X<\n> X<\r> X<\f> X<\a> X<\l> X<\u> X<\L> X<\U> X<\E> X<\Q>
197X<\0> X<\c> X<\N> X<\x>
a0d0e21e 198
0f36ee90
PP
199 \t tab (HT, TAB)
200 \n newline (LF, NL)
201 \r return (CR)
202 \f form feed (FF)
203 \a alarm (bell) (BEL)
204 \e escape (think troff) (ESC)
cb1a09d0
AD
205 \033 octal char (think of a PDP-11)
206 \x1B hex char
a0ed51b3 207 \x{263a} wide hex char (Unicode SMILEY)
a0d0e21e 208 \c[ control char
4a2d328f 209 \N{name} named char
cb1a09d0
AD
210 \l lowercase next char (think vi)
211 \u uppercase next char (think vi)
212 \L lowercase till \E (think vi)
213 \U uppercase till \E (think vi)
214 \E end case modification (think vi)
5a964f20 215 \Q quote (disable) pattern metacharacters till \E
a0d0e21e 216
a034a98d 217If C<use locale> is in effect, the case map used by C<\l>, C<\L>, C<\u>
423cee85 218and C<\U> is taken from the current locale. See L<perllocale>. For
4a2d328f 219documentation of C<\N{name}>, see L<charnames>.
a034a98d 220
1d2dff63
GS
221You cannot include a literal C<$> or C<@> within a C<\Q> sequence.
222An unescaped C<$> or C<@> interpolates the corresponding variable,
223while escaping will cause the literal string C<\$> to be matched.
224You'll need to write something like C<m/\Quser\E\@\Qhost/>.
225
04838cea
RGS
226=head3 Character classes
227
a0d0e21e 228In addition, Perl defines the following:
d74e8afc
ITB
229X<metacharacter>
230X<\w> X<\W> X<\s> X<\S> X<\d> X<\D> X<\X> X<\p> X<\P> X<\C>
231X<word> X<whitespace>
a0d0e21e 232
81714fb9
YO
233 \w Match a "word" character (alphanumeric plus "_")
234 \W Match a non-"word" character
235 \s Match a whitespace character
236 \S Match a non-whitespace character
237 \d Match a digit character
238 \D Match a non-digit character
239 \pP Match P, named property. Use \p{Prop} for longer names.
240 \PP Match non-P
241 \X Match eXtended Unicode "combining character sequence",
242 equivalent to (?:\PM\pM*)
243 \C Match a single C char (octet) even under Unicode.
244 NOTE: breaks up characters into their UTF-8 bytes,
245 so you may end up with malformed pieces of UTF-8.
246 Unsupported in lookbehind.
247 \1 Backreference to a a specific group.
248 '1' may actually be any positive integer
249 \k<name> Named backreference
250 \N{name} Named unicode character, or unicode escape.
251 \x12 Hexadecimal escape sequence
252 \x{1234} Long hexadecimal escape sequence
a0d0e21e 253
08ce8fc6
JH
254A C<\w> matches a single alphanumeric character (an alphabetic
255character, or a decimal digit) or C<_>, not a whole word. Use C<\w+>
256to match a string of Perl-identifier characters (which isn't the same
257as matching an English word). If C<use locale> is in effect, the list
258of alphabetic characters generated by C<\w> is taken from the current
259locale. See L<perllocale>. You may use C<\w>, C<\W>, C<\s>, C<\S>,
1209ba90 260C<\d>, and C<\D> within character classes, but if you try to use them
08ce8fc6
JH
261as endpoints of a range, that's not a range, the "-" is understood
262literally. If Unicode is in effect, C<\s> matches also "\x{85}",
263"\x{2028}, and "\x{2029}", see L<perlunicode> for more details about
491fd90a 264C<\pP>, C<\PP>, and C<\X>, and L<perluniintro> about Unicode in general.
fa11829f 265You can define your own C<\p> and C<\P> properties, see L<perlunicode>.
d74e8afc 266X<\w> X<\W> X<word>
a0d0e21e 267
b8c5462f 268The POSIX character class syntax
d74e8afc 269X<character class>
b8c5462f 270
820475bd 271 [:class:]
b8c5462f 272
5496314a
SP
273is also available. Note that the C<[> and C<]> braces are I<literal>;
274they must always be used within a character class expression.
275
276 # this is correct:
277 $string =~ /[[:alpha:]]/;
278
279 # this is not, and will generate a warning:
280 $string =~ /[:alpha:]/;
281
282The available classes and their backslash equivalents (if available) are
283as follows:
d74e8afc
ITB
284X<character class>
285X<alpha> X<alnum> X<ascii> X<blank> X<cntrl> X<digit> X<graph>
286X<lower> X<print> X<punct> X<space> X<upper> X<word> X<xdigit>
b8c5462f
JH
287
288 alpha
289 alnum
290 ascii
aaa51d5e 291 blank [1]
b8c5462f
JH
292 cntrl
293 digit \d
294 graph
295 lower
296 print
297 punct
aaa51d5e 298 space \s [2]
b8c5462f 299 upper
aaa51d5e 300 word \w [3]
b8c5462f
JH
301 xdigit
302
07698885
RGS
303=over
304
305=item [1]
306
b432a672 307A GNU extension equivalent to C<[ \t]>, "all horizontal whitespace".
07698885
RGS
308
309=item [2]
310
311Not exactly equivalent to C<\s> since the C<[[:space:]]> includes
b432a672 312also the (very rare) "vertical tabulator", "\ck", chr(11).
07698885
RGS
313
314=item [3]
315
08ce8fc6 316A Perl extension, see above.
07698885
RGS
317
318=back
aaa51d5e 319
26b44a0a 320For example use C<[:upper:]> to match all the uppercase characters.
aaa51d5e
JF
321Note that the C<[]> are part of the C<[::]> construct, not part of the
322whole character class. For example:
b8c5462f 323
820475bd 324 [01[:alpha:]%]
b8c5462f 325
593df60c 326matches zero, one, any alphabetic character, and the percentage sign.
b8c5462f 327
72ff2908
JH
328The following equivalences to Unicode \p{} constructs and equivalent
329backslash character classes (if available), will hold:
d74e8afc 330X<character class> X<\p> X<\p{}>
72ff2908 331
5496314a 332 [[:...:]] \p{...} backslash
b8c5462f
JH
333
334 alpha IsAlpha
335 alnum IsAlnum
336 ascii IsASCII
b432a672 337 blank IsSpace
b8c5462f 338 cntrl IsCntrl
3bec3564 339 digit IsDigit \d
b8c5462f
JH
340 graph IsGraph
341 lower IsLower
342 print IsPrint
343 punct IsPunct
344 space IsSpace
3bec3564 345 IsSpacePerl \s
b8c5462f
JH
346 upper IsUpper
347 word IsWord
348 xdigit IsXDigit
349
5496314a 350For example C<[[:lower:]]> and C<\p{IsLower}> are equivalent.
b8c5462f
JH
351
352If the C<utf8> pragma is not used but the C<locale> pragma is, the
aaa51d5e 353classes correlate with the usual isalpha(3) interface (except for
b432a672 354"word" and "blank").
b8c5462f
JH
355
356The assumedly non-obviously named classes are:
357
358=over 4
359
360=item cntrl
d74e8afc 361X<cntrl>
b8c5462f 362
820475bd
GS
363Any control character. Usually characters that don't produce output as
364such but instead control the terminal somehow: for example newline and
365backspace are control characters. All characters with ord() less than
593df60c 36632 are most often classified as control characters (assuming ASCII,
7be5a6cf
JF
367the ISO Latin character sets, and Unicode), as is the character with
368the ord() value of 127 (C<DEL>).
b8c5462f
JH
369
370=item graph
d74e8afc 371X<graph>
b8c5462f 372
f1cbbd6e 373Any alphanumeric or punctuation (special) character.
b8c5462f
JH
374
375=item print
d74e8afc 376X<print>
b8c5462f 377
f79b3095 378Any alphanumeric or punctuation (special) character or the space character.
b8c5462f
JH
379
380=item punct
d74e8afc 381X<punct>
b8c5462f 382
f1cbbd6e 383Any punctuation (special) character.
b8c5462f
JH
384
385=item xdigit
d74e8afc 386X<xdigit>
b8c5462f 387
593df60c 388Any hexadecimal digit. Though this may feel silly ([0-9A-Fa-f] would
820475bd 389work just fine) it is included for completeness.
