Debian Codegraph @

upstream/debian-unstable/man1/perlre.1

raw
1 .\" -*- mode: troff; coding: utf-8 -*-
2 .\" Automatically generated by Pod::Man 5.0102 (Pod::Simple 3.45)
3 .\"
4 .\" Standard preamble:
5 .\" ========================================================================
6 .de Sp \" Vertical space (when we can't use .PP)
7 .if t .sp .5v
8 .if n .sp
9 ..
10 .de Vb \" Begin verbatim text
11 .ft CW
12 .nf
13 .ne \\$1
14 ..
15 .de Ve \" End verbatim text
16 .ft R
17 .fi
18 ..
19 .\" \*(C` and \*(C' are quotes in nroff, nothing in troff, for use with C<>.
20 .ie n \{\
21 . ds C` ""
22 . ds C' ""
23 'br\}
24 .el\{\
25 . ds C`
26 . ds C'
27 'br\}
28 .\"
29 .\" Escape single quotes in literal strings from groff's Unicode transform.
30 .ie \n(.g .ds Aq \(aq
31 .el .ds Aq '
32 .\"
33 .\" If the F register is >0, we'll generate index entries on stderr for
34 .\" titles (.TH), headers (.SH), subsections (.SS), items (.Ip), and index
35 .\" entries marked with X<> in POD. Of course, you'll have to process the
36 .\" output yourself in some meaningful fashion.
37 .\"
38 .\" Avoid warning from groff about undefined register 'F'.
39 .de IX
40 ..
41 .nr rF 0
42 .if \n(.g .if rF .nr rF 1
43 .if (\n(rF:(\n(.g==0)) \{\
44 . if \nF \{\
45 . de IX
46 . tm Index:\\$1\t\\n%\t"\\$2"
47 ..
48 . if !\nF==2 \{\
49 . nr % 0
50 . nr F 2
51 . \}
52 . \}
53 .\}
54 .rr rF
55 .\" ========================================================================
56 .\"
57 .IX Title "PERLRE 1"
58 .TH PERLRE 1 2025-11-16 "perl v5.40.1" "Perl Programmers Reference Guide"
59 .\" For nroff, turn off justification. Always turn off hyphenation; it makes
60 .\" way too many mistakes in technical documents.
61 .if n .ad l
62 .nh
63 .SH NAME
64 perlre \- Perl regular expressions
65 .IX Xref "regular expression regex regexp"
66 .SH DESCRIPTION
67 .IX Header "DESCRIPTION"
68 This page describes the syntax of regular expressions in Perl.
69 .PP
70 If you haven't used regular expressions before, a tutorial introduction
71 is available in perlretut. If you know just a little about them,
72 a quick-start introduction is available in perlrequick.
73 .PP
74 Except for "The Basics" section, this page assumes you are familiar
75 with regular expression basics, like what is a "pattern", what does it
76 look like, and how it is basically used. For a reference on how they
77 are used, plus various examples of the same, see discussions of \f(CW\*(C`m//\*(C'\fR,
78 \&\f(CW\*(C`s///\*(C'\fR, \f(CW\*(C`qr//\*(C'\fR and \f(CW"??"\fR in "Regexp Quote-Like Operators" in perlop.
79 .PP
80 New in v5.22, \f(CW\*(C`use re \*(Aqstrict\*(Aq\*(C'\fR applies stricter
81 rules than otherwise when compiling regular expression patterns. It can
82 find things that, while legal, may not be what you intended.
83 .SS "The Basics"
84 .IX Xref "regular expression, version 8 regex, version 8 regexp, version 8"
85 .IX Subsection "The Basics"
86 Regular expressions are strings with the very particular syntax and
87 meaning described in this document and auxiliary documents referred to
88 by this one. The strings are called "patterns". Patterns are used to
89 determine if some other string, called the "target", has (or doesn't
90 have) the characteristics specified by the pattern. We call this
91 "matching" the target string against the pattern. Usually the match is
92 done by having the target be the first operand, and the pattern be the
93 second operand, of one of the two binary operators \f(CW\*(C`=~\*(C'\fR and \f(CW\*(C`!~\*(C'\fR,
94 listed in "Binding Operators" in perlop; and the pattern will have been
95 converted from an ordinary string by one of the operators in
96 "Regexp Quote-Like Operators" in perlop, like so:
97 .PP
98 .Vb 1
99 \& $foo =~ m/abc/
100 .Ve
101 .PP
102 This evaluates to true if and only if the string in the variable \f(CW$foo\fR
103 contains somewhere in it, the sequence of characters "a", "b", then "c".
104 (The \f(CW\*(C`=~ m\*(C'\fR, or match operator, is described in
105 "m/PATTERN/msixpodualngc" in perlop.)
106 .PP
107 Patterns that aren't already stored in some variable must be delimited,
108 at both ends, by delimiter characters. These are often, as in the
109 example above, forward slashes, and the typical way a pattern is written
110 in documentation is with those slashes. In most cases, the delimiter
111 is the same character, fore and aft, but there are a few cases where a
112 character looks like it has a mirror-image mate, where the opening
113 version is the beginning delimiter, and the closing one is the ending
114 delimiter, like
115 .PP
116 .Vb 1
117 \& $foo =~ m<abc>
118 .Ve
119 .PP
120 Most times, the pattern is evaluated in double-quotish context, but it
121 is possible to choose delimiters to force single-quotish, like
122 .PP
123 .Vb 1
124 \& $foo =~ m\*(Aqabc\*(Aq
125 .Ve
126 .PP
127 If the pattern contains its delimiter within it, that delimiter must be
128 escaped. Prefixing it with a backslash (\fIe.g.\fR, \f(CW"/foo\e/bar/"\fR)
129 serves this purpose.
130 .PP
131 Any single character in a pattern matches that same character in the
132 target string, unless the character is a \fImetacharacter\fR with a special
133 meaning described in this document. A sequence of non-metacharacters
134 matches the same sequence in the target string, as we saw above with
135 \&\f(CW\*(C`m/abc/\*(C'\fR.
136 .PP
137 Only a few characters (all of them being ASCII punctuation characters)
138 are metacharacters. The most commonly used one is a dot \f(CW"."\fR, which
139 normally matches almost any character (including a dot itself).
