3 perlre - Perl regular expressions
7 This page describes the syntax of regular expressions in Perl.
9 if you haven't used regular expressions before, a quick-start
10 introduction is available in L<perlrequick>, and a longer tutorial
11 introduction is available in L<perlretut>.
13 For reference on how regular expressions are used in matching
14 operations, plus various examples of the same, see discussions of
15 C<m//>, C<s///>, C<qr//> and C<??> in L<perlop/"Regexp Quote-Like
18 Matching operations can have various modifiers. Modifiers
19 that relate to the interpretation of the regular expression inside
20 are listed below. Modifiers that alter the way a regular expression
21 is used by Perl are detailed in L<perlop/"Regexp Quote-Like Operators"> and
22 L<perlop/"Gory details of parsing quoted constructs">.
28 Do case-insensitive pattern matching.
30 If C<use locale> is in effect, the case map is taken from the current
31 locale. See L<perllocale>.
35 Treat string as multiple lines. That is, change "^" and "$" from matching
36 the start or end of the string to matching the start or end of any
37 line anywhere within the string.
41 Treat string as single line. That is, change "." to match any character
42 whatsoever, even a newline, which normally it would not match.
44 The C</s> and C</m> modifiers both override the C<$*> setting. That
45 is, no matter what C<$*> contains, C</s> without C</m> will force
46 "^" to match only at the beginning of the string and "$" to match
47 only at the end (or just before a newline at the end) of the string.
48 Together, as /ms, they let the "." match any character whatsoever,
49 while still allowing "^" and "$" to match, respectively, just after
50 and just before newlines within the string.
54 Extend your pattern's legibility by permitting whitespace and comments.
58 These are usually written as "the C</x> modifier", even though the delimiter
59 in question might not really be a slash. Any of these
60 modifiers may also be embedded within the regular expression itself using
61 the C<(?...)> construct. See below.
63 The C</x> modifier itself needs a little more explanation. It tells
64 the regular expression parser to ignore whitespace that is neither
65 backslashed nor within a character class. You can use this to break up
66 your regular expression into (slightly) more readable parts. The C<#>
67 character is also treated as a metacharacter introducing a comment,
68 just as in ordinary Perl code. This also means that if you want real
69 whitespace or C<#> characters in the pattern (outside a character
70 class, where they are unaffected by C</x>), that you'll either have to
71 escape them or encode them using octal or hex escapes. Taken together,
72 these features go a long way towards making Perl's regular expressions
73 more readable. Note that you have to be careful not to include the
74 pattern delimiter in the comment--perl has no way of knowing you did
75 not intend to close the pattern early. See the C-comment deletion code
78 =head2 Regular Expressions
80 The patterns used in Perl pattern matching derive from supplied in
81 the Version 8 regex routines. (The routines are derived
82 (distantly) from Henry Spencer's freely redistributable reimplementation
83 of the V8 routines.) See L<Version 8 Regular Expressions> for
86 In particular the following metacharacters have their standard I<egrep>-ish
89 \ Quote the next metacharacter
90 ^ Match the beginning of the line
91 . Match any character (except newline)
92 $ Match the end of the line (or before newline at the end)
97 By default, the "^" character is guaranteed to match only the
98 beginning of the string, the "$" character only the end (or before the
99 newline at the end), and Perl does certain optimizations with the
100 assumption that the string contains only one line. Embedded newlines
101 will not be matched by "^" or "$". You may, however, wish to treat a
102 string as a multi-line buffer, such that the "^" will match after any
103 newline within the string, and "$" will match before any newline. At the
104 cost of a little more overhead, you can do this by using the /m modifier
105 on the pattern match operator. (Older programs did this by setting C<$*>,
106 but this practice is now deprecated.)
108 To simplify multi-line substitutions, the "." character never matches a
109 newline unless you use the C</s> modifier, which in effect tells Perl to pretend
110 the string is a single line--even if it isn't. The C</s> modifier also
111 overrides the setting of C<$*>, in case you have some (badly behaved) older
112 code that sets it in another module.
114 The following standard quantifiers are recognized:
116 * Match 0 or more times
117 + Match 1 or more times
119 {n} Match exactly n times
120 {n,} Match at least n times
121 {n,m} Match at least n but not more than m times
123 (If a curly bracket occurs in any other context, it is treated
124 as a regular character.) The "*" modifier is equivalent to C<{0,}>, the "+"
125 modifier to C<{1,}>, and the "?" modifier to C<{0,1}>. n and m are limited
126 to integral values less than a preset limit defined when perl is built.
127 This is usually 32766 on the most common platforms. The actual limit can
128 be seen in the error message generated by code such as this:
130 $_ **= $_ , / {$_} / for 2 .. 42;
132 By default, a quantified subpattern is "greedy", that is, it will match as
133 many times as possible (given a particular starting location) while still
134 allowing the rest of the pattern to match. If you want it to match the
135 minimum number of times possible, follow the quantifier with a "?". Note
136 that the meanings don't change, just the "greediness":
138 *? Match 0 or more times
139 +? Match 1 or more times
141 {n}? Match exactly n times
142 {n,}? Match at least n times
143 {n,m}? Match at least n but not more than m times
145 Because patterns are processed as double quoted strings, the following
152 \a alarm (bell) (BEL)
153 \e escape (think troff) (ESC)
154 \033 octal char (think of a PDP-11)
156 \x{263a} wide hex char (Unicode SMILEY)
159 \l lowercase next char (think vi)
160 \u uppercase next char (think vi)
161 \L lowercase till \E (think vi)
162 \U uppercase till \E (think vi)
163 \E end case modification (think vi)
164 \Q quote (disable) pattern metacharacters till \E
166 If C<use locale> is in effect, the case map used by C<\l>, C<\L>, C<\u>
167 and C<\U> is taken from the current locale. See L<perllocale>. For
168 documentation of C<\N{name}>, see L<charnames>.