b8c5462f 390
b8c5462f
JH
391=back
392
393You can negate the [::] character classes by prefixing the class name
394with a '^'. This is a Perl extension. For example:
d74e8afc 395X<character class, negation>
b8c5462f 396
5496314a 397 POSIX traditional Unicode
93733859 398
5496314a
SP
399 [[:^digit:]] \D \P{IsDigit}
400 [[:^space:]] \S \P{IsSpace}
401 [[:^word:]] \W \P{IsWord}
b8c5462f 402
54c18d04
MK
403Perl respects the POSIX standard in that POSIX character classes are
404only supported within a character class. The POSIX character classes
405[.cc.] and [=cc=] are recognized but B<not> supported and trying to
406use them will cause an error.
b8c5462f 407
04838cea
RGS
408=head3 Assertions
409
a0d0e21e 410Perl defines the following zero-width assertions:
d74e8afc
ITB
411X<zero-width assertion> X<assertion> X<regex, zero-width assertion>
412X<regexp, zero-width assertion>
413X<regular expression, zero-width assertion>
414X<\b> X<\B> X<\A> X<\Z> X<\z> X<\G>
a0d0e21e
LW
415
416 \b Match a word boundary
417 \B Match a non-(word boundary)
b85d18e9
IZ
418 \A Match only at beginning of string
419 \Z Match only at end of string, or before newline at the end
420 \z Match only at end of string
9da458fc
IZ
421 \G Match only at pos() (e.g. at the end-of-match position
422 of prior m//g)
a0d0e21e 423
14218588 424A word boundary (C<\b>) is a spot between two characters
19799a22
GS
425that has a C<\w> on one side of it and a C<\W> on the other side
426of it (in either order), counting the imaginary characters off the
427beginning and end of the string as matching a C<\W>. (Within
428character classes C<\b> represents backspace rather than a word
429boundary, just as it normally does in any double-quoted string.)
430The C<\A> and C<\Z> are just like "^" and "$", except that they
431won't match multiple times when the C</m> modifier is used, while
432"^" and "$" will match at every internal line boundary. To match
433the actual end of the string and not ignore an optional trailing
434newline, use C<\z>.
d74e8afc 435X<\b> X<\A> X<\Z> X<\z> X</m>
19799a22
GS
436
437The C<\G> assertion can be used to chain global matches (using
438C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
439It is also useful when writing C<lex>-like scanners, when you have
440several patterns that you want to match against consequent substrings
441of your string, see the previous reference. The actual location
442where C<\G> will match can also be influenced by using C<pos()> as
25cf8c22
HS
443an lvalue: see L<perlfunc/pos>. Currently C<\G> is only fully
444supported when anchored to the start of the pattern; while it
445is permitted to use it elsewhere, as in C</(?<=\G..)./g>, some
446such uses (C</.\G/g>, for example) currently cause problems, and
447it is recommended that you avoid such usage for now.
d74e8afc 448X<\G>
c47ff5f1 449
04838cea
RGS
450=head3 Capture buffers
451
14218588 452The bracketing construct C<( ... )> creates capture buffers. To
c47ff5f1 453refer to the digit'th buffer use \<digit> within the
14218588 454match. Outside the match use "$" instead of "\". (The
81714fb9 455\<digit> notation works in certain circumstances outside
14218588
GS
456the match. See the warning below about \1 vs $1 for details.)
457Referring back to another part of the match is called a
458I<backreference>.
d74e8afc
ITB
459X<regex, capture buffer> X<regexp, capture buffer>
460X<regular expression, capture buffer> X<backreference>
14218588
GS
461
462There is no limit to the number of captured substrings that you may
463use. However Perl also uses \10, \11, etc. as aliases for \010,
fb55449c
JH
464\011, etc. (Recall that 0 means octal, so \011 is the character at
465number 9 in your coded character set; which would be the 10th character,
81714fb9
YO
466a horizontal tab under ASCII.) Perl resolves this
467ambiguity by interpreting \10 as a backreference only if at least 10
468left parentheses have opened before it. Likewise \11 is a
469backreference only if at least 11 left parentheses have opened
470before it. And so on. \1 through \9 are always interpreted as
fb55449c 471backreferences.
14218588 472
81714fb9
YO
473Additionally, as of Perl 5.10 you may use named capture buffers and named
474backreferences. The notation is C<< (?<name>...) >> and C<< \k<name> >>
475(you may also use single quotes instead of angle brackets to quote the
476name). The only difference with named capture buffers and unnamed ones is
477that multiple buffers may have the same name and that the contents of
478named capture buffers is available via the C<%+> hash. When multiple
479groups share the same name C<$+{name}> and C<< \k<name> >> refer to the
480leftmost defined group, thus it's possible to do things with named capture
481buffers that would otherwise require C<(??{})> code to accomplish. Named
482capture buffers are numbered just as normal capture buffers are and may be
483referenced via the magic numeric variables or via numeric backreferences
484as well as by name.
485
14218588 486Examples:
a0d0e21e
LW
487
488 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
489
81714fb9
YO
490 /(.)\1/ # find first doubled char
491 and print "'$1' is the first doubled character\n";
492
493 /(?<char>.)\k<char>/ # ... a different way
494 and print "'$+{char}' is the first doubled character\n";
495
496 /(?<char>.)\1/ # ... mix and match
497 and print "'$1' is the first doubled character\n";
c47ff5f1 498
14218588 499 if (/Time: (..):(..):(..)/) { # parse out values
a0d0e21e
LW
500 $hours = $1;
501 $minutes = $2;
502 $seconds = $3;
503 }
c47ff5f1 504
14218588
GS
505Several special variables also refer back to portions of the previous
506match. C<$+> returns whatever the last bracket match matched.
507C<$&> returns the entire matched string. (At one point C<$0> did
508also, but now it returns the name of the program.) C<$`> returns
77ea4f6d
JV
509everything before the matched string. C<$'> returns everything
510after the matched string. And C<$^N> contains whatever was matched by
511the most-recently closed group (submatch). C<$^N> can be used in
512extended patterns (see below), for example to assign a submatch to a
81714fb9 513variable.
d74e8afc 514X<$+> X<$^N> X<$&> X<$`> X<$'>
14218588 515
665e98b9 516The numbered match variables ($1, $2, $3, etc.) and the related punctuation
77ea4f6d 517set (C<$+>, C<$&>, C<$`>, C<$'>, and C<$^N>) are all dynamically scoped
14218588
GS
518until the end of the enclosing block or until the next successful
519match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
d74e8afc
ITB
520X<$+> X<$^N> X<$&> X<$`> X<$'>
521X<$1> X<$2> X<$3> X<$4> X<$5> X<$6> X<$7> X<$8> X<$9>
522
14218588 523
665e98b9 524B<NOTE>: failed matches in Perl do not reset the match variables,
5146ce24 525which makes it easier to write code that tests for a series of more
665e98b9
JH
526specific cases and remembers the best match.
527
14218588
GS
528B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
529C<$'> anywhere in the program, it has to provide them for every
530pattern match. This may substantially slow your program. Perl
531uses the same mechanism to produce $1, $2, etc, so you also pay a
532price for each pattern that contains capturing parentheses. (To
533avoid this cost while retaining the grouping behaviour, use the
534extended regular expression C<(?: ... )> instead.) But if you never
535use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
536parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
537if you can, but if you can't (and some algorithms really appreciate
538them), once you've used them once, use them at will, because you've
539already paid the price. As of 5.005, C<$&> is not so costly as the
540other two.
d74e8afc 541X<$&> X<$`> X<$'>
68dc0745 542
19799a22
GS
543Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
544C<\w>, C<\n>. Unlike some other regular expression languages, there
545are no backslashed symbols that aren't alphanumeric. So anything
c47ff5f1 546that looks like \\, \(, \), \<, \>, \{, or \} is always
19799a22
GS
547interpreted as a literal character, not a metacharacter. This was
548once used in a common idiom to disable or quote the special meanings
549of regular expression metacharacters in a string that you want to
36bbe248 550use for a pattern. Simply quote all non-"word" characters:
a0d0e21e
LW
551
552 $pattern =~ s/(\W)/\\$1/g;
553
f1cbbd6e 554(If C<use locale> is set, then this depends on the current locale.)
14218588
GS
555Today it is more common to use the quotemeta() function or the C<\Q>
556metaquoting escape sequence to disable all metacharacters' special
557meanings like this:
a0d0e21e
LW
558
559 /$unquoted\Q$quoted\E$unquoted/
560
9da458fc
IZ
561Beware that if you put literal backslashes (those not inside
562interpolated variables) between C<\Q> and C<\E>, double-quotish
563backslash interpolation may lead to confusing results. If you
564I<need> to use literal backslashes within C<\Q...\E>,
565consult L<perlop/"Gory details of parsing quoted constructs">.