140 .PP
141 You can cause characters that normally function as metacharacters to be
142 interpreted literally by prefixing them with a \f(CW"\e"\fR, just like the
143 pattern's delimiter must be escaped if it also occurs within the
144 pattern. Thus, \f(CW"\e."\fR matches just a literal dot, \f(CW"."\fR instead of
145 its normal meaning. This means that the backslash is also a
146 metacharacter, so \f(CW"\e\e"\fR matches a single \f(CW"\e"\fR. And a sequence that
147 contains an escaped metacharacter matches the same sequence (but without
148 the escape) in the target string. So, the pattern \f(CW\*(C`/blur\e\efl/\*(C'\fR would
149 match any target string that contains the sequence \f(CW"blur\efl"\fR.
150 .PP
151 The metacharacter \f(CW"|"\fR is used to match one thing or another. Thus
152 .PP
153 .Vb 1
154 \& $foo =~ m/this|that/
155 .Ve
156 .PP
157 is TRUE if and only if \f(CW$foo\fR contains either the sequence \f(CW"this"\fR or
158 the sequence \f(CW"that"\fR. Like all metacharacters, prefixing the \f(CW"|"\fR
159 with a backslash makes it match the plain punctuation character; in its
160 case, the VERTICAL LINE.
161 .PP
162 .Vb 1
163 \& $foo =~ m/this\e|that/
164 .Ve
165 .PP
166 is TRUE if and only if \f(CW$foo\fR contains the sequence \f(CW"this|that"\fR.
167 .PP
168 You aren't limited to just a single \f(CW"|"\fR.
169 .PP
170 .Vb 1
171 \& $foo =~ m/fee|fie|foe|fum/
172 .Ve
173 .PP
174 is TRUE if and only if \f(CW$foo\fR contains any of those 4 sequences from
175 the children's story "Jack and the Beanstalk".
176 .PP
177 As you can see, the \f(CW"|"\fR binds less tightly than a sequence of
178 ordinary characters. We can override this by using the grouping
179 metacharacters, the parentheses \f(CW"("\fR and \f(CW")"\fR.
180 .PP
181 .Vb 1
182 \& $foo =~ m/th(is|at) thing/
183 .Ve
184 .PP
185 is TRUE if and only if \f(CW$foo\fR contains either the sequence \f(CW"this\ thing"\fR or the sequence \f(CW"that\ thing"\fR. The portions of the string
186 that match the portions of the pattern enclosed in parentheses are
187 normally made available separately for use later in the pattern,
188 substitution, or program. This is called "capturing", and it can get
189 complicated. See "Capture groups".
190 .PP
191 The first alternative includes everything from the last pattern
192 delimiter (\f(CW"("\fR, \f(CW"(?:"\fR (described later), \fIetc\fR. or the beginning
193 of the pattern) up to the first \f(CW"|"\fR, and the last alternative
194 contains everything from the last \f(CW"|"\fR to the next closing pattern
195 delimiter. That's why it's common practice to include alternatives in
196 parentheses: to minimize confusion about where they start and end.
197 .PP
198 Alternatives are tried from left to right, so the first
199 alternative found for which the entire expression matches, is the one that
200 is chosen. This means that alternatives are not necessarily greedy. For
201 example: when matching \f(CW\*(C`foo|foot\*(C'\fR against \f(CW"barefoot"\fR, only the \f(CW"foo"\fR
202 part will match, as that is the first alternative tried, and it successfully
203 matches the target string. (This might not seem important, but it is
204 important when you are capturing matched text using parentheses.)
205 .PP
206 Besides taking away the special meaning of a metacharacter, a prefixed
207 backslash changes some letter and digit characters away from matching
208 just themselves to instead have special meaning. These are called
209 "escape sequences", and all such are described in perlrebackslash. A
210 backslash sequence (of a letter or digit) that doesn't currently have
211 special meaning to Perl will raise a warning if warnings are enabled,
212 as those are reserved for potential future use.
213 .PP
214 One such sequence is \f(CW\*(C`\eb\*(C'\fR, which matches a boundary of some sort.
215 \&\f(CW\*(C`\eb{wb}\*(C'\fR and a few others give specialized types of boundaries.
216 (They are all described in detail starting at
217 "\eb{}, \eb, \eB{}, \eB" in perlrebackslash.) Note that these don't match
218 characters, but the zero-width spaces between characters. They are an
219 example of a zero-width assertion. Consider again,
220 .PP
221 .Vb 1
222 \& $foo =~ m/fee|fie|foe|fum/
223 .Ve
224 .PP
225 It evaluates to TRUE if, besides those 4 words, any of the sequences
226 "feed", "field", "Defoe", "fume", and many others are in \f(CW$foo\fR. By
227 judicious use of \f(CW\*(C`\eb\*(C'\fR (or better (because it is designed to handle
228 natural language) \f(CW\*(C`\eb{wb}\*(C'\fR), we can make sure that only the Giant's
229 words are matched:
230 .PP
231 .Vb 2
232 \& $foo =~ m/\eb(fee|fie|foe|fum)\eb/
233 \& $foo =~ m/\eb{wb}(fee|fie|foe|fum)\eb{wb}/
234 .Ve
235 .PP
236 The final example shows that the characters \f(CW"{"\fR and \f(CW"}"\fR are
237 metacharacters.
238 .PP
239 Another use for escape sequences is to specify characters that cannot
240 (or which you prefer not to) be written literally. These are described
241 in detail in "Character Escapes" in perlrebackslash, but the next three
242 paragraphs briefly describe some of them.
243 .PP
244 Various control characters can be written in C language style: \f(CW"\en"\fR
245 matches a newline, \f(CW"\et"\fR a tab, \f(CW"\er"\fR a carriage return, \f(CW"\ef"\fR a
246 form feed, \fIetc\fR.
247 .PP
248 More generally, \f(CW\*(C`\e\fR\f(CInnn\fR\f(CW\*(C'\fR, where \fInnn\fR is a string of three octal
249 digits, matches the character whose native code point is \fInnn\fR. You
250 can easily run into trouble if you don't have exactly three digits. So
251 always use three, or since Perl 5.14, you can use \f(CW\*(C`\eo{...}\*(C'\fR to specify
252 any number of octal digits.