170 You cannot include a literal C<$> or C<@> within a C<\Q> sequence.
171 An unescaped C<$> or C<@> interpolates the corresponding variable,
172 while escaping will cause the literal string C<\$> to be matched.
173 You'll need to write something like C<m/\Quser\E\@\Qhost/>.
175 In addition, Perl defines the following:
177 \w Match a "word" character (alphanumeric plus "_")
178 \W Match a non-"word" character
179 \s Match a whitespace character
180 \S Match a non-whitespace character
181 \d Match a digit character
182 \D Match a non-digit character
183 \pP Match P, named property. Use \p{Prop} for longer names.
185 \X Match eXtended Unicode "combining character sequence",
186 equivalent to (?:\PM\pM*)
187 \C Match a single C char (octet) even under Unicode.
188 NOTE: breaks up characters into their UTF-8 bytes,
189 so you may end up with malformed pieces of UTF-8.
191 A C<\w> matches a single alphanumeric character or C<_>, not a whole word.
192 Use C<\w+> to match a string of Perl-identifier characters (which isn't
193 the same as matching an English word). If C<use locale> is in effect, the
194 list of alphabetic characters generated by C<\w> is taken from the
195 current locale. See L<perllocale>. You may use C<\w>, C<\W>, C<\s>, C<\S>,
196 C<\d>, and C<\D> within character classes, but if you try to use them
197 as endpoints of a range, that's not a range, the "-" is understood literally.
198 See L<perlunicode> for details about C<\pP>, C<\PP>, and C<\X>.
200 The POSIX character class syntax
204 is also available. The available classes and their backslash
205 equivalents (if available) are as follows:
226 A GNU extension equivalent to C<[ \t]>, `all horizontal whitespace'.
230 Not exactly equivalent to C<\s> since the C<[[:space:]]> includes
231 also the (very rare) `vertical tabulator', \ck", chr(11).
239 For example use C<[:upper:]> to match all the uppercase characters.
240 Note that the C<[]> are part of the C<[::]> construct, not part of the
241 whole character class. For example:
245 matches zero, one, any alphabetic character, and the percentage sign.
247 The following equivalences to Unicode \p{} constructs and equivalent
248 backslash character classes (if available), will hold:
250 [:...:] \p{...} backslash
268 For example C<[:lower:]> and C<\p{IsLower}> are equivalent.
270 If the C<utf8> pragma is not used but the C<locale> pragma is, the
271 classes correlate with the usual isalpha(3) interface (except for
274 The assumedly non-obviously named classes are:
280 Any control character. Usually characters that don't produce output as
281 such but instead control the terminal somehow: for example newline and
282 backspace are control characters. All characters with ord() less than
283 32 are most often classified as control characters (assuming ASCII,
284 the ISO Latin character sets, and Unicode), as is the character with
285 the ord() value of 127 (C<DEL>).
289 Any alphanumeric or punctuation (special) character.
293 Any alphanumeric or punctuation (special) character or the space character.
297 Any punctuation (special) character.
301 Any hexadecimal digit. Though this may feel silly ([0-9A-Fa-f] would
302 work just fine) it is included for completeness.
306 You can negate the [::] character classes by prefixing the class name
307 with a '^'. This is a Perl extension. For example:
309 POSIX traditional Unicode
311 [:^digit:] \D \P{IsDigit}
312 [:^space:] \S \P{IsSpace}
313 [:^word:] \W \P{IsWord}
315 The POSIX character classes [.cc.] and [=cc=] are recognized but
316 B<not> supported and trying to use them will cause an error.
318 Perl defines the following zero-width assertions:
320 \b Match a word boundary
321 \B Match a non-(word boundary)
322 \A Match only at beginning of string
323 \Z Match only at end of string, or before newline at the end
324 \z Match only at end of string
325 \G Match only at pos() (e.g. at the end-of-match position
328 A word boundary (C<\b>) is a spot between two characters
329 that has a C<\w> on one side of it and a C<\W> on the other side
330 of it (in either order), counting the imaginary characters off the
331 beginning and end of the string as matching a C<\W>. (Within
332 character classes C<\b> represents backspace rather than a word
333 boundary, just as it normally does in any double-quoted string.)
334 The C<\A> and C<\Z> are just like "^" and "$", except that they
335 won't match multiple times when the C</m> modifier is used, while
336 "^" and "$" will match at every internal line boundary. To match
337 the actual end of the string and not ignore an optional trailing
340 The C<\G> assertion can be used to chain global matches (using
341 C<m//g>), as described in L<perlop/"Regexp Quote-Like Operators">.
342 It is also useful when writing C<lex>-like scanners, when you have
343 several patterns that you want to match against consequent substrings
344 of your string, see the previous reference. The actual location
345 where C<\G> will match can also be influenced by using C<pos()> as
346 an lvalue. See L<perlfunc/pos>.
348 The bracketing construct C<( ... )> creates capture buffers. To
349 refer to the digit'th buffer use \<digit> within the
350 match. Outside the match use "$" instead of "\". (The
351 \<digit> notation works in certain circumstances outside
352 the match. See the warning below about \1 vs $1 for details.)