566
19799a22
GS
567=head2 Extended Patterns
568
14218588
GS
569Perl also defines a consistent extension syntax for features not
570found in standard tools like B<awk> and B<lex>. The syntax is a
571pair of parentheses with a question mark as the first thing within
572the parentheses. The character after the question mark indicates
573the extension.
19799a22 574
14218588
GS
575The stability of these extensions varies widely. Some have been
576part of the core language for many years. Others are experimental
577and may change without warning or be completely removed. Check
578the documentation on an individual feature to verify its current
579status.
19799a22 580
14218588
GS
581A question mark was chosen for this and for the minimal-matching
582construct because 1) question marks are rare in older regular
583expressions, and 2) whenever you see one, you should stop and
584"question" exactly what is going on. That's psychology...
a0d0e21e
LW
585
586=over 10
587
cc6b7395 588=item C<(?#text)>
d74e8afc 589X<(?#)>
a0d0e21e 590
14218588 591A comment. The text is ignored. If the C</x> modifier enables
19799a22 592whitespace formatting, a simple C<#> will suffice. Note that Perl closes
259138e3
GS
593the comment as soon as it sees a C<)>, so there is no way to put a literal
594C<)> in the comment.
a0d0e21e 595
19799a22 596=item C<(?imsx-imsx)>
d74e8afc 597X<(?)>
19799a22 598
0b6d1084
JH
599One or more embedded pattern-match modifiers, to be turned on (or
600turned off, if preceded by C<->) for the remainder of the pattern or
601the remainder of the enclosing pattern group (if any). This is
602particularly useful for dynamic patterns, such as those read in from a
603configuration file, read in as an argument, are specified in a table
604somewhere, etc. Consider the case that some of which want to be case
605sensitive and some do not. The case insensitive ones need to include
606merely C<(?i)> at the front of the pattern. For example:
19799a22
GS
607
608 $pattern = "foobar";
609 if ( /$pattern/i ) { }
610
611 # more flexible:
612
613 $pattern = "(?i)foobar";
614 if ( /$pattern/ ) { }
615
0b6d1084 616These modifiers are restored at the end of the enclosing group. For example,
19799a22
GS
617
618 ( (?i) blah ) \s+ \1
619
620will match a repeated (I<including the case>!) word C<blah> in any
14218588 621case, assuming C<x> modifier, and no C<i> modifier outside this
19799a22
GS
622group.
623
5a964f20 624=item C<(?:pattern)>
d74e8afc 625X<(?:)>
a0d0e21e 626
ca9dfc88
IZ
627=item C<(?imsx-imsx:pattern)>
628
5a964f20
TC
629This is for clustering, not capturing; it groups subexpressions like
630"()", but doesn't make backreferences as "()" does. So
a0d0e21e 631
5a964f20 632 @fields = split(/\b(?:a|b|c)\b/)
a0d0e21e
LW
633
634is like
635
5a964f20 636 @fields = split(/\b(a|b|c)\b/)
a0d0e21e 637
19799a22
GS
638but doesn't spit out extra fields. It's also cheaper not to capture
639characters if you don't need to.
a0d0e21e 640
19799a22
GS
641Any letters between C<?> and C<:> act as flags modifiers as with
642C<(?imsx-imsx)>. For example,
ca9dfc88
IZ
643
644 /(?s-i:more.*than).*million/i
645
14218588 646is equivalent to the more verbose
ca9dfc88
IZ
647
648 /(?:(?s-i)more.*than).*million/i
649
5a964f20 650=item C<(?=pattern)>
d74e8afc 651X<(?=)> X<look-ahead, positive> X<lookahead, positive>
a0d0e21e 652
19799a22 653A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
a0d0e21e
LW
654matches a word followed by a tab, without including the tab in C<$&>.
655
5a964f20 656=item C<(?!pattern)>
d74e8afc 657X<(?!)> X<look-ahead, negative> X<lookahead, negative>
a0d0e21e 658
19799a22 659A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
a0d0e21e 660matches any occurrence of "foo" that isn't followed by "bar". Note
19799a22
GS
661however that look-ahead and look-behind are NOT the same thing. You cannot
662use this for look-behind.
7b8d334a 663
5a964f20 664If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
7b8d334a
GS
665will not do what you want. That's because the C<(?!foo)> is just saying that
666the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
667match. You would have to do something like C</(?!foo)...bar/> for that. We
668say "like" because there's the case of your "bar" not having three characters
669before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
670Sometimes it's still easier just to say:
a0d0e21e 671
a3cb178b 672 if (/bar/ && $` !~ /foo$/)
a0d0e21e 673
19799a22 674For look-behind see below.
c277df42 675
c47ff5f1 676=item C<(?<=pattern)>
d74e8afc 677X<(?<=)> X<look-behind, positive> X<lookbehind, positive>
c277df42 678
c47ff5f1 679A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
19799a22
GS
680matches a word that follows a tab, without including the tab in C<$&>.
681Works only for fixed-width look-behind.
c277df42 682
5a964f20 683=item C<(?<!pattern)>
d74e8afc 684X<(?<!)> X<look-behind, negative> X<lookbehind, negative>
c277df42 685
19799a22
GS
686A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
687matches any occurrence of "foo" that does not follow "bar". Works
688only for fixed-width look-behind.
c277df42 689
81714fb9
YO
690=item C<(?'NAME'pattern)>
691
692=item C<< (?<NAME>pattern) >>
693X<< (?<NAME>) >> X<(?'NAME')> X<named capture> X<capture>
694
695A named capture buffer. Identical in every respect to normal capturing
696parens C<()> but for the additional fact that C<%+> may be used after
697a succesful match to refer to a named buffer. See C<perlvar> for more
698details on the C<%+> hash.
699
700If multiple distinct capture buffers have the same name then the
701$+{NAME} will refer to the leftmost defined buffer in the match.
702
703The forms C<(?'NAME'pattern)> and C<(?<NAME>pattern)> are equivalent.
704
705B<NOTE:> While the notation of this construct is the same as the similar
706function in .NET regexes, the behavior is not, in Perl the buffers are
707numbered sequentially regardless of being named or not. Thus in the
708pattern
709
710 /(x)(?<foo>y)(z)/
711
712$+{foo} will be the same as $2, and $3 will contain 'z' instead of
713the opposite which is what a .NET regex hacker might expect.
714
715Currently NAME is restricted to word chars only. In other words, it
716must match C</^\w+$/>.
717
718=item C<< \k<name> >>
719
720=item C<< \k'name' >>
721
722Named backreference. Similar to numeric backreferences, except that
723the group is designated by name and not number. If multiple groups
724have the same name then it refers to the leftmost defined group in
725the current match.
726
727It is an error to refer to a name not defined by a C<(?<NAME>)>
728earlier in the pattern.
729
730Both forms are equivalent.
731
cc6b7395 732=item C<(?{ code })>
d74e8afc 733X<(?{})> X<regex, code in> X<regexp, code in> X<regular expression, code in>
c277df42 734
19799a22 735B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
736experimental, and may be changed without notice. Code executed that
737has side effects may not perform identically from version to version
738due to the effect of future optimisations in the regex engine.
c277df42 739
cc46d5f2 740This zero-width assertion evaluates any embedded Perl code. It
19799a22
GS
741always succeeds, and its C<code> is not interpolated. Currently,
742the rules to determine where the C<code> ends are somewhat convoluted.
743
77ea4f6d
JV
744This feature can be used together with the special variable C<$^N> to
745capture the results of submatches in variables without having to keep
746track of the number of nested parentheses. For example:
747
748 $_ = "The brown fox jumps over the lazy dog";
749 /the (\S+)(?{ $color = $^N }) (\S+)(?{ $animal = $^N })/i;
750 print "color = $color, animal = $animal\n";
751
754091cb
RGS
752Inside the C<(?{...})> block, C<$_> refers to the string the regular
753expression is matching against. You can also use C<pos()> to know what is
fa11829f 754the current position of matching within this string.
754091cb 755
19799a22
GS
756The C<code> is properly scoped in the following sense: If the assertion
757is backtracked (compare L<"Backtracking">), all changes introduced after
758C<local>ization are undone, so that
b9ac3b5b
GS
759
760 $_ = 'a' x 8;
761 m<
762 (?{ $cnt = 0 }) # Initialize $cnt.