253 .PP
254 Similarly, \f(CW\*(C`\ex\fR\f(CInn\fR\f(CW\*(C'\fR, where \fInn\fR are hexadecimal digits, matches the
255 character whose native ordinal is \fInn\fR. Again, not using exactly two
256 digits is a recipe for disaster, but you can use \f(CW\*(C`\ex{...}\*(C'\fR to specify
257 any number of hex digits.
258 .PP
259 Besides being a metacharacter, the \f(CW"."\fR is an example of a "character
260 class", something that can match any single character of a given set of
261 them. In its case, the set is just about all possible characters. Perl
262 predefines several character classes besides the \f(CW"."\fR; there is a
263 separate reference page about just these, perlrecharclass.
264 .PP
265 You can define your own custom character classes, by putting into your
266 pattern in the appropriate place(s), a list of all the characters you
267 want in the set. You do this by enclosing the list within \f(CW\*(C`[]\*(C'\fR bracket
268 characters. These are called "bracketed character classes" when we are
269 being precise, but often the word "bracketed" is dropped. (Dropping it
270 usually doesn't cause confusion.) This means that the \f(CW"["\fR character
271 is another metacharacter. It doesn't match anything just by itself; it
272 is used only to tell Perl that what follows it is a bracketed character
273 class. If you want to match a literal left square bracket, you must
274 escape it, like \f(CW"\e["\fR. The matching \f(CW"]"\fR is also a metacharacter;
275 again it doesn't match anything by itself, but just marks the end of
276 your custom class to Perl. It is an example of a "sometimes
277 metacharacter". It isn't a metacharacter if there is no corresponding
278 \&\f(CW"["\fR, and matches its literal self:
279 .PP
280 .Vb 1
281 \& print "]" =~ /]/; # prints 1
282 .Ve
283 .PP
284 The list of characters within the character class gives the set of
285 characters matched by the class. \f(CW"[abc]"\fR matches a single "a" or "b"
286 or "c". But if the first character after the \f(CW"["\fR is \f(CW"^"\fR, the
287 class instead matches any character not in the list. Within a list, the
288 \&\f(CW"\-"\fR character specifies a range of characters, so that \f(CW\*(C`a\-z\*(C'\fR
289 represents all characters between "a" and "z", inclusive. If you want
290 either \f(CW"\-"\fR or \f(CW"]"\fR itself to be a member of a class, put it at the
291 start of the list (possibly after a \f(CW"^"\fR), or escape it with a
292 backslash. \f(CW"\-"\fR is also taken literally when it is at the end of the
293 list, just before the closing \f(CW"]"\fR. (The following all specify the
294 same class of three characters: \f(CW\*(C`[\-az]\*(C'\fR, \f(CW\*(C`[az\-]\*(C'\fR, and \f(CW\*(C`[a\e\-z]\*(C'\fR. All
295 are different from \f(CW\*(C`[a\-z]\*(C'\fR, which specifies a class containing
296 twenty-six characters, even on EBCDIC-based character sets.)
297 .PP
298 There is lots more to bracketed character classes; full details are in
299 "Bracketed Character Classes" in perlrecharclass.
300 .PP
301 \fIMetacharacters\fR
302 .IX Xref "metacharacter \\ ^ . $ | ( () [ []"
303 .IX Subsection "Metacharacters"
304 .PP
305 "The Basics" introduced some of the metacharacters. This section
306 gives them all. Most of them have the same meaning as in the \fIegrep\fR
307 command.
308 .PP
309 Only the \f(CW"\e"\fR is always a metacharacter. The others are metacharacters
310 just sometimes. The following tables lists all of them, summarizes
311 their use, and gives the contexts where they are metacharacters.
312 Outside those contexts or if prefixed by a \f(CW"\e"\fR, they match their
313 corresponding punctuation character. In some cases, their meaning
314 varies depending on various pattern modifiers that alter the default
315 behaviors. See "Modifiers".
316 .PP
317 .Vb 10
318 \& PURPOSE WHERE
319 \& \e Escape the next character Always, except when
320 \& escaped by another \e
321 \& ^ Match the beginning of the string Not in []
322 \& (or line, if /m is used)
323 \& ^ Complement the [] class At the beginning of []
324 \& . Match any single character except newline Not in []
325 \& (under /s, includes newline)
326 \& $ Match the end of the string Not in [], but can
327 \& (or before newline at the end of the mean interpolate a
328 \& string; or before any newline if /m is scalar
329 \& used)
330 \& | Alternation Not in []
331 \& () Grouping Not in []
332 \& [ Start Bracketed Character class Not in []
333 \& ] End Bracketed Character class Only in [], and
334 \& not first
335 \& * Matches the preceding element 0 or more Not in []
336 \& times
337 \& + Matches the preceding element 1 or more Not in []
338 \& times
339 \& ? Matches the preceding element 0 or 1 Not in []
340 \& times
341 \& { Starts a sequence that gives number(s) Not in []
342 \& of times the preceding element can be
343 \& matched
344 \& { when following certain escape sequences
345 \& starts a modifier to the meaning of the
346 \& sequence
347 \& } End sequence started by {
348 \& \- Indicates a range Only in [] interior
349 \& # Beginning of comment, extends to line end Only with /x modifier
350 .Ve
351 .PP
352 Notice that most of the metacharacters lose their special meaning when
353 they occur in a bracketed character class, except \f(CW"^"\fR has a different
354 meaning when it is at the beginning of such a class. And \f(CW"\-"\fR and \f(CW"]"\fR
355 are metacharacters only at restricted positions within bracketed
356 character classes; while \f(CW"}"\fR is a metacharacter only when closing a
357 special construct started by \f(CW"{"\fR.
358 .PP
359 In double-quotish context, as is usually the case, you need to be
360 careful about \f(CW"$"\fR and the non-metacharacter \f(CW"@"\fR. Those could
361 interpolate variables, which may or may not be what you intended.