353 Referring back to another part of the match is called a
356 There is no limit to the number of captured substrings that you may
357 use. However Perl also uses \10, \11, etc. as aliases for \010,
358 \011, etc. (Recall that 0 means octal, so \011 is the character at
359 number 9 in your coded character set; which would be the 10th character,
360 a horizontal tab under ASCII.) Perl resolves this
361 ambiguity by interpreting \10 as a backreference only if at least 10
362 left parentheses have opened before it. Likewise \11 is a
363 backreference only if at least 11 left parentheses have opened
364 before it. And so on. \1 through \9 are always interpreted as
369 s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
371 if (/(.)\1/) { # find first doubled char
372 print "'$1' is the first doubled character\n";
375 if (/Time: (..):(..):(..)/) { # parse out values
381 Several special variables also refer back to portions of the previous
382 match. C<$+> returns whatever the last bracket match matched.
383 C<$&> returns the entire matched string. (At one point C<$0> did
384 also, but now it returns the name of the program.) C<$`> returns
385 everything before the matched string. And C<$'> returns everything
386 after the matched string.
388 The numbered variables ($1, $2, $3, etc.) and the related punctuation
389 set (C<$+>, C<$&>, C<$`>, and C<$'>) are all dynamically scoped
390 until the end of the enclosing block or until the next successful
391 match, whichever comes first. (See L<perlsyn/"Compound Statements">.)
393 B<WARNING>: Once Perl sees that you need one of C<$&>, C<$`>, or
394 C<$'> anywhere in the program, it has to provide them for every
395 pattern match. This may substantially slow your program. Perl
396 uses the same mechanism to produce $1, $2, etc, so you also pay a
397 price for each pattern that contains capturing parentheses. (To
398 avoid this cost while retaining the grouping behaviour, use the
399 extended regular expression C<(?: ... )> instead.) But if you never
400 use C<$&>, C<$`> or C<$'>, then patterns I<without> capturing
401 parentheses will not be penalized. So avoid C<$&>, C<$'>, and C<$`>
402 if you can, but if you can't (and some algorithms really appreciate
403 them), once you've used them once, use them at will, because you've
404 already paid the price. As of 5.005, C<$&> is not so costly as the
407 Backslashed metacharacters in Perl are alphanumeric, such as C<\b>,
408 C<\w>, C<\n>. Unlike some other regular expression languages, there
409 are no backslashed symbols that aren't alphanumeric. So anything
410 that looks like \\, \(, \), \<, \>, \{, or \} is always
411 interpreted as a literal character, not a metacharacter. This was
412 once used in a common idiom to disable or quote the special meanings
413 of regular expression metacharacters in a string that you want to
414 use for a pattern. Simply quote all non-"word" characters:
416 $pattern =~ s/(\W)/\\$1/g;
418 (If C<use locale> is set, then this depends on the current locale.)
419 Today it is more common to use the quotemeta() function or the C<\Q>
420 metaquoting escape sequence to disable all metacharacters' special
423 /$unquoted\Q$quoted\E$unquoted/
425 Beware that if you put literal backslashes (those not inside
426 interpolated variables) between C<\Q> and C<\E>, double-quotish
427 backslash interpolation may lead to confusing results. If you
428 I<need> to use literal backslashes within C<\Q...\E>,
429 consult L<perlop/"Gory details of parsing quoted constructs">.
431 =head2 Extended Patterns
433 Perl also defines a consistent extension syntax for features not
434 found in standard tools like B<awk> and B<lex>. The syntax is a
435 pair of parentheses with a question mark as the first thing within
436 the parentheses. The character after the question mark indicates
439 The stability of these extensions varies widely. Some have been
440 part of the core language for many years. Others are experimental
441 and may change without warning or be completely removed. Check
442 the documentation on an individual feature to verify its current
445 A question mark was chosen for this and for the minimal-matching
446 construct because 1) question marks are rare in older regular
447 expressions, and 2) whenever you see one, you should stop and
448 "question" exactly what is going on. That's psychology...
454 A comment. The text is ignored. If the C</x> modifier enables
455 whitespace formatting, a simple C<#> will suffice. Note that Perl closes
456 the comment as soon as it sees a C<)>, so there is no way to put a literal
459 =item C<(?imsx-imsx)>
461 One or more embedded pattern-match modifiers. This is particularly
462 useful for dynamic patterns, such as those read in from a configuration
463 file, read in as an argument, are specified in a table somewhere,
464 etc. Consider the case that some of which want to be case sensitive
465 and some do not. The case insensitive ones need to include merely
466 C<(?i)> at the front of the pattern. For example:
469 if ( /$pattern/i ) { }
473 $pattern = "(?i)foobar";
474 if ( /$pattern/ ) { }
476 Letters after a C<-> turn those modifiers off. These modifiers are
477 localized inside an enclosing group (if any). For example,
481 will match a repeated (I<including the case>!) word C<blah> in any
482 case, assuming C<x> modifier, and no C<i> modifier outside this
487 =item C<(?imsx-imsx:pattern)>
489 This is for clustering, not capturing; it groups subexpressions like
490 "()", but doesn't make backreferences as "()" does. So
492 @fields = split(/\b(?:a|b|c)\b/)
496 @fields = split(/\b(a|b|c)\b/)
498 but doesn't spit out extra fields. It's also cheaper not to capture
499 characters if you don't need to.
501 Any letters between C<?> and C<:> act as flags modifiers as with
502 C<(?imsx-imsx)>. For example,
504 /(?s-i:more.*than).*million/i
506 is equivalent to the more verbose
508 /(?:(?s-i)more.*than).*million/i
512 A zero-width positive look-ahead assertion. For example, C</\w+(?=\t)/>
513 matches a word followed by a tab, without including the tab in C<$&>.
517 A zero-width negative look-ahead assertion. For example C</foo(?!bar)/>
518 matches any occurrence of "foo" that isn't followed by "bar". Note
519 however that look-ahead and look-behind are NOT the same thing. You cannot
520 use this for look-behind.