763 (
764 a
765 (?{
766 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
767 })
768 )*
769 aaaa
770 (?{ $res = $cnt }) # On success copy to non-localized
771 # location.
772 >x;
773
19799a22 774will set C<$res = 4>. Note that after the match, $cnt returns to the globally
14218588 775introduced value, because the scopes that restrict C<local> operators
b9ac3b5b
GS
776are unwound.
777
19799a22
GS
778This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
779switch. If I<not> used in this way, the result of evaluation of
780C<code> is put into the special variable C<$^R>. This happens
781immediately, so C<$^R> can be used from other C<(?{ code })> assertions
782inside the same regular expression.
b9ac3b5b 783
19799a22
GS
784The assignment to C<$^R> above is properly localized, so the old
785value of C<$^R> is restored if the assertion is backtracked; compare
786L<"Backtracking">.
b9ac3b5b 787
61528107
SP
788Due to an unfortunate implementation issue, the Perl code contained in these
789blocks is treated as a compile time closure that can have seemingly bizarre
6bda09f9 790consequences when used with lexically scoped variables inside of subroutines
61528107
SP
791or loops. There are various workarounds for this, including simply using
792global variables instead. If you are using this construct and strange results
6bda09f9
YO
793occur then check for the use of lexically scoped variables.
794
19799a22
GS
795For reasons of security, this construct is forbidden if the regular
796expression involves run-time interpolation of variables, unless the
797perilous C<use re 'eval'> pragma has been used (see L<re>), or the
798variables contain results of C<qr//> operator (see
799L<perlop/"qr/STRING/imosx">).
871b0233 800
14218588 801This restriction is because of the wide-spread and remarkably convenient
19799a22 802custom of using run-time determined strings as patterns. For example:
871b0233
IZ
803
804 $re = <>;
805 chomp $re;
806 $string =~ /$re/;
807
14218588
GS
808Before Perl knew how to execute interpolated code within a pattern,
809this operation was completely safe from a security point of view,
810although it could raise an exception from an illegal pattern. If
811you turn on the C<use re 'eval'>, though, it is no longer secure,
812so you should only do so if you are also using taint checking.
813Better yet, use the carefully constrained evaluation within a Safe
cc46d5f2 814compartment. See L<perlsec> for details about both these mechanisms.
871b0233 815
8988a1bb
DD
816Because perl's regex engine is not currently re-entrant, interpolated
817code may not invoke the regex engine either directly with C<m//> or C<s///>),
818or indirectly with functions such as C<split>.
819
14455d6c 820=item C<(??{ code })>
d74e8afc
ITB
821X<(??{})>
822X<regex, postponed> X<regexp, postponed> X<regular expression, postponed>
0f5d15d6 823
19799a22 824B<WARNING>: This extended regular expression feature is considered
b9b4dddf
YO
825experimental, and may be changed without notice. Code executed that
826has side effects may not perform identically from version to version
827due to the effect of future optimisations in the regex engine.
0f5d15d6 828
19799a22
GS
829This is a "postponed" regular subexpression. The C<code> is evaluated
830at run time, at the moment this subexpression may match. The result
831of evaluation is considered as a regular expression and matched as
61528107 832if it were inserted instead of this construct. Note that this means
6bda09f9
YO
833that the contents of capture buffers defined inside an eval'ed pattern
834are not available outside of the pattern, and vice versa, there is no
835way for the inner pattern to refer to a capture buffer defined outside.
836Thus,
837
838 ('a' x 100)=~/(??{'(.)' x 100})/
839
81714fb9 840B<will> match, it will B<not> set $1.
0f5d15d6 841
428594d9 842The C<code> is not interpolated. As before, the rules to determine
19799a22
GS
843where the C<code> ends are currently somewhat convoluted.
844
845The following pattern matches a parenthesized group:
0f5d15d6
IZ
846
847 $re = qr{
848 \(
849 (?:
850 (?> [^()]+ ) # Non-parens without backtracking
851 |
14455d6c 852 (??{ $re }) # Group with matching parens
0f5d15d6
IZ
853 )*
854 \)
855 }x;
856
6bda09f9
YO
857See also C<(?PARNO)> for a different, more efficient way to accomplish
858the same task.
859
8988a1bb
DD
860Because perl's regex engine is not currently re-entrant, delayed
861code may not invoke the regex engine either directly with C<m//> or C<s///>),
862or indirectly with functions such as C<split>.
863
6bda09f9 864Recursing deeper than 50 times without consuming any input string will
61528107 865result in a fatal error. The maximum depth is compiled into perl, so
6bda09f9
YO
866changing it requires a custom build.
867
894be9b7
YO
868=item C<(?PARNO)> C<(?R)> C<(?0)>
869X<(?PARNO)> X<(?1)> X<(?R)> X<(?0)>
6bda09f9
YO
870X<regex, recursive> X<regexp, recursive> X<regular expression, recursive>
871
81714fb9
YO
872Similar to C<(??{ code })> except it does not involve compiling any code,
873instead it treats the contents of a capture buffer as an independent
61528107 874pattern that must match at the current position. Capture buffers
81714fb9 875contained by the pattern will have the value as determined by the
6bda09f9
YO
876outermost recursion.
877
894be9b7
YO
878PARNO is a sequence of digits (not starting with 0) whose value reflects
879the paren-number of the capture buffer to recurse to. C<(?R)> recurses to
880the beginning of the whole pattern. C<(?0)> is an alternate syntax for
881C<(?R)>.
6bda09f9 882
81714fb9 883The following pattern matches a function foo() which may contain
f145b7e9 884balanced parentheses as the argument.
6bda09f9
YO
885
886 $re = qr{ ( # paren group 1 (full function)
81714fb9 887 foo
6bda09f9
YO
888 ( # paren group 2 (parens)
889 \(
890 ( # paren group 3 (contents of parens)
891 (?:
892 (?> [^()]+ ) # Non-parens without backtracking
893 |
894 (?2) # Recurse to start of paren group 2
895 )*
896 )
897 \)
898 )
899 )
900 }x;
901
902If the pattern was used as follows
903
904 'foo(bar(baz)+baz(bop))'=~/$re/
905 and print "\$1 = $1\n",
906 "\$2 = $2\n",
907 "\$3 = $3\n";
908
909the output produced should be the following:
910
911 $1 = foo(bar(baz)+baz(bop))
912 $2 = (bar(baz)+baz(bop))
81714fb9 913 $3 = bar(baz)+baz(bop)
6bda09f9 914
81714fb9 915If there is no corresponding capture buffer defined, then it is a
61528107 916fatal error. Recursing deeper than 50 times without consuming any input
81714fb9 917string will also result in a fatal error. The maximum depth is compiled
6bda09f9
YO
918into perl, so changing it requires a custom build.
919
81714fb9 920B<Note> that this pattern does not behave the same way as the equivalent
6bda09f9
YO
921PCRE or Python construct of the same form. In perl you can backtrack into
922a recursed group, in PCRE and Python the recursed into group is treated
81714fb9
YO
923as atomic. Also, constructs like (?i:(?1)) or (?:(?i)(?1)) do not affect
924the pattern being recursed into.
6bda09f9 925
894be9b7
YO
926=item C<(?&NAME)>
927X<(?&NAME)>
928
929Recurse to a named subpattern. Identical to (?PARNO) except that the
930parenthesis to recurse to is determined by name. If multiple parens have
931the same name, then it recurses to the leftmost.
932
933It is an error to refer to a name that is not declared somewhere in the
934pattern.
935
286f584a
RGS
936=item C<(?FAIL)> C<(?F)>
937X<(?FAIL)> X<(?F)>
938
939This pattern matches nothing and always fails. It can be used to force the
940engine to backtrack. It is equivalent to C<(?!)>, but easier to read. In
941fact, C<(?!)> gets optimised into C<(?FAIL)> internally.
942
943It is probably useful only when combined with C<(?{})> or C<(??{})>.
944
c47ff5f1 945=item C<< (?>pattern) >>
6bda09f9 946X<backtrack> X<backtracking> X<atomic> X<possessive>
5a964f20 947
19799a22
GS
948An "independent" subexpression, one which matches the substring
949that a I<standalone> C<pattern> would match if anchored at the given
9da458fc 950position, and it matches I<nothing other than this substring>. This
19799a22
GS
951construct is useful for optimizations of what would otherwise be
952"eternal" matches, because it will not backtrack (see L<"Backtracking">).
9da458fc
IZ
953It may also be useful in places where the "grab all you can, and do not
954give anything back" semantic is desirable.