362 .PP
363 These rules were designed for compactness of expression, rather than
364 legibility and maintainability. The "/x and /xx" pattern
365 modifiers allow you to insert white space to improve readability. And
366 use of \f(CW\*(C`re\ \*(Aqstrict\*(Aq\*(C'\fR adds extra checking to
367 catch some typos that might silently compile into something unintended.
368 .PP
369 By default, the \f(CW"^"\fR character is guaranteed to match only the
370 beginning of the string, the \f(CW"$"\fR character only the end (or before the
371 newline at the end), and Perl does certain optimizations with the
372 assumption that the string contains only one line. Embedded newlines
373 will not be matched by \f(CW"^"\fR or \f(CW"$"\fR. You may, however, wish to treat a
374 string as a multi-line buffer, such that the \f(CW"^"\fR will match after any
375 newline within the string (except if the newline is the last character in
376 the string), and \f(CW"$"\fR will match before any newline. At the
377 cost of a little more overhead, you can do this by using the
378 \&\f(CW"/m"\fR modifier on the pattern match operator. (Older programs
379 did this by setting \f(CW$*\fR, but this option was removed in perl 5.10.)
380 .IX Xref "^ $ m"
381 .PP
382 To simplify multi-line substitutions, the \f(CW"."\fR character never matches a
383 newline unless you use the \f(CW\*(C`/s\*(C'\fR modifier, which in effect tells
384 Perl to pretend the string is a single line\-\-even if it isn't.
385 .IX Xref ". s"
386 .SS Modifiers
387 .IX Subsection "Modifiers"
388 \fIOverview\fR
389 .IX Subsection "Overview"
390 .PP
391 The default behavior for matching can be changed, using various
392 modifiers. Modifiers that relate to the interpretation of the pattern
393 are listed just below. Modifiers that alter the way a pattern is used
394 by Perl are detailed in "Regexp Quote-Like Operators" in perlop and
395 "Gory details of parsing quoted constructs" in perlop. Modifiers can be added
396 dynamically; see "Extended Patterns" below.
397 .ie n .IP "\fR\fB""m""\fR\fB\fR" 4
398 .el .IP \fR\f(CBm\fR\fB\fR 4
399 .IX Xref " m regex, multiline regexp, multiline regular expression, multiline"
400 .IX Item "m"
401 Treat the string being matched against as multiple lines. That is, change \f(CW"^"\fR and \f(CW"$"\fR from matching
402 the start of the string's first line and the end of its last line to
403 matching the start and end of each line within the string.
404 .ie n .IP "\fR\fB""s""\fR\fB\fR" 4
405 .el .IP \fR\f(CBs\fR\fB\fR 4
406 .IX Xref " s regex, single-line regexp, single-line regular expression, single-line"
407 .IX Item "s"
408 Treat the string as single line. That is, change \f(CW"."\fR to match any character
409 whatsoever, even a newline, which normally it would not match.
410 .Sp
411 Used together, as \f(CW\*(C`/ms\*(C'\fR, they let the \f(CW"."\fR match any character whatsoever,
412 while still allowing \f(CW"^"\fR and \f(CW"$"\fR to match, respectively, just after
413 and just before newlines within the string.
414 .ie n .IP "\fR\fB""i""\fR\fB\fR" 4
415 .el .IP \fR\f(CBi\fR\fB\fR 4
416 .IX Xref " i regex, case-insensitive regexp, case-insensitive regular expression, case-insensitive"
417 .IX Item "i"
418 Do case-insensitive pattern matching. For example, "A" will match "a"
419 under \f(CW\*(C`/i\*(C'\fR.
420 .Sp
421 If locale matching rules are in effect, the case map is taken from the
422 current
423 locale for code points less than 255, and from Unicode rules for larger
424 code points. However, matches that would cross the Unicode
425 rules/non\-Unicode rules boundary (ords 255/256) will not succeed, unless
426 the locale is a UTF\-8 one. See perllocale.
427 .Sp
428 There are a number of Unicode characters that match a sequence of
429 multiple characters under \f(CW\*(C`/i\*(C'\fR. For example,
430 \&\f(CW\*(C`LATIN SMALL LIGATURE FI\*(C'\fR should match the sequence \f(CW\*(C`fi\*(C'\fR. Perl is not
431 currently able to do this when the multiple characters are in the pattern and
432 are split between groupings, or when one or more are quantified. Thus
433 .Sp
434 .Vb 3
435 \& "\eN{LATIN SMALL LIGATURE FI}" =~ /fi/i; # Matches
436 \& "\eN{LATIN SMALL LIGATURE FI}" =~ /[fi][fi]/i; # Doesn\*(Aqt match!
437 \& "\eN{LATIN SMALL LIGATURE FI}" =~ /fi*/i; # Doesn\*(Aqt match!
438 \&
439 \& # The below doesn\*(Aqt match, and it isn\*(Aqt clear what $1 and $2 would
440 \& # be even if it did!!
441 \& "\eN{LATIN SMALL LIGATURE FI}" =~ /(f)(i)/i; # Doesn\*(Aqt match!
442 .Ve
443 .Sp
444 Perl doesn't match multiple characters in a bracketed
445 character class unless the character that maps to them is explicitly
446 mentioned, and it doesn't match them at all if the character class is
447 inverted, which otherwise could be highly confusing. See
448 "Bracketed Character Classes" in perlrecharclass, and
449 "Negation" in perlrecharclass.
450 .ie n .IP "\fR\fB""x""\fR\fB\fR and \fB\fR\fB""xx""\fR\fB\fR" 4
451 .el .IP "\fR\f(CBx\fR\fB\fR and \fB\fR\f(CBxx\fR\fB\fR" 4
452 .IX Xref " x"
453 .IX Item "x and xx"
454 Extend your pattern's legibility by permitting whitespace and comments.
455 Details in "/x and /xx"
456 .ie n .IP "\fR\fB""p""\fR\fB\fR" 4
457 .el .IP \fR\f(CBp\fR\fB\fR 4
458 .IX Xref " p regex, preserve regexp, preserve"
459 .IX Item "p"
460 Preserve the string matched such that \f(CW\*(C`${^PREMATCH}\*(C'\fR, \f(CW\*(C`${^MATCH}\*(C'\fR, and
461 \&\f(CW\*(C`${^POSTMATCH}\*(C'\fR are available for use after matching.