522 If you are looking for a "bar" that isn't preceded by a "foo", C</(?!foo)bar/>
523 will not do what you want. That's because the C<(?!foo)> is just saying that
524 the next thing cannot be "foo"--and it's not, it's a "bar", so "foobar" will
525 match. You would have to do something like C</(?!foo)...bar/> for that. We
526 say "like" because there's the case of your "bar" not having three characters
527 before it. You could cover that this way: C</(?:(?!foo)...|^.{0,2})bar/>.
528 Sometimes it's still easier just to say:
530 if (/bar/ && $` !~ /foo$/)
532 For look-behind see below.
534 =item C<(?<=pattern)>
536 A zero-width positive look-behind assertion. For example, C</(?<=\t)\w+/>
537 matches a word that follows a tab, without including the tab in C<$&>.
538 Works only for fixed-width look-behind.
540 =item C<(?<!pattern)>
542 A zero-width negative look-behind assertion. For example C</(?<!bar)foo/>
543 matches any occurrence of "foo" that does not follow "bar". Works
544 only for fixed-width look-behind.
548 B<WARNING>: This extended regular expression feature is considered
549 highly experimental, and may be changed or deleted without notice.
551 This zero-width assertion evaluate any embedded Perl code. It
552 always succeeds, and its C<code> is not interpolated. Currently,
553 the rules to determine where the C<code> ends are somewhat convoluted.
555 The C<code> is properly scoped in the following sense: If the assertion
556 is backtracked (compare L<"Backtracking">), all changes introduced after
557 C<local>ization are undone, so that
561 (?{ $cnt = 0 }) # Initialize $cnt.
565 local $cnt = $cnt + 1; # Update $cnt, backtracking-safe.
569 (?{ $res = $cnt }) # On success copy to non-localized
573 will set C<$res = 4>. Note that after the match, $cnt returns to the globally
574 introduced value, because the scopes that restrict C<local> operators
577 This assertion may be used as a C<(?(condition)yes-pattern|no-pattern)>
578 switch. If I<not> used in this way, the result of evaluation of
579 C<code> is put into the special variable C<$^R>. This happens
580 immediately, so C<$^R> can be used from other C<(?{ code })> assertions
581 inside the same regular expression.
583 The assignment to C<$^R> above is properly localized, so the old
584 value of C<$^R> is restored if the assertion is backtracked; compare
587 For reasons of security, this construct is forbidden if the regular
588 expression involves run-time interpolation of variables, unless the
589 perilous C<use re 'eval'> pragma has been used (see L<re>), or the
590 variables contain results of C<qr//> operator (see
591 L<perlop/"qr/STRING/imosx">).
593 This restriction is because of the wide-spread and remarkably convenient
594 custom of using run-time determined strings as patterns. For example:
600 Before Perl knew how to execute interpolated code within a pattern,
601 this operation was completely safe from a security point of view,
602 although it could raise an exception from an illegal pattern. If
603 you turn on the C<use re 'eval'>, though, it is no longer secure,
604 so you should only do so if you are also using taint checking.
605 Better yet, use the carefully constrained evaluation within a Safe
606 module. See L<perlsec> for details about both these mechanisms.
608 =item C<(??{ code })>
610 B<WARNING>: This extended regular expression feature is considered
611 highly experimental, and may be changed or deleted without notice.
612 A simplified version of the syntax may be introduced for commonly
615 This is a "postponed" regular subexpression. The C<code> is evaluated
616 at run time, at the moment this subexpression may match. The result
617 of evaluation is considered as a regular expression and matched as
618 if it were inserted instead of this construct.
620 The C<code> is not interpolated. As before, the rules to determine
621 where the C<code> ends are currently somewhat convoluted.
623 The following pattern matches a parenthesized group:
628 (?> [^()]+ ) # Non-parens without backtracking
630 (??{ $re }) # Group with matching parens
635 =item C<< (?>pattern) >>
637 B<WARNING>: This extended regular expression feature is considered
638 highly experimental, and may be changed or deleted without notice.
640 An "independent" subexpression, one which matches the substring
641 that a I<standalone> C<pattern> would match if anchored at the given
642 position, and it matches I<nothing other than this substring>. This
643 construct is useful for optimizations of what would otherwise be
644 "eternal" matches, because it will not backtrack (see L<"Backtracking">).
645 It may also be useful in places where the "grab all you can, and do not
646 give anything back" semantic is desirable.
648 For example: C<< ^(?>a*)ab >> will never match, since C<< (?>a*) >>
649 (anchored at the beginning of string, as above) will match I<all>
650 characters C<a> at the beginning of string, leaving no C<a> for
651 C<ab> to match. In contrast, C<a*ab> will match the same as C<a+b>,
652 since the match of the subgroup C<a*> is influenced by the following
653 group C<ab> (see L<"Backtracking">). In particular, C<a*> inside
654 C<a*ab> will match fewer characters than a standalone C<a*>, since
655 this makes the tail match.
657 An effect similar to C<< (?>pattern) >> may be achieved by writing
658 C<(?=(pattern))\1>. This matches the same substring as a standalone
659 C<a+>, and the following C<\1> eats the matched string; it therefore
660 makes a zero-length assertion into an analogue of C<< (?>...) >>.
661 (The difference between these two constructs is that the second one
662 uses a capturing group, thus shifting ordinals of backreferences
663 in the rest of a regular expression.)