19799a22 955
c47ff5f1 956For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
19799a22
GS
957(anchored at the beginning of string, as above) will match I<all>
958characters C<a> at the beginning of string, leaving no C<a> for
959C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
960since the match of the subgroup C<a*> is influenced by the following
961group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
962C<a*ab> will match fewer characters than a standalone C<a*>, since
963this makes the tail match.
964
c47ff5f1 965An effect similar to C<< (?>pattern) >> may be achieved by writing
19799a22
GS
966C<(?=(pattern))\1>. This matches the same substring as a standalone
967C<a+>, and the following C<\1> eats the matched string; it therefore
c47ff5f1 968makes a zero-length assertion into an analogue of C<< (?>...) >>.
19799a22
GS
969(The difference between these two constructs is that the second one
970uses a capturing group, thus shifting ordinals of backreferences
971in the rest of a regular expression.)
972
973Consider this pattern:
c277df42 974
871b0233
IZ
975 m{ \(
976 (
9da458fc 977 [^()]+ # x+
871b0233
IZ
978 |
979 \( [^()]* \)
980 )+
981 \)
982 }x
5a964f20 983
19799a22
GS
984That will efficiently match a nonempty group with matching parentheses
985two levels deep or less. However, if there is no such group, it
986will take virtually forever on a long string. That's because there
987are so many different ways to split a long string into several
988substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
989to a subpattern of the above pattern. Consider how the pattern
990above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
991seconds, but that each extra letter doubles this time. This
992exponential performance will make it appear that your program has
14218588 993hung. However, a tiny change to this pattern
5a964f20 994
871b0233
IZ
995 m{ \(
996 (
9da458fc 997 (?> [^()]+ ) # change x+ above to (?> x+ )
871b0233
IZ
998 |
999 \( [^()]* \)
1000 )+
1001 \)
1002 }x
c277df42 1003
c47ff5f1 1004which uses C<< (?>...) >> matches exactly when the one above does (verifying
5a964f20
TC
1005this yourself would be a productive exercise), but finishes in a fourth
1006the time when used on a similar string with 1000000 C<a>s. Be aware,
1007however, that this pattern currently triggers a warning message under
9f1b1f2d 1008the C<use warnings> pragma or B<-w> switch saying it
6bab786b 1009C<"matches null string many times in regex">.
c277df42 1010
c47ff5f1 1011On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
19799a22 1012effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
c277df42
IZ
1013This was only 4 times slower on a string with 1000000 C<a>s.
1014
9da458fc
IZ
1015The "grab all you can, and do not give anything back" semantic is desirable
1016in many situations where on the first sight a simple C<()*> looks like
1017the correct solution. Suppose we parse text with comments being delimited
1018by C<#> followed by some optional (horizontal) whitespace. Contrary to
4375e838 1019its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
9da458fc
IZ
1020the comment delimiter, because it may "give up" some whitespace if
1021the remainder of the pattern can be made to match that way. The correct
1022answer is either one of these:
1023
1024 (?>#[ \t]*)
1025 #[ \t]*(?![ \t])
1026
1027For example, to grab non-empty comments into $1, one should use either
1028one of these:
1029
1030 / (?> \# [ \t]* ) ( .+ ) /x;
1031 / \# [ \t]* ( [^ \t] .* ) /x;
1032
1033Which one you pick depends on which of these expressions better reflects
1034the above specification of comments.
1035
6bda09f9
YO
1036In some literature this construct is called "atomic matching" or
1037"possessive matching".
1038
b9b4dddf
YO
1039Possessive quantifiers are equivalent to putting the item they are applied
1040to inside of one of these constructs. The following equivalences apply:
1041
1042 Quantifier Form Bracketing Form
1043 --------------- ---------------
1044 PAT*+ (?>PAT*)
1045 PAT++ (?>PAT+)
1046 PAT?+ (?>PAT?)
1047 PAT{min,max}+ (?>PAT{min,max})
1048
54612592
YO
1049=item C<(?COMMIT)>
1050X<(?COMMIT)>
1051
1052This zero-width pattern commits the match at the current point, preventing
1053the engine from back-tracking on failure to the left of the commit point.
1054Consider the pattern C<A (?COMMIT) B>, where A and B are complex patterns.
1055Until the C<(?COMMIT)> is reached, A may backtrack as necessary to match.
1056Once it is reached, matching continues in B, which may also backtrack as
1057necessary; however, should B not match, then no further backtracking will
1058take place, and the pattern will fail outright at that starting position.
1059
1060The following example counts all the possible matching strings in a
1061pattern (without actually matching any of them).
1062
1063 'aaab'=~/a+b?(?{print "$&\n"; $count++})(?FAIL)/;
1064 print "Count=$count\n";
1065
1066which produces:
1067
1068 aaab
1069 aaa
1070 aa
1071 a
1072 aab
1073 aa
1074 a
1075 ab
1076 a
1077 Count=9
1078
1079If we add a C<(?COMMIT)> before the count like the following
1080
1081 'aaab'=~/a+b?(?COMMIT)(?{print "$&\n"; $count++})(?FAIL)/;
1082 print "Count=$count\n";
1083
1084we prevent backtracking and find the count of the longest matching
1085at each matching startpoint like so:
1086
1087 aaab
1088 aab
1089 ab
1090 Count=3
1091
1092Any number of C<(?COMMIT)> assertions may be used in a pattern.
1093
1094See also C<< (?>pattern) >> and possessive quantifiers for other
1095ways to control backtracking.
1096
24b23f37
YO
1097=item C<(?CUT)>
1098X<(?CUT)>
1099
1100This zero-width pattern is similar to C<(?COMMIT)>, except that on
1101failure it also signifies that whatever text that was matched leading
1102up to the C<(?CUT)> pattern cannot match, I<even from another
1103starting point>.
1104
1105Compare the following to the examples in C<(?COMMIT)>, note the string
1106is twice as long:
1107
1108 'aaabaaab'=~/a+b?(?CUT)(?{print "$&\n"; $count++})(?FAIL)/;
1109 print "Count=$count\n";
1110
1111outputs
1112
1113 aaab
1114 aaab
1115 Count=2
1116
1117Once the 'aaab' at the start of the string has matched and the C<(?CUT)>
1118executed the next startpoint will be where the cursor was when the
1119C<(?CUT)> was executed.
1120
1121=item C<(?ERROR)>
1122X<(?ERROR)>
1123
1124This zero-width pattern is similar to C<(?CUT)> except that it causes
1125the match to fail outright. No attempts to match will occur again.
1126
1127 'aaabaaab'=~/a+b?(?ERROR)(?{print "$&\n"; $count++})(?FAIL)/;
1128 print "Count=$count\n";
1129
1130outputs
1131
1132 aaab
1133 Count=1
1134
1135In other words, once the C<(?ERROR)> has been entered and then pattern
1136does not match then the regex engine will not try any further matching at
1137all on the rest of the string.
1138
5a964f20 1139=item C<(?(condition)yes-pattern|no-pattern)>
d74e8afc 1140X<(?()>
c277df42 1141
5a964f20 1142=item C<(?(condition)yes-pattern)>
c277df42
IZ
1143
1144Conditional expression. C<(condition)> should be either an integer in
1145parentheses (which is valid if the corresponding pair of parentheses
0a4db386
YO
1146matched), a look-ahead/look-behind/evaluate zero-width assertion, a
1147name in angle brackets or single quotes (which is valid if a buffer
1148with the given name matched), the special symbol (R) (true when
1149evaluated inside of recursion or eval). Additionally the R may be
1150followed by a number, (which will be true when evaluated when recursing
1151inside of the appropriate group), or by C<&NAME> in which case it will
9af228c6
YO
1152be true only when evaluated during recursion in the named group.
1153
1154Here's a summary of the possible predicates:
1155
1156=over 4
1157
1158=item (1) (2) ...
1159
1160Checks if the numbered capturing buffer has matched something.
1161
1162=item (<NAME>) ('NAME')
1163
1164Checks if a buffer with the given name has matched something.
1165
1166=item (?{ CODE })
1167
1168Treats the code block as the condition
1169
1170=item (R)
1171
1172Checks if the expression has been evaluated inside of recursion.
1173
1174=item (R1) (R2) ...
1175
1176Checks if the expression has been evaluated while executing directly
1177inside of the n-th capture group. This check is the regex equivalent of
1178
1179 if ((caller(0))[3] eq 'subname') { .. }
1180
1181In other words, it does not check the full recursion stack.
1182
1183=item (R&NAME)
1184
1185Similar to C<(R1)>, this predicate checks to see if we're executing
1186directly inside of the leftmost group with a given name (this is the same
1187logic used by C<(?&NAME)> to disambiguate). It does not check the full
1188stack, but only the name of the innermost active recursion.