462 .Sp
463 In Perl 5.20 and higher this is ignored. Due to a new copy-on-write
464 mechanism, \f(CW\*(C`${^PREMATCH}\*(C'\fR, \f(CW\*(C`${^MATCH}\*(C'\fR, and \f(CW\*(C`${^POSTMATCH}\*(C'\fR will be available
465 after the match regardless of the modifier.
466 .ie n .IP "\fR\fB""a""\fR\fB\fR, \fB\fR\fB""d""\fR\fB\fR, \fB\fR\fB""l""\fR\fB\fR, and \fB\fR\fB""u""\fR\fB\fR" 4
467 .el .IP "\fR\f(CBa\fR\fB\fR, \fB\fR\f(CBd\fR\fB\fR, \fB\fR\f(CBl\fR\fB\fR, and \fB\fR\f(CBu\fR\fB\fR" 4
468 .IX Xref " a d l u"
469 .IX Item "a, d, l, and u"
470 These modifiers, all new in 5.14, affect which character-set rules
471 (Unicode, \fIetc\fR.) are used, as described below in
472 "Character set modifiers".
473 .ie n .IP "\fR\fB""n""\fR\fB\fR" 4
474 .el .IP \fR\f(CBn\fR\fB\fR 4
475 .IX Xref " n regex, non-capture regexp, non-capture regular expression, non-capture"
476 .IX Item "n"
477 Prevent the grouping metacharacters \f(CW\*(C`()\*(C'\fR from capturing. This modifier,
478 new in 5.22, will stop \f(CW$1\fR, \f(CW$2\fR, \fIetc\fR... from being filled in.
479 .Sp
480 .Vb 2
481 \& "hello" =~ /(hi|hello)/; # $1 is "hello"
482 \& "hello" =~ /(hi|hello)/n; # $1 is undef
483 .Ve
484 .Sp
485 This is equivalent to putting \f(CW\*(C`?:\*(C'\fR at the beginning of every capturing group:
486 .Sp
487 .Vb 1
488 \& "hello" =~ /(?:hi|hello)/; # $1 is undef
489 .Ve
490 .Sp
491 \&\f(CW\*(C`/n\*(C'\fR can be negated on a per-group basis. Alternatively, named captures
492 may still be used.
493 .Sp
494 .Vb 3
495 \& "hello" =~ /(?\-n:(hi|hello))/n; # $1 is "hello"
496 \& "hello" =~ /(?<greet>hi|hello)/n; # $1 is "hello", $+{greet} is
497 \& # "hello"
498 .Ve
499 .IP "Other Modifiers" 4
500 .IX Item "Other Modifiers"
501 There are a number of flags that can be found at the end of regular
502 expression constructs that are \fInot\fR generic regular expression flags, but
503 apply to the operation being performed, like matching or substitution (\f(CW\*(C`m//\*(C'\fR
504 or \f(CW\*(C`s///\*(C'\fR respectively).
505 .Sp
506 Flags described further in
507 "Using regular expressions in Perl" in perlretut are:
508 .Sp
509 .Vb 2
510 \& c \- keep the current position during repeated matching
511 \& g \- globally match the pattern repeatedly in the string
512 .Ve
513 .Sp
514 Substitution-specific modifiers described in
515 "s/PATTERN/REPLACEMENT/msixpodualngcer" in perlop are:
516 .Sp
517 .Vb 4
518 \& e \- evaluate the right\-hand side as an expression
519 \& ee \- evaluate the right side as a string then eval the result
520 \& o \- pretend to optimize your code, but actually introduce bugs
521 \& r \- perform non\-destructive substitution and return the new value
522 .Ve
523 .PP
524 Regular expression modifiers are usually written in documentation
525 as \fIe.g.\fR, "the \f(CW\*(C`/x\*(C'\fR modifier", even though the delimiter
526 in question might not really be a slash. The modifiers \f(CW\*(C`/imnsxadlup\*(C'\fR
527 may also be embedded within the regular expression itself using
528 the \f(CW\*(C`(?...)\*(C'\fR construct, see "Extended Patterns" below.
529 .PP
530 \fIDetails on some modifiers\fR
531 .IX Subsection "Details on some modifiers"
532 .PP
533 Some of the modifiers require more explanation than given in the
534 "Overview" above.
535 .PP
536 \f(CW\*(C`/x\*(C'\fR and \f(CW\*(C`/xx\*(C'\fR
537 .IX Subsection "/x and /xx"
538 .PP
539 A single \f(CW\*(C`/x\*(C'\fR tells
540 the regular expression parser to ignore most whitespace that is neither
541 backslashed nor within a bracketed character class, nor within the characters
542 of a multi-character metapattern like \f(CW\*(C`(?i: ... )\*(C'\fR. You can use this to
543 break up your regular expression into more readable parts.
544 Also, the \f(CW"#"\fR character is treated as a metacharacter introducing a
545 comment that runs up to the pattern's closing delimiter, or to the end
546 of the current line if the pattern extends onto the next line. Hence,
547 this is very much like an ordinary Perl code comment. (You can include
548 the closing delimiter within the comment only if you precede it with a
549 backslash, so be careful!)
550 .PP
551 Use of \f(CW\*(C`/x\*(C'\fR means that if you want real
552 whitespace or \f(CW"#"\fR characters in the pattern (outside a bracketed character
553 class, which is unaffected by \f(CW\*(C`/x\*(C'\fR), then you'll either have to
554 escape them (using backslashes or \f(CW\*(C`\eQ...\eE\*(C'\fR) or encode them using octal,
555 hex, or \f(CW\*(C`\eN{}\*(C'\fR or \f(CW\*(C`\ep{name=...}\*(C'\fR escapes.
556 It is ineffective to try to continue a comment onto the next line by
557 escaping the \f(CW\*(C`\en\*(C'\fR with a backslash or \f(CW\*(C`\eQ\*(C'\fR.
558 .PP
559 You can use "(?#text)" to create a comment that ends earlier than the
560 end of the current line, but \f(CW\*(C`text\*(C'\fR also can't contain the closing
561 delimiter unless escaped with a backslash.