665 Consider this pattern:
676 That will efficiently match a nonempty group with matching parentheses
677 two levels deep or less. However, if there is no such group, it
678 will take virtually forever on a long string. That's because there
679 are so many different ways to split a long string into several
680 substrings. This is what C<(.+)+> is doing, and C<(.+)+> is similar
681 to a subpattern of the above pattern. Consider how the pattern
682 above detects no-match on C<((()aaaaaaaaaaaaaaaaaa> in several
683 seconds, but that each extra letter doubles this time. This
684 exponential performance will make it appear that your program has
685 hung. However, a tiny change to this pattern
689 (?> [^()]+ ) # change x+ above to (?> x+ )
696 which uses C<< (?>...) >> matches exactly when the one above does (verifying
697 this yourself would be a productive exercise), but finishes in a fourth
698 the time when used on a similar string with 1000000 C<a>s. Be aware,
699 however, that this pattern currently triggers a warning message under
700 the C<use warnings> pragma or B<-w> switch saying it
701 C<"matches null string many times in regex">.
703 On simple groups, such as the pattern C<< (?> [^()]+ ) >>, a comparable
704 effect may be achieved by negative look-ahead, as in C<[^()]+ (?! [^()] )>.
705 This was only 4 times slower on a string with 1000000 C<a>s.
707 The "grab all you can, and do not give anything back" semantic is desirable
708 in many situations where on the first sight a simple C<()*> looks like
709 the correct solution. Suppose we parse text with comments being delimited
710 by C<#> followed by some optional (horizontal) whitespace. Contrary to
711 its appearance, C<#[ \t]*> I<is not> the correct subexpression to match
712 the comment delimiter, because it may "give up" some whitespace if
713 the remainder of the pattern can be made to match that way. The correct
714 answer is either one of these:
719 For example, to grab non-empty comments into $1, one should use either
722 / (?> \# [ \t]* ) ( .+ ) /x;
723 / \# [ \t]* ( [^ \t] .* ) /x;
725 Which one you pick depends on which of these expressions better reflects
726 the above specification of comments.
728 =item C<(?(condition)yes-pattern|no-pattern)>
730 =item C<(?(condition)yes-pattern)>
732 B<WARNING>: This extended regular expression feature is considered
733 highly experimental, and may be changed or deleted without notice.
735 Conditional expression. C<(condition)> should be either an integer in
736 parentheses (which is valid if the corresponding pair of parentheses
737 matched), or look-ahead/look-behind/evaluate zero-width assertion.
746 matches a chunk of non-parentheses, possibly included in parentheses
753 NOTE: This section presents an abstract approximation of regular
754 expression behavior. For a more rigorous (and complicated) view of
755 the rules involved in selecting a match among possible alternatives,
756 see L<Combining pieces together>.
758 A fundamental feature of regular expression matching involves the
759 notion called I<backtracking>, which is currently used (when needed)
760 by all regular expression quantifiers, namely C<*>, C<*?>, C<+>,
761 C<+?>, C<{n,m}>, and C<{n,m}?>. Backtracking is often optimized
762 internally, but the general principle outlined here is valid.
764 For a regular expression to match, the I<entire> regular expression must
765 match, not just part of it. So if the beginning of a pattern containing a
766 quantifier succeeds in a way that causes later parts in the pattern to
767 fail, the matching engine backs up and recalculates the beginning
768 part--that's why it's called backtracking.
770 Here is an example of backtracking: Let's say you want to find the
771 word following "foo" in the string "Food is on the foo table.":
773 $_ = "Food is on the foo table.";
774 if ( /\b(foo)\s+(\w+)/i ) {
775 print "$2 follows $1.\n";
778 When the match runs, the first part of the regular expression (C<\b(foo)>)
779 finds a possible match right at the beginning of the string, and loads up
780 $1 with "Foo". However, as soon as the matching engine sees that there's
781 no whitespace following the "Foo" that it had saved in $1, it realizes its
782 mistake and starts over again one character after where it had the
783 tentative match. This time it goes all the way until the next occurrence
784 of "foo". The complete regular expression matches this time, and you get
785 the expected output of "table follows foo."
787 Sometimes minimal matching can help a lot. Imagine you'd like to match
788 everything between "foo" and "bar". Initially, you write something
791 $_ = "The food is under the bar in the barn.";
792 if ( /foo(.*)bar/ ) {
796 Which perhaps unexpectedly yields:
798 got <d is under the bar in the >
800 That's because C<.*> was greedy, so you get everything between the
801 I<first> "foo" and the I<last> "bar". Here it's more effective
802 to use minimal matching to make sure you get the text between a "foo"
803 and the first "bar" thereafter.
805 if ( /foo(.*?)bar/ ) { print "got <$1>\n" }
806 got <d is under the >
808 Here's another example: let's say you'd like to match a number at the end
809 of a string, and you also want to keep the preceding part of the match.
812 $_ = "I have 2 numbers: 53147";
813 if ( /(.*)(\d*)/ ) { # Wrong!
814 print "Beginning is <$1>, number is <$2>.\n";
817 That won't work at all, because C<.*> was greedy and gobbled up the
818 whole string. As C<\d*> can match on an empty string the complete
819 regular expression matched successfully.
821 Beginning is <I have 2 numbers: 53147>, number is <>.
823 Here are some variants, most of which don't work:
825 $_ = "I have 2 numbers: 53147";
838 printf "%-12s ", $pat;
848 (.*)(\d*) <I have 2 numbers: 53147> <>
849 (.*)(\d+) <I have 2 numbers: 5314> <7>
851 (.*?)(\d+) <I have > <2>
852 (.*)(\d+)$ <I have 2 numbers: 5314> <7>
853 (.*?)(\d+)$ <I have 2 numbers: > <53147>
854 (.*)\b(\d+)$ <I have 2 numbers: > <53147>
855 (.*\D)(\d+)$ <I have 2 numbers: > <53147>
857 As you see, this can be a bit tricky. It's important to realize that a
858 regular expression is merely a set of assertions that gives a definition
859 of success. There may be 0, 1, or several different ways that the
860 definition might succeed against a particular string. And if there are
861 multiple ways it might succeed, you need to understand backtracking to
862 know which variety of success you will achieve.