1189
1190=item (DEFINE)
1191
1192In this case, the yes-pattern is never directly executed, and no
1193no-pattern is allowed. Similar in spirit to C<(?{0})> but more efficient.
1194See below for details.
1195
1196=back
c277df42 1197
19799a22 1198For example:
c277df42 1199
0a4db386
YO
1200 m{ ( \( )?
1201 [^()]+
1202 (?(1) \) )
871b0233 1203 }x
c277df42
IZ
1204
1205matches a chunk of non-parentheses, possibly included in parentheses
1206themselves.
a0d0e21e 1207
9af228c6
YO
1208A special form is the C<(DEFINE)> predicate, which never executes directly
1209its yes-pattern, and does not allow a no-pattern. This allows to define
1210subpatterns which will be executed only by using the recursion mechanism.
1211This way, you can define a set of regular expression rules that can be
1212bundled into any pattern you choose.
1213
1214It is recommended that for this usage you put the DEFINE block at the
1215end of the pattern, and that you name any subpatterns defined within it.
1216
1217Also, it's worth noting that patterns defined this way probably will
1218not be as efficient, as the optimiser is not very clever about
1219handling them. YMMV.
1220
1221An example of how this might be used is as follows:
1222
1223 /(?<NAME>(&NAME_PAT))(?<ADDR>(&ADDRESS_PAT))
1224 (?(DEFINE)
1225 (<NAME_PAT>....)
1226 (<ADRESS_PAT>....)
1227 )/x
1228
1229Note that capture buffers matched inside of recursion are not accessible
1230after the recursion returns, so the extra layer of capturing buffers are
1231necessary. Thus C<$+{NAME_PAT}> would not be defined even though
1232C<$+{NAME}> would be.
0a4db386 1233
a0d0e21e
LW
1234=back
1235
c07a80fd 1236=head2 Backtracking
d74e8afc 1237X<backtrack> X<backtracking>
c07a80fd 1238
35a734be
IZ
1239NOTE: This section presents an abstract approximation of regular
1240expression behavior. For a more rigorous (and complicated) view of
1241the rules involved in selecting a match among possible alternatives,
1242see L<Combining pieces together>.
1243
c277df42 1244A fundamental feature of regular expression matching involves the
5a964f20 1245notion called I<backtracking>, which is currently used (when needed)
c277df42 1246by all regular expression quantifiers, namely C<*>, C<*?>, C<+>,
9da458fc
IZ
1247C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
1248internally, but the general principle outlined here is valid.
c07a80fd
PP
1249
1250For a regular expression to match, the I<entire> regular expression must
1251match, not just part of it. So if the beginning of a pattern containing a
1252quantifier succeeds in a way that causes later parts in the pattern to
1253fail, the matching engine backs up and recalculates the beginning
1254part--that's why it's called backtracking.
1255
1256Here is an example of backtracking: Let's say you want to find the
1257word following "foo" in the string "Food is on the foo table.":
1258
1259 $_ = "Food is on the foo table.";
1260 if ( /\b(foo)\s+(\w+)/i ) {
1261 print "$2 follows $1.\n";
1262 }
1263
1264When the match runs, the first part of the regular expression (C<\b(foo)>)
1265finds a possible match right at the beginning of the string, and loads up
1266$1 with "Foo". However, as soon as the matching engine sees that there's
1267no whitespace following the "Foo" that it had saved in $1, it realizes its
68dc0745 1268mistake and starts over again one character after where it had the
c07a80fd
PP
1269tentative match. This time it goes all the way until the next occurrence
1270of "foo". The complete regular expression matches this time, and you get
1271the expected output of "table follows foo."
1272
1273Sometimes minimal matching can help a lot. Imagine you'd like to match
1274everything between "foo" and "bar". Initially, you write something
1275like this:
1276
1277 $_ = "The food is under the bar in the barn.";
1278 if ( /foo(.*)bar/ ) {
1279 print "got <$1>\n";
1280 }
1281
1282Which perhaps unexpectedly yields:
1283
1284 got <d is under the bar in the >
1285
1286That's because C<.*> was greedy, so you get everything between the
14218588 1287I<first> "foo" and the I<last> "bar". Here it's more effective
c07a80fd
PP
1288to use minimal matching to make sure you get the text between a "foo"
1289and the first "bar" thereafter.
1290
1291 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
1292 got <d is under the >
1293
1294Here's another example: let's say you'd like to match a number at the end
b6e13d97 1295of a string, and you also want to keep the preceding part of the match.
c07a80fd
PP
1296So you write this:
1297
1298 $_ = "I have 2 numbers: 53147";
1299 if ( /(.*)(\d*)/ ) { # Wrong!
1300 print "Beginning is <$1>, number is <$2>.\n";
1301 }
1302
1303That won't work at all, because C<.*> was greedy and gobbled up the
1304whole string. As C<\d*> can match on an empty string the complete
1305regular expression matched successfully.
1306
8e1088bc 1307 Beginning is <I have 2 numbers: 53147>, number is <>.
c07a80fd
PP
1308
1309Here are some variants, most of which don't work:
1310
1311 $_ = "I have 2 numbers: 53147";
1312 @pats = qw{
1313 (.*)(\d*)
1314 (.*)(\d+)
1315 (.*?)(\d*)
1316 (.*?)(\d+)
1317 (.*)(\d+)$
1318 (.*?)(\d+)$
1319 (.*)\b(\d+)$
1320 (.*\D)(\d+)$
1321 };
1322
1323 for $pat (@pats) {
1324 printf "%-12s ", $pat;
1325 if ( /$pat/ ) {
1326 print "<$1> <$2>\n";
1327 } else {
1328 print "FAIL\n";
1329 }
1330 }
1331
1332That will print out:
1333
1334 (.*)(\d*) <I have 2 numbers: 53147> <>
1335 (.*)(\d+) <I have 2 numbers: 5314> <7>
1336 (.*?)(\d*) <> <>
1337 (.*?)(\d+) <I have > <2>
1338 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
1339 (.*?)(\d+)$ <I have 2 numbers: > <53147>
1340 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
1341 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
1342
1343As you see, this can be a bit tricky. It's important to realize that a
1344regular expression is merely a set of assertions that gives a definition
1345of success. There may be 0, 1, or several different ways that the
1346definition might succeed against a particular string. And if there are
5a964f20
TC
1347multiple ways it might succeed, you need to understand backtracking to
1348know which variety of success you will achieve.
c07a80fd 1349
19799a22 1350When using look-ahead assertions and negations, this can all get even
8b19b778 1351trickier. Imagine you'd like to find a sequence of non-digits not
c07a80fd
PP
1352followed by "123". You might try to write that as
1353
871b0233
IZ
1354 $_ = "ABC123";
1355 if ( /^\D*(?!123)/ ) { # Wrong!
1356 print "Yup, no 123 in $_\n";
1357 }
c07a80fd
PP
1358
1359But that isn't going to match; at least, not the way you're hoping. It
1360claims that there is no 123 in the string. Here's a clearer picture of
9b9391b2 1361why that pattern matches, contrary to popular expectations:
c07a80fd 1362
4358a253
SS
1363 $x = 'ABC123';
1364 $y = 'ABC445';
c07a80fd 1365
4358a253
SS
1366 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/;
1367 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/;
c07a80fd 1368
4358a253
SS
1369 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/;
1370 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/;
c07a80fd
PP
1371
1372This prints
1373
1374 2: got ABC
1375 3: got AB
1376 4: got ABC
1377
5f05dabc 1378You might have expected test 3 to fail because it seems to a more
c07a80fd
PP
1379general purpose version of test 1. The important difference between
1380them is that test 3 contains a quantifier (C<\D*>) and so can use
1381backtracking, whereas test 1 will not. What's happening is
1382that you've asked "Is it true that at the start of $x, following 0 or more
5f05dabc 1383non-digits, you have something that's not 123?" If the pattern matcher had
c07a80fd 1384let C<\D*> expand to "ABC", this would have caused the whole pattern to
54310121 1385fail.
14218588 1386
c07a80fd 1387The search engine will initially match C<\D*> with "ABC". Then it will
14218588 1388try to match C<(?!123> with "123", which fails. But because
c07a80fd
PP
1389a quantifier (C<\D*>) has been used in the regular expression, the
1390search engine can backtrack and retry the match differently
54310121 1391in the hope of matching the complete regular expression.
c07a80fd 1392
5a964f20
TC
1393The pattern really, I<really> wants to succeed, so it uses the
1394standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
c07a80fd 1395time. Now there's indeed something following "AB" that is not
14218588 1396"123". It's "C123", which suffices.
c07a80fd 1397
14218588
GS
1398We can deal with this by using both an assertion and a negation.