562 .PP
563 A common pitfall is to forget that \f(CW"#"\fR characters (outside a
564 bracketed character class) begin a comment under \f(CW\*(C`/x\*(C'\fR and are not
565 matched literally. Just keep that in mind when trying to puzzle out why
566 a particular \f(CW\*(C`/x\*(C'\fR pattern isn't working as expected.
567 Inside a bracketed character class, \f(CW"#"\fR retains its non-special,
568 literal meaning.
569 .PP
570 Starting in Perl v5.26, if the modifier has a second \f(CW"x"\fR within it,
571 the effect of a single \f(CW\*(C`/x\*(C'\fR is increased. The only difference is that
572 inside bracketed character classes, non-escaped (by a backslash) SPACE
573 and TAB characters are not added to the class, and hence can be inserted
574 to make the classes more readable:
575 .PP
576 .Vb 2
577 \& / [d\-e g\-i 3\-7]/xx
578 \& /[ ! @ " # $ % ^ & * () = ? <> \*(Aq ]/xx
579 .Ve
580 .PP
581 may be easier to grasp than the squashed equivalents
582 .PP
583 .Vb 2
584 \& /[d\-eg\-i3\-7]/
585 \& /[!@"#$%^&*()=?<>\*(Aq]/
586 .Ve
587 .PP
588 Note that this unfortunately doesn't mean that your bracketed classes
589 can contain comments or extend over multiple lines. A \f(CW\*(C`#\*(C'\fR inside a
590 character class is still just a literal \f(CW\*(C`#\*(C'\fR, and doesn't introduce a
591 comment. And, unless the closing bracket is on the same line as the
592 opening one, the newline character (and everything on the next line(s)
593 until terminated by a \f(CW\*(C`]\*(C'\fR will be part of the class, just as if you'd
594 written \f(CW\*(C`\en\*(C'\fR.
595 .PP
596 Taken together, these features go a long way towards
597 making Perl's regular expressions more readable. Here's an example:
598 .PP
599 .Vb 6
600 \& # Delete (most) C comments.
601 \& $program =~ s {
602 \& /\e* # Match the opening delimiter.
603 \& .*? # Match a minimal number of characters.
604 \& \e*/ # Match the closing delimiter.
605 \& } []gsx;
606 .Ve
607 .PP
608 Note that anything inside
609 a \f(CW\*(C`\eQ...\eE\*(C'\fR stays unaffected by \f(CW\*(C`/x\*(C'\fR. And note that \f(CW\*(C`/x\*(C'\fR doesn't affect
610 space interpretation within a single multi-character construct. For
611 example \f(CW\*(C`(?:...)\*(C'\fR can't have a space between the \f(CW"("\fR,
612 \&\f(CW"?"\fR, and \f(CW":"\fR. Within any delimiters for such a construct, allowed
613 spaces are not affected by \f(CW\*(C`/x\*(C'\fR, and depend on the construct. For
614 example, all constructs using curly braces as delimiters, such as
615 \&\f(CW\*(C`\ex{...}\*(C'\fR can have blanks within but adjacent to the braces, but not
616 elsewhere, and no non-blank space characters. An exception are Unicode
617 properties which follow Unicode rules, for which see
618 "Properties accessible through \ep{} and \eP{}" in perluniprops.
619 .IX Xref " x"
620 .PP
621 The set of characters that are deemed whitespace are those that Unicode
622 calls "Pattern White Space", namely:
623 .PP
624 .Vb 11
625 \& U+0009 CHARACTER TABULATION
626 \& U+000A LINE FEED
627 \& U+000B LINE TABULATION
628 \& U+000C FORM FEED
629 \& U+000D CARRIAGE RETURN
630 \& U+0020 SPACE
631 \& U+0085 NEXT LINE
632 \& U+200E LEFT\-TO\-RIGHT MARK
633 \& U+200F RIGHT\-TO\-LEFT MARK
634 \& U+2028 LINE SEPARATOR
635 \& U+2029 PARAGRAPH SEPARATOR
636 .Ve
637 .PP
638 Character set modifiers
639 .IX Subsection "Character set modifiers"
640 .PP
641 \&\f(CW\*(C`/d\*(C'\fR, \f(CW\*(C`/u\*(C'\fR, \f(CW\*(C`/a\*(C'\fR, and \f(CW\*(C`/l\*(C'\fR, available starting in 5.14, are called
642 the character set modifiers; they affect the character set rules
643 used for the regular expression.
644 .PP
645 The \f(CW\*(C`/d\*(C'\fR, \f(CW\*(C`/u\*(C'\fR, and \f(CW\*(C`/l\*(C'\fR modifiers are not likely to be of much use
646 to you, and so you need not worry about them very much. They exist for
647 Perl's internal use, so that complex regular expression data structures
648 can be automatically serialized and later exactly reconstituted,
649 including all their nuances. But, since Perl can't keep a secret, and
650 there may be rare instances where they are useful, they are documented
651 here.
652 .PP
653 The \f(CW\*(C`/a\*(C'\fR modifier, on the other hand, may be useful. Its purpose is to
654 allow code that is to work mostly on ASCII data to not have to concern
655 itself with Unicode.
656 .PP
657 Briefly, \f(CW\*(C`/l\*(C'\fR sets the character set to that of whatever \fBL\fRocale is in
658 effect at the time of the execution of the pattern match.
659 .PP
660 \&\f(CW\*(C`/u\*(C'\fR sets the character set to \fBU\fRnicode.
661 .PP
662 \&\f(CW\*(C`/a\*(C'\fR also sets the character set to Unicode, BUT adds several
663 restrictions for \fBA\fRSCII-safe matching.
664 .PP
665 \&\f(CW\*(C`/d\*(C'\fR is the old, problematic, pre\-5.14 \fBD\fRefault character set
666 behavior. Its only use is to force that old behavior.
667 .PP
668 At any given time, exactly one of these modifiers is in effect. Their
669 existence allows Perl to keep the originally compiled behavior of a
670 regular expression, regardless of what rules are in effect when it is
671 actually executed. And if it is interpolated into a larger regex, the
672 original's rules continue to apply to it, and don't affect the other
673 parts.