864 When using look-ahead assertions and negations, this can all get even
865 tricker. Imagine you'd like to find a sequence of non-digits not
866 followed by "123". You might try to write that as
869 if ( /^\D*(?!123)/ ) { # Wrong!
870 print "Yup, no 123 in $_\n";
873 But that isn't going to match; at least, not the way you're hoping. It
874 claims that there is no 123 in the string. Here's a clearer picture of
875 why that pattern matches, contrary to popular expectations:
880 print "1: got $1\n" if $x =~ /^(ABC)(?!123)/ ;
881 print "2: got $1\n" if $y =~ /^(ABC)(?!123)/ ;
883 print "3: got $1\n" if $x =~ /^(\D*)(?!123)/ ;
884 print "4: got $1\n" if $y =~ /^(\D*)(?!123)/ ;
892 You might have expected test 3 to fail because it seems to a more
893 general purpose version of test 1. The important difference between
894 them is that test 3 contains a quantifier (C<\D*>) and so can use
895 backtracking, whereas test 1 will not. What's happening is
896 that you've asked "Is it true that at the start of $x, following 0 or more
897 non-digits, you have something that's not 123?" If the pattern matcher had
898 let C<\D*> expand to "ABC", this would have caused the whole pattern to
901 The search engine will initially match C<\D*> with "ABC". Then it will
902 try to match C<(?!123> with "123", which fails. But because
903 a quantifier (C<\D*>) has been used in the regular expression, the
904 search engine can backtrack and retry the match differently
905 in the hope of matching the complete regular expression.
907 The pattern really, I<really> wants to succeed, so it uses the
908 standard pattern back-off-and-retry and lets C<\D*> expand to just "AB" this
909 time. Now there's indeed something following "AB" that is not
910 "123". It's "C123", which suffices.
912 We can deal with this by using both an assertion and a negation.
913 We'll say that the first part in $1 must be followed both by a digit
914 and by something that's not "123". Remember that the look-aheads
915 are zero-width expressions--they only look, but don't consume any
916 of the string in their match. So rewriting this way produces what
917 you'd expect; that is, case 5 will fail, but case 6 succeeds:
919 print "5: got $1\n" if $x =~ /^(\D*)(?=\d)(?!123)/ ;
920 print "6: got $1\n" if $y =~ /^(\D*)(?=\d)(?!123)/ ;
924 In other words, the two zero-width assertions next to each other work as though
925 they're ANDed together, just as you'd use any built-in assertions: C</^$/>
926 matches only if you're at the beginning of the line AND the end of the
927 line simultaneously. The deeper underlying truth is that juxtaposition in
928 regular expressions always means AND, except when you write an explicit OR
929 using the vertical bar. C</ab/> means match "a" AND (then) match "b",
930 although the attempted matches are made at different positions because "a"
931 is not a zero-width assertion, but a one-width assertion.
933 B<WARNING>: particularly complicated regular expressions can take
934 exponential time to solve because of the immense number of possible
935 ways they can use backtracking to try match. For example, without
936 internal optimizations done by the regular expression engine, this will
937 take a painfully long time to run:
939 'aaaaaaaaaaaa' =~ /((a{0,5}){0,5})*[c]/
941 And if you used C<*>'s in the internal groups instead of limiting them
942 to 0 through 5 matches, then it would take forever--or until you ran
943 out of stack space. Moreover, these internal optimizations are not
944 always applicable. For example, if you put C<{0,5}> instead of C<*>
945 on the external group, no current optimization is applicable, and the
946 match takes a long time to finish.
948 A powerful tool for optimizing such beasts is what is known as an
950 which does not backtrack (see L<C<< (?>pattern) >>>). Note also that
951 zero-length look-ahead/look-behind assertions will not backtrack to make
952 the tail match, since they are in "logical" context: only
953 whether they match is considered relevant. For an example
954 where side-effects of look-ahead I<might> have influenced the
955 following match, see L<C<< (?>pattern) >>>.
957 =head2 Version 8 Regular Expressions
959 In case you're not familiar with the "regular" Version 8 regex
960 routines, here are the pattern-matching rules not described above.
962 Any single character matches itself, unless it is a I<metacharacter>
963 with a special meaning described here or above. You can cause
964 characters that normally function as metacharacters to be interpreted
965 literally by prefixing them with a "\" (e.g., "\." matches a ".", not any
966 character; "\\" matches a "\"). A series of characters matches that
967 series of characters in the target string, so the pattern C<blurfl>
968 would match "blurfl" in the target string.
970 You can specify a character class, by enclosing a list of characters
971 in C<[]>, which will match any one character from the list. If the
972 first character after the "[" is "^", the class matches any character not
973 in the list. Within a list, the "-" character specifies a
974 range, so that C<a-z> represents all characters between "a" and "z",
975 inclusive. If you want either "-" or "]" itself to be a member of a
976 class, put it at the start of the list (possibly after a "^"), or
977 escape it with a backslash. "-" is also taken literally when it is
978 at the end of the list, just before the closing "]". (The
979 following all specify the same class of three characters: C<[-az]>,
980 C<[az-]>, and C<[a\-z]>. All are different from C<[a-z]>, which
981 specifies a class containing twenty-six characters, even on EBCDIC
982 based coded character sets.) Also, if you try to use the character
983 classes C<\w>, C<\W>, C<\s>, C<\S>, C<\d>, or C<\D> as endpoints of
984 a range, that's not a range, the "-" is understood literally.