1399We'll say that the first part in $1 must be followed both by a digit
1400and by something that's not "123". Remember that the look-aheads
1401are zero-width expressions--they only look, but don't consume any
1402of the string in their match. So rewriting this way produces what
c07a80fd
PP
1403you'd expect; that is, case 5 will fail, but case 6 succeeds:
1404
4358a253
SS
1405 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/;
1406 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/;
c07a80fd
PP
1407
1408 6: got ABC
1409
5a964f20 1410In other words, the two zero-width assertions next to each other work as though
19799a22 1411they're ANDed together, just as you'd use any built-in assertions: C</^$/>
c07a80fd
PP
1412matches only if you're at the beginning of the line AND the end of the
1413line simultaneously. The deeper underlying truth is that juxtaposition in
1414regular expressions always means AND, except when you write an explicit OR
1415using the vertical bar. C</ab/> means match "a" AND (then) match "b",
1416although the attempted matches are made at different positions because "a"
1417is not a zero-width assertion, but a one-width assertion.
1418
19799a22 1419B<WARNING>: particularly complicated regular expressions can take
14218588 1420exponential time to solve because of the immense number of possible
9da458fc
IZ
1421ways they can use backtracking to try match. For example, without
1422internal optimizations done by the regular expression engine, this will
1423take a painfully long time to run:
c07a80fd 1424
e1901655
IZ
1425 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
1426
1427And if you used C<*>'s in the internal groups instead of limiting them
1428to 0 through 5 matches, then it would take forever--or until you ran
1429out of stack space. Moreover, these internal optimizations are not
1430always applicable. For example, if you put C<{0,5}> instead of C<*>
1431on the external group, no current optimization is applicable, and the
1432match takes a long time to finish.
c07a80fd 1433
9da458fc
IZ
1434A powerful tool for optimizing such beasts is what is known as an
1435"independent group",
c47ff5f1 1436which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
9da458fc 1437zero-length look-ahead/look-behind assertions will not backtrack to make
14218588
GS
1438the tail match, since they are in "logical" context: only
1439whether they match is considered relevant. For an example
9da458fc 1440where side-effects of look-ahead I<might> have influenced the
c47ff5f1 1441following match, see L<C<< (?>pattern) >>>.
c277df42 1442
a0d0e21e 1443=head2 Version 8 Regular Expressions
d74e8afc 1444X<regular expression, version 8> X<regex, version 8> X<regexp, version 8>
a0d0e21e 1445
5a964f20 1446In case you're not familiar with the "regular" Version 8 regex
a0d0e21e
LW
1447routines, here are the pattern-matching rules not described above.
1448
54310121 1449Any single character matches itself, unless it is a I<metacharacter>
a0d0e21e 1450with a special meaning described here or above. You can cause
5a964f20 1451characters that normally function as metacharacters to be interpreted
5f05dabc 1452literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
a0d0e21e
LW
1453character; "\\" matches a "\"). A series of characters matches that
1454series of characters in the target string, so the pattern C<blurfl>
1455would match "blurfl" in the target string.
1456
1457You can specify a character class, by enclosing a list of characters
5a964f20 1458in C<[]>, which will match any one character from the list. If the
a0d0e21e 1459first character after the "[" is "^", the class matches any character not
14218588 1460in the list. Within a list, the "-" character specifies a
5a964f20 1461range, so that C<a-z> represents all characters between "a" and "z",
8a4f6ac2
GS
1462inclusive. If you want either "-" or "]" itself to be a member of a
1463class, put it at the start of the list (possibly after a "^"), or
1464escape it with a backslash. "-" is also taken literally when it is
1465at the end of the list, just before the closing "]". (The
84850974
DD
1466following all specify the same class of three characters: C<[-az]>,
1467C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
fb55449c
JH
1468specifies a class containing twenty-six characters, even on EBCDIC
1469based coded character sets.) Also, if you try to use the character
1470classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
1471a range, that's not a range, the "-" is understood literally.
a0d0e21e 1472
8ada0baa
JH
1473Note also that the whole range idea is rather unportable between
1474character sets--and even within character sets they may cause results
1475you probably didn't expect. A sound principle is to use only ranges
1476that begin from and end at either alphabets of equal case ([a-e],
1477[A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
1478spell out the character sets in full.
1479
54310121 1480Characters may be specified using a metacharacter syntax much like that
a0d0e21e
LW
1481used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
1482"\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
fb55449c
JH
1483of octal digits, matches the character whose coded character set value
1484is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
1485matches the character whose numeric value is I<nn>. The expression \cI<x>
1486matches the character control-I<x>. Finally, the "." metacharacter
1487matches any character except "\n" (unless you use C</s>).
a0d0e21e
LW
1488
1489You can specify a series of alternatives for a pattern using "|" to
1490separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
5a964f20 1491or "foe" in the target string (as would C<f(e|i|o)e>). The
a0d0e21e
LW
1492first alternative includes everything from the last pattern delimiter
1493("(", "[", or the beginning of the pattern) up to the first "|", and
1494the last alternative contains everything from the last "|" to the next
14218588
GS
1495pattern delimiter. That's why it's common practice to include
1496alternatives in parentheses: to minimize confusion about where they
a3cb178b
GS
1497start and end.
1498
5a964f20 1499Alternatives are tried from left to right, so the first
a3cb178b
GS
1500alternative found for which the entire expression matches, is the one that
1501is chosen. This means that alternatives are not necessarily greedy. For
628afcb5 1502example: when matching C<foo|foot> against "barefoot", only the "foo"
a3cb178b
GS
1503part will match, as that is the first alternative tried, and it successfully
1504matches the target string. (This might not seem important, but it is
1505important when you are capturing matched text using parentheses.)
1506
5a964f20 1507Also remember that "|" is interpreted as a literal within square brackets,
a3cb178b 1508so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
a0d0e21e 1509
14218588
GS
1510Within a pattern, you may designate subpatterns for later reference
1511by enclosing them in parentheses, and you may refer back to the
1512I<n>th subpattern later in the pattern using the metacharacter
1513\I<n>. Subpatterns are numbered based on the left to right order
1514of their opening parenthesis. A backreference matches whatever
1515actually matched the subpattern in the string being examined, not
1516the rules for that subpattern. Therefore, C<(0|0x)\d*\s\1\d*> will
1517match "0x1234 0x4321", but not "0x1234 01234", because subpattern
15181 matched "0x", even though the rule C<0|0x> could potentially match
1519the leading 0 in the second number.
cb1a09d0 1520
19799a22 1521=head2 Warning on \1 vs $1
cb1a09d0 1522
5a964f20 1523Some people get too used to writing things like:
cb1a09d0
AD
1524
1525 $pattern =~ s/(\W)/\\\1/g;
1526
1527This is grandfathered for the RHS of a substitute to avoid shocking the
1528B<sed> addicts, but it's a dirty habit to get into. That's because in
d1be9408 1529PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
cb1a09d0
AD
1530the usual double-quoted string means a control-A. The customary Unix
1531meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1532of doing that, you get yourself into trouble if you then add an C</e>
1533modifier.
1534
5a964f20 1535 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
cb1a09d0
AD
1536
1537Or if you try to do
1538
1539 s/(\d+)/\1000/;
1540
1541You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
14218588 1542C<${1}000>. The operation of interpolation should not be confused
cb1a09d0
AD
1543with the operation of matching a backreference. Certainly they mean two
1544different things on the I<left> side of the C<s///>.
9fa51da4 1545
c84d73f1
IZ
1546=head2 Repeated patterns matching zero-length substring
1547
19799a22 1548B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
c84d73f1
IZ
1549
1550Regular expressions provide a terse and powerful programming language. As
1551with most other power tools, power comes together with the ability
1552to wreak havoc.
1553
1554A common abuse of this power stems from the ability to make infinite
628afcb5 1555loops using regular expressions, with something as innocuous as:
c84d73f1
IZ
1556
1557 'foo' =~ m{ ( o? )* }x;
1558
1559The C<o?> can match at the beginning of C<'foo'>, and since the position
1560in the string is not moved by the match, C<o?> would match again and again
14218588 1561because of the C<*> modifier. Another common way to create a similar cycle
c84d73f1
IZ
1562is with the looping modifier C<//g>:
1563
1564 @matches = ( 'foo' =~ m{ o? }xg );
1565
1566or
1567
1568 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1569
1570or the loop implied by split().