674 .PP
675 The \f(CW\*(C`/l\*(C'\fR and \f(CW\*(C`/u\*(C'\fR modifiers are automatically selected for
676 regular expressions compiled within the scope of various pragmas,
677 and we recommend that in general, you use those pragmas instead of
678 specifying these modifiers explicitly. For one thing, the modifiers
679 affect only pattern matching, and do not extend to even any replacement
680 done, whereas using the pragmas gives consistent results for all
681 appropriate operations within their scopes. For example,
682 .PP
683 .Vb 1
684 \& s/foo/\eUbar/il
685 .Ve
686 .PP
687 will match "foo" using the locale's rules for case-insensitive matching,
688 but the \f(CW\*(C`/l\*(C'\fR does not affect how the \f(CW\*(C`\eU\*(C'\fR operates. Most likely you
689 want both of them to use locale rules. To do this, instead compile the
690 regular expression within the scope of \f(CW\*(C`use locale\*(C'\fR. This both
691 implicitly adds the \f(CW\*(C`/l\*(C'\fR, and applies locale rules to the \f(CW\*(C`\eU\*(C'\fR. The
692 lesson is to \f(CW\*(C`use locale\*(C'\fR, and not \f(CW\*(C`/l\*(C'\fR explicitly.
693 .PP
694 Similarly, it would be better to use \f(CW\*(C`use feature \*(Aqunicode_strings\*(Aq\*(C'\fR
695 instead of,
696 .PP
697 .Vb 1
698 \& s/foo/\eLbar/iu
699 .Ve
700 .PP
701 to get Unicode rules, as the \f(CW\*(C`\eL\*(C'\fR in the former (but not necessarily
702 the latter) would also use Unicode rules.
703 .PP
704 More detail on each of the modifiers follows. Most likely you don't
705 need to know this detail for \f(CW\*(C`/l\*(C'\fR, \f(CW\*(C`/u\*(C'\fR, and \f(CW\*(C`/d\*(C'\fR, and can skip ahead
706 to /a.
707 .PP
708 /l
709 .IX Subsection "/l"
710 .PP
711 means to use the current locale's rules (see perllocale) when pattern
712 matching. For example, \f(CW\*(C`\ew\*(C'\fR will match the "word" characters of that
713 locale, and \f(CW"/i"\fR case-insensitive matching will match according to
714 the locale's case folding rules. The locale used will be the one in
715 effect at the time of execution of the pattern match. This may not be
716 the same as the compilation-time locale, and can differ from one match
717 to another if there is an intervening call of the
718 \&\fBsetlocale()\fR function.
719 .PP
720 Prior to v5.20, Perl did not support multi-byte locales. Starting then,
721 UTF\-8 locales are supported. No other multi byte locales are ever
722 likely to be supported. However, in all locales, one can have code
723 points above 255 and these will always be treated as Unicode no matter
724 what locale is in effect.
725 .PP
726 Under Unicode rules, there are a few case-insensitive matches that cross
727 the 255/256 boundary. Except for UTF\-8 locales in Perls v5.20 and
728 later, these are disallowed under \f(CW\*(C`/l\*(C'\fR. For example, 0xFF (on ASCII
729 platforms) does not caselessly match the character at 0x178, \f(CW\*(C`LATIN
730 CAPITAL LETTER Y WITH DIAERESIS\*(C'\fR, because 0xFF may not be \f(CW\*(C`LATIN SMALL
731 LETTER Y WITH DIAERESIS\*(C'\fR in the current locale, and Perl has no way of
732 knowing if that character even exists in the locale, much less what code
733 point it is.
734 .PP
735 In a UTF\-8 locale in v5.20 and later, the only visible difference
736 between locale and non-locale in regular expressions should be tainting,
737 if your perl supports taint checking (see perlsec).
738 .PP
739 This modifier may be specified to be the default by \f(CW\*(C`use locale\*(C'\fR, but
740 see "Which character set modifier is in effect?".
741 .IX Xref " l"
742 .PP
743 /u
744 .IX Subsection "/u"
745 .PP
746 means to use Unicode rules when pattern matching. On ASCII platforms,
747 this means that the code points between 128 and 255 take on their
748 Latin\-1 (ISO\-8859\-1) meanings (which are the same as Unicode's).
749 (Otherwise Perl considers their meanings to be undefined.) Thus,
750 under this modifier, the ASCII platform effectively becomes a Unicode
751 platform; and hence, for example, \f(CW\*(C`\ew\*(C'\fR will match any of the more than
752 100_000 word characters in Unicode.
753 .PP
754 Unlike most locales, which are specific to a language and country pair,
755 Unicode classifies all the characters that are letters \fIsomewhere\fR in
756 the world as
757 \&\f(CW\*(C`\ew\*(C'\fR. For example, your locale might not think that \f(CW\*(C`LATIN SMALL
758 LETTER ETH\*(C'\fR is a letter (unless you happen to speak Icelandic), but
759 Unicode does. Similarly, all the characters that are decimal digits
760 somewhere in the world will match \f(CW\*(C`\ed\*(C'\fR; this is hundreds, not 10,
761 possible matches. And some of those digits look like some of the 10
762 ASCII digits, but mean a different number, so a human could easily think
763 a number is a different quantity than it really is. For example,
764 \&\f(CW\*(C`BENGALI DIGIT FOUR\*(C'\fR (U+09EA) looks very much like an
765 \&\f(CW\*(C`ASCII DIGIT EIGHT\*(C'\fR (U+0038), and \f(CW\*(C`LEPCHA DIGIT SIX\*(C'\fR (U+1C46) looks
766 very much like an \f(CW\*(C`ASCII DIGIT FIVE\*(C'\fR (U+0035). And, \f(CW\*(C`\ed+\*(C'\fR, may match
767 strings of digits that are a mixture from different writing systems,
768 creating a security issue. A fraudulent website, for example, could
769 display the price of something using U+1C46, and it would appear to the
770 user that something cost 500 units, but it really costs 600. A browser
771 that enforced script runs ("Script Runs") would prevent that
772 fraudulent display. "\fBnum()\fR" in Unicode::UCD can also be used to sort this
773 out. Or the \f(CW\*(C`/a\*(C'\fR modifier can be used to force \f(CW\*(C`\ed\*(C'\fR to match just the
774 ASCII 0 through 9.
775 .PP
776 Also, under this modifier, case-insensitive matching works on the full
777 set of Unicode
778 characters. The \f(CW\*(C`KELVIN SIGN\*(C'\fR, for example matches the letters "k" and
779 "K"; and \f(CW\*(C`LATIN SMALL LIGATURE FF\*(C'\fR matches the sequence "ff", which,
780 if you're not prepared, might make it look like a hexadecimal constant,
781 presenting another potential security issue. See
782 <https://unicode.org/reports/tr36> for a detailed discussion of Unicode
783 security issues.