986 Note also that the whole range idea is rather unportable between
987 character sets--and even within character sets they may cause results
988 you probably didn't expect. A sound principle is to use only ranges
989 that begin from and end at either alphabets of equal case ([a-e],
990 [A-E]), or digits ([0-9]). Anything else is unsafe. If in doubt,
991 spell out the character sets in full.
993 Characters may be specified using a metacharacter syntax much like that
994 used in C: "\n" matches a newline, "\t" a tab, "\r" a carriage return,
995 "\f" a form feed, etc. More generally, \I<nnn>, where I<nnn> is a string
996 of octal digits, matches the character whose coded character set value
997 is I<nnn>. Similarly, \xI<nn>, where I<nn> are hexadecimal digits,
998 matches the character whose numeric value is I<nn>. The expression \cI<x>
999 matches the character control-I<x>. Finally, the "." metacharacter
1000 matches any character except "\n" (unless you use C</s>).
1002 You can specify a series of alternatives for a pattern using "|" to
1003 separate them, so that C<fee|fie|foe> will match any of "fee", "fie",
1004 or "foe" in the target string (as would C<f(e|i|o)e>). The
1005 first alternative includes everything from the last pattern delimiter
1006 ("(", "[", or the beginning of the pattern) up to the first "|", and
1007 the last alternative contains everything from the last "|" to the next
1008 pattern delimiter. That's why it's common practice to include
1009 alternatives in parentheses: to minimize confusion about where they
1012 Alternatives are tried from left to right, so the first
1013 alternative found for which the entire expression matches, is the one that
1014 is chosen. This means that alternatives are not necessarily greedy. For
1015 example: when matching C<foo|foot> against "barefoot", only the "foo"
1016 part will match, as that is the first alternative tried, and it successfully
1017 matches the target string. (This might not seem important, but it is
1018 important when you are capturing matched text using parentheses.)
1020 Also remember that "|" is interpreted as a literal within square brackets,
1021 so if you write C<[fee|fie|foe]> you're really only matching C<[feio|]>.
1023 Within a pattern, you may designate subpatterns for later reference
1024 by enclosing them in parentheses, and you may refer back to the
1025 I<n>th subpattern later in the pattern using the metacharacter
1026 \I<n>. Subpatterns are numbered based on the left to right order
1027 of their opening parenthesis. A backreference matches whatever
1028 actually matched the subpattern in the string being examined, not
1029 the rules for that subpattern. Therefore, C<(0|0x)\d*\s\1\d*> will
1030 match "0x1234 0x4321", but not "0x1234 01234", because subpattern
1031 1 matched "0x", even though the rule C<0|0x> could potentially match
1032 the leading 0 in the second number.
1034 =head2 Warning on \1 vs $1
1036 Some people get too used to writing things like:
1038 $pattern =~ s/(\W)/\\\1/g;
1040 This is grandfathered for the RHS of a substitute to avoid shocking the
1041 B<sed> addicts, but it's a dirty habit to get into. That's because in
1042 PerlThink, the righthand side of an C<s///> is a double-quoted string. C<\1> in
1043 the usual double-quoted string means a control-A. The customary Unix
1044 meaning of C<\1> is kludged in for C<s///>. However, if you get into the habit
1045 of doing that, you get yourself into trouble if you then add an C</e>
1048 s/(\d+)/ \1 + 1 /eg; # causes warning under -w
1054 You can't disambiguate that by saying C<\{1}000>, whereas you can fix it with
1055 C<${1}000>. The operation of interpolation should not be confused
1056 with the operation of matching a backreference. Certainly they mean two
1057 different things on the I<left> side of the C<s///>.
1059 =head2 Repeated patterns matching zero-length substring
1061 B<WARNING>: Difficult material (and prose) ahead. This section needs a rewrite.
1063 Regular expressions provide a terse and powerful programming language. As
1064 with most other power tools, power comes together with the ability
1067 A common abuse of this power stems from the ability to make infinite
1068 loops using regular expressions, with something as innocuous as:
1070 'foo' =~ m{ ( o? )* }x;
1072 The C<o?> can match at the beginning of C<'foo'>, and since the position
1073 in the string is not moved by the match, C<o?> would match again and again
1074 because of the C<*> modifier. Another common way to create a similar cycle
1075 is with the looping modifier C<//g>:
1077 @matches = ( 'foo' =~ m{ o? }xg );
1081 print "match: <$&>\n" while 'foo' =~ m{ o? }xg;
1083 or the loop implied by split().
1085 However, long experience has shown that many programming tasks may
1086 be significantly simplified by using repeated subexpressions that
1087 may match zero-length substrings. Here's a simple example being:
1089 @chars = split //, $string; # // is not magic in split
1090 ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
1092 Thus Perl allows such constructs, by I<forcefully breaking
1093 the infinite loop>. The rules for this are different for lower-level
1094 loops given by the greedy modifiers C<*+{}>, and for higher-level
1095 ones like the C</g> modifier or split() operator.
1097 The lower-level loops are I<interrupted> (that is, the loop is
1098 broken) when Perl detects that a repeated expression matched a
1099 zero-length substring. Thus
1101 m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
1103 is made equivalent to
1105 m{ (?: NON_ZERO_LENGTH )*
1110 The higher level-loops preserve an additional state between iterations:
1111 whether the last match was zero-length. To break the loop, the following
1112 match after a zero-length match is prohibited to have a length of zero.
1113 This prohibition interacts with backtracking (see L<"Backtracking">),
1114 and so the I<second best> match is chosen if the I<best> match is of
1122 results in C<< <><b><><a><><r><> >>. At each position of the string the best
1123 match given by non-greedy C<??> is the zero-length match, and the I<second
1124 best> match is what is matched by C<\w>. Thus zero-length matches
1125 alternate with one-character-long matches.
1127 Similarly, for repeated C<m/()/g> the second-best match is the match at the
1128 position one notch further in the string.