1571
1572However, long experience has shown that many programming tasks may
14218588
GS
1573be significantly simplified by using repeated subexpressions that
1574may match zero-length substrings. Here's a simple example being:
c84d73f1
IZ
1575
1576 @chars = split //, $string; # // is not magic in split
1577 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1578
9da458fc 1579Thus Perl allows such constructs, by I<forcefully breaking
c84d73f1
IZ
1580the infinite loop>. The rules for this are different for lower-level
1581loops given by the greedy modifiers C<*+{}>, and for higher-level
1582ones like the C</g> modifier or split() operator.
1583
19799a22
GS
1584The lower-level loops are I<interrupted> (that is, the loop is
1585broken) when Perl detects that a repeated expression matched a
1586zero-length substring. Thus
c84d73f1
IZ
1587
1588 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1589
1590is made equivalent to
1591
1592 m{ (?: NON_ZERO_LENGTH )*
1593 |
1594 (?: ZERO_LENGTH )?
1595 }x;
1596
1597The higher level-loops preserve an additional state between iterations:
1598whether the last match was zero-length. To break the loop, the following
1599match after a zero-length match is prohibited to have a length of zero.
1600This prohibition interacts with backtracking (see L<"Backtracking">),
1601and so the I<second best> match is chosen if the I<best> match is of
1602zero length.
1603
19799a22 1604For example:
c84d73f1
IZ
1605
1606 $_ = 'bar';
1607 s/\w??/<$&>/g;
1608
20fb949f 1609results in C<< <><b><><a><><r><> >>. At each position of the string the best
c84d73f1
IZ
1610match given by non-greedy C<??> is the zero-length match, and the I<second
1611best> match is what is matched by C<\w>. Thus zero-length matches
1612alternate with one-character-long matches.
1613
1614Similarly, for repeated C<m/()/g> the second-best match is the match at the
1615position one notch further in the string.
1616
19799a22 1617The additional state of being I<matched with zero-length> is associated with
c84d73f1 1618the matched string, and is reset by each assignment to pos().
9da458fc
IZ
1619Zero-length matches at the end of the previous match are ignored
1620during C<split>.
c84d73f1 1621
35a734be
IZ
1622=head2 Combining pieces together
1623
1624Each of the elementary pieces of regular expressions which were described
1625before (such as C<ab> or C<\Z>) could match at most one substring
1626at the given position of the input string. However, in a typical regular
1627expression these elementary pieces are combined into more complicated
1628patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1629(in these examples C<S> and C<T> are regular subexpressions).
1630
1631Such combinations can include alternatives, leading to a problem of choice:
1632if we match a regular expression C<a|ab> against C<"abc">, will it match
1633substring C<"a"> or C<"ab">? One way to describe which substring is
1634actually matched is the concept of backtracking (see L<"Backtracking">).
1635However, this description is too low-level and makes you think
1636in terms of a particular implementation.
1637
1638Another description starts with notions of "better"/"worse". All the
1639substrings which may be matched by the given regular expression can be
1640sorted from the "best" match to the "worst" match, and it is the "best"
1641match which is chosen. This substitutes the question of "what is chosen?"
1642by the question of "which matches are better, and which are worse?".
1643
1644Again, for elementary pieces there is no such question, since at most
1645one match at a given position is possible. This section describes the
1646notion of better/worse for combining operators. In the description
1647below C<S> and C<T> are regular subexpressions.
1648
13a2d996 1649=over 4
35a734be
IZ
1650
1651=item C<ST>
1652
1653Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
1654substrings which can be matched by C<S>, C<B> and C<B'> are substrings
1655which can be matched by C<T>.
1656
1657If C<A> is better match for C<S> than C<A'>, C<AB> is a better
1658match than C<A'B'>.
1659
1660If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
1661C<B> is better match for C<T> than C<B'>.
1662
1663=item C<S|T>
1664
1665When C<S> can match, it is a better match than when only C<T> can match.
1666
1667Ordering of two matches for C<S> is the same as for C<S>. Similar for
1668two matches for C<T>.
1669
1670=item C<S{REPEAT_COUNT}>
1671
1672Matches as C<SSS...S> (repeated as many times as necessary).
1673
1674=item C<S{min,max}>
1675
1676Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
1677
1678=item C<S{min,max}?>
1679
1680Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
1681
1682=item C<S?>, C<S*>, C<S+>
1683
1684Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
1685
1686=item C<S??>, C<S*?>, C<S+?>
1687
1688Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
1689
c47ff5f1 1690=item C<< (?>S) >>
35a734be
IZ
1691
1692Matches the best match for C<S> and only that.
1693
1694=item C<(?=S)>, C<(?<=S)>
1695
1696Only the best match for C<S> is considered. (This is important only if
1697C<S> has capturing parentheses, and backreferences are used somewhere
1698else in the whole regular expression.)
1699
1700=item C<(?!S)>, C<(?<!S)>
1701
1702For this grouping operator there is no need to describe the ordering, since
1703only whether or not C<S> can match is important.
1704
6bda09f9 1705=item C<(??{ EXPR })>, C<(?PARNO)>
35a734be
IZ
1706
1707The ordering is the same as for the regular expression which is
6bda09f9 1708the result of EXPR, or the pattern contained by capture buffer PARNO.
35a734be
IZ
1709
1710=item C<(?(condition)yes-pattern|no-pattern)>
1711
1712Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
1713already determined. The ordering of the matches is the same as for the
1714chosen subexpression.
1715
1716=back
1717
1718The above recipes describe the ordering of matches I<at a given position>.
1719One more rule is needed to understand how a match is determined for the
1720whole regular expression: a match at an earlier position is always better
1721than a match at a later position.
1722
c84d73f1
IZ
1723=head2 Creating custom RE engines
1724
1725Overloaded constants (see L<overload>) provide a simple way to extend
1726the functionality of the RE engine.
1727
1728Suppose that we want to enable a new RE escape-sequence C<\Y|> which
6b0ac556 1729matches at boundary between whitespace characters and non-whitespace
c84d73f1
IZ
1730characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
1731at these positions, so we want to have each C<\Y|> in the place of the
1732more complicated version. We can create a module C<customre> to do
1733this:
1734
1735 package customre;
1736 use overload;
1737
1738 sub import {
1739 shift;
1740 die "No argument to customre::import allowed" if @_;
1741 overload::constant 'qr' => \&convert;
1742 }
1743
1744 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
1745
580a9fe1
RGS
1746 # We must also take care of not escaping the legitimate \\Y|
1747 # sequence, hence the presence of '\\' in the conversion rules.
141db969 1748 my %rules = ( '\\' => '\\\\',
c84d73f1
IZ
1749 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
1750 sub convert {
1751 my $re = shift;
1752 $re =~ s{
1753 \\ ( \\ | Y . )
1754 }
1755 { $rules{$1} or invalid($re,$1) }sgex;
1756 return $re;
1757 }
1758
1759Now C<use customre> enables the new escape in constant regular
1760expressions, i.e., those without any runtime variable interpolations.
1761As documented in L<overload>, this conversion will work only over
1762literal parts of regular expressions. For C<\Y|$re\Y|> the variable
1763part of this regular expression needs to be converted explicitly
1764(but only if the special meaning of C<\Y|> should be enabled inside $re):
1765
1766 use customre;
1767 $re = <>;
1768 chomp $re;
1769 $re = customre::convert $re;
1770 /\Y|$re\Y|/;
1771
19799a22
GS
1772=head1 BUGS
1773
9da458fc
IZ
1774This document varies from difficult to understand to completely
1775and utterly opaque. The wandering prose riddled with jargon is
1776hard to fathom in several places.
1777
1778This document needs a rewrite that separates the tutorial content
1779from the reference content.
19799a22
GS
1780
1781=head1 SEE ALSO
9fa51da4 1782
91e0c79e
MJD
1783L<perlrequick>.
1784
1785L<perlretut>.
1786
9b599b2a
GS
1787L<perlop/"Regexp Quote-Like Operators">.
1788
1e66bd83
PP
1789L<perlop/"Gory details of parsing quoted constructs">.
1790
14218588
GS
1791L<perlfaq6>.
1792
9b599b2a
GS
1793L<perlfunc/pos>.
1794
1795L<perllocale>.
1796
fb55449c
JH
1797L<perlebcdic>.
1798
14218588
GS
1799I<Mastering Regular Expressions> by Jeffrey Friedl, published
1800by O'Reilly and Associates.