1130 The additional state of being I<matched with zero-length> is associated with
1131 the matched string, and is reset by each assignment to pos().
1132 Zero-length matches at the end of the previous match are ignored
1135 =head2 Combining pieces together
1137 Each of the elementary pieces of regular expressions which were described
1138 before (such as C<ab> or C<\Z>) could match at most one substring
1139 at the given position of the input string. However, in a typical regular
1140 expression these elementary pieces are combined into more complicated
1141 patterns using combining operators C<ST>, C<S|T>, C<S*> etc
1142 (in these examples C<S> and C<T> are regular subexpressions).
1144 Such combinations can include alternatives, leading to a problem of choice:
1145 if we match a regular expression C<a|ab> against C<"abc">, will it match
1146 substring C<"a"> or C<"ab">? One way to describe which substring is
1147 actually matched is the concept of backtracking (see L<"Backtracking">).
1148 However, this description is too low-level and makes you think
1149 in terms of a particular implementation.
1151 Another description starts with notions of "better"/"worse". All the
1152 substrings which may be matched by the given regular expression can be
1153 sorted from the "best" match to the "worst" match, and it is the "best"
1154 match which is chosen. This substitutes the question of "what is chosen?"
1155 by the question of "which matches are better, and which are worse?".
1157 Again, for elementary pieces there is no such question, since at most
1158 one match at a given position is possible. This section describes the
1159 notion of better/worse for combining operators. In the description
1160 below C<S> and C<T> are regular subexpressions.
1166 Consider two possible matches, C<AB> and C<A'B'>, C<A> and C<A'> are
1167 substrings which can be matched by C<S>, C<B> and C<B'> are substrings
1168 which can be matched by C<T>.
1170 If C<A> is better match for C<S> than C<A'>, C<AB> is a better
1173 If C<A> and C<A'> coincide: C<AB> is a better match than C<AB'> if
1174 C<B> is better match for C<T> than C<B'>.
1178 When C<S> can match, it is a better match than when only C<T> can match.
1180 Ordering of two matches for C<S> is the same as for C<S>. Similar for
1181 two matches for C<T>.
1183 =item C<S{REPEAT_COUNT}>
1185 Matches as C<SSS...S> (repeated as many times as necessary).
1189 Matches as C<S{max}|S{max-1}|...|S{min+1}|S{min}>.
1191 =item C<S{min,max}?>
1193 Matches as C<S{min}|S{min+1}|...|S{max-1}|S{max}>.
1195 =item C<S?>, C<S*>, C<S+>
1197 Same as C<S{0,1}>, C<S{0,BIG_NUMBER}>, C<S{1,BIG_NUMBER}> respectively.
1199 =item C<S??>, C<S*?>, C<S+?>
1201 Same as C<S{0,1}?>, C<S{0,BIG_NUMBER}?>, C<S{1,BIG_NUMBER}?> respectively.
1205 Matches the best match for C<S> and only that.
1207 =item C<(?=S)>, C<(?<=S)>
1209 Only the best match for C<S> is considered. (This is important only if
1210 C<S> has capturing parentheses, and backreferences are used somewhere
1211 else in the whole regular expression.)
1213 =item C<(?!S)>, C<(?<!S)>
1215 For this grouping operator there is no need to describe the ordering, since
1216 only whether or not C<S> can match is important.
1218 =item C<(??{ EXPR })>
1220 The ordering is the same as for the regular expression which is
1223 =item C<(?(condition)yes-pattern|no-pattern)>
1225 Recall that which of C<yes-pattern> or C<no-pattern> actually matches is
1226 already determined. The ordering of the matches is the same as for the
1227 chosen subexpression.
1231 The above recipes describe the ordering of matches I<at a given position>.
1232 One more rule is needed to understand how a match is determined for the
1233 whole regular expression: a match at an earlier position is always better
1234 than a match at a later position.
1236 =head2 Creating custom RE engines
1238 Overloaded constants (see L<overload>) provide a simple way to extend
1239 the functionality of the RE engine.
1241 Suppose that we want to enable a new RE escape-sequence C<\Y|> which
1242 matches at boundary between white-space characters and non-whitespace
1243 characters. Note that C<(?=\S)(?<!\S)|(?!\S)(?<=\S)> matches exactly
1244 at these positions, so we want to have each C<\Y|> in the place of the
1245 more complicated version. We can create a module C<customre> to do
1253 die "No argument to customre::import allowed" if @_;
1254 overload::constant 'qr' => \&convert;
1257 sub invalid { die "/$_[0]/: invalid escape '\\$_[1]'"}
1259 my %rules = ( '\\' => '\\',
1260 'Y|' => qr/(?=\S)(?<!\S)|(?!\S)(?<=\S)/ );
1266 { $rules{$1} or invalid($re,$1) }sgex;
1270 Now C<use customre> enables the new escape in constant regular
1271 expressions, i.e., those without any runtime variable interpolations.
1272 As documented in L<overload>, this conversion will work only over
1273 literal parts of regular expressions. For C<\Y|$re\Y|> the variable
1274 part of this regular expression needs to be converted explicitly
1275 (but only if the special meaning of C<\Y|> should be enabled inside $re):
1280 $re = customre::convert $re;
1285 This document varies from difficult to understand to completely
1286 and utterly opaque. The wandering prose riddled with jargon is
1287 hard to fathom in several places.
1289 This document needs a rewrite that separates the tutorial content
1290 from the reference content.
1298 L<perlop/"Regexp Quote-Like Operators">.
1300 L<perlop/"Gory details of parsing quoted constructs">.
1310 I<Mastering Regular Expressions> by Jeffrey Friedl, published
1311 by O'Reilly and Associates.