784 .PP
785 This modifier may be specified to be the default by \f(CW\*(C`use feature
786 \&\*(Aqunicode_strings\*(C'\fR, \f(CW\*(C`use locale \*(Aq:not_characters\*(Aq\*(C'\fR, or
787 \&\f(CW\*(C`use v5.12\*(C'\fR (or higher),
788 but see "Which character set modifier is in effect?".
789 .IX Xref " u"
790 .PP
791 /d
792 .IX Subsection "/d"
793 .PP
794 \&\fBIMPORTANT:\fR Because of the unpredictable behaviors this
795 modifier causes, only use it to maintain weird backward compatibilities.
796 Use the
797 \&\f(CW\*(C`unicode_strings\*(C'\fR
798 feature
799 in new code to avoid inadvertently enabling this modifier by default.
800 .PP
801 What does this modifier do? It "Depends"!
802 .PP
803 This modifier means to use platform-native matching rules
804 except when there is cause to use Unicode rules instead, as follows:
805 .IP 1. 4
806 the target string's UTF8 flag
807 (see below) is set; or
808 .IP 2. 4
809 the pattern's UTF8 flag
810 (see below) is set; or
811 .IP 3. 4
812 the pattern explicitly mentions a code point that is above 255 (say by
813 \&\f(CW\*(C`\ex{100}\*(C'\fR); or
814 .IP 4. 4
815 the pattern uses a Unicode name (\f(CW\*(C`\eN{...}\*(C'\fR); or
816 .IP 5. 4
817 the pattern uses a Unicode property (\f(CW\*(C`\ep{...}\*(C'\fR or \f(CW\*(C`\eP{...}\*(C'\fR); or
818 .IP 6. 4
819 the pattern uses a Unicode break (\f(CW\*(C`\eb{...}\*(C'\fR or \f(CW\*(C`\eB{...}\*(C'\fR); or
820 .IP 7. 4
821 the pattern uses \f(CW"(?[ ])"\fR
822 .IP 8. 4
823 the pattern uses \f(CW\*(C`(*script_run: ...)\*(C'\fR
824 .PP
825 Regarding the "UTF8 flag" references above: normally Perl applications
826 shouldn't think about that flag. It's part of Perl's internals,
827 so it can change whenever Perl wants. \f(CW\*(C`/d\*(C'\fR may thus cause unpredictable
828 results. See "The "Unicode Bug"" in perlunicode. This bug
829 has become rather infamous, leading to yet other (without swearing) names
830 for this modifier like "Dicey" and "Dodgy".
831 .PP
832 Here are some examples of how that works on an ASCII platform:
833 .PP
834 .Vb 3
835 \& $str = "\exDF"; #
836 \& utf8::downgrade($str); # $str is not UTF8\-flagged.
837 \& $str =~ /^\ew/; # No match, since no UTF8 flag.
838 \&
839 \& $str .= "\ex{0e0b}"; # Now $str is UTF8\-flagged.
840 \& $str =~ /^\ew/; # Match! $str is now UTF8\-flagged.
841 \& chop $str;
842 \& $str =~ /^\ew/; # Still a match! $str retains its UTF8 flag.
843 .Ve
844 .PP
845 Under Perl's default configuration this modifier is automatically
846 selected by default when none of the others are, so yet another name
847 for it (unfortunately) is "Default".
848 .PP
849 Whenever you can, use the
850 \&\f(CW\*(C`unicode_strings\*(C'\fR
851 to cause to be the default instead.
852 .IX Xref " u"
853 .PP
854 /a (and /aa)
855 .IX Subsection "/a (and /aa)"
856 .PP
857 This modifier stands for ASCII-restrict (or ASCII-safe). This modifier
858 may be doubled-up to increase its effect.
859 .PP
860 When it appears singly, it causes the sequences \f(CW\*(C`\ed\*(C'\fR, \f(CW\*(C`\es\*(C'\fR, \f(CW\*(C`\ew\*(C'\fR, and
861 the Posix character classes to match only in the ASCII range. They thus
862 revert to their pre\-5.6, pre-Unicode meanings. Under \f(CW\*(C`/a\*(C'\fR, \f(CW\*(C`\ed\*(C'\fR
863 always means precisely the digits \f(CW"0"\fR to \f(CW"9"\fR; \f(CW\*(C`\es\*(C'\fR means the five
864 characters \f(CW\*(C`[ \ef\en\er\et]\*(C'\fR, and starting in Perl v5.18, the vertical tab;
865 \&\f(CW\*(C`\ew\*(C'\fR means the 63 characters
866 \&\f(CW\*(C`[A\-Za\-z0\-9_]\*(C'\fR; and likewise, all the Posix classes such as
867 \&\f(CW\*(C`[[:print:]]\*(C'\fR match only the appropriate ASCII-range characters.
868 .PP
869 This modifier is useful for people who only incidentally use Unicode,
870 and who do not wish to be burdened with its complexities and security
871 concerns.
872 .PP
873 With \f(CW\*(C`/a\*(C'\fR, one can write \f(CW\*(C`\ed\*(C'\fR with confidence that it will only match
874 ASCII characters, and should the need arise to match beyond ASCII, you
875 can instead use \f(CW\*(C`\ep{Digit}\*(C'\fR (or \f(CW\*(C`\ep{Word}\*(C'\fR for \f(CW\*(C`\ew\*(C'\fR). There are
876 similar \f(CW\*(C`\ep{...}\*(C'\fR constructs that can match beyond ASCII both white
877 space (see "Whitespace" in perlrecharclass), and Posix classes (see
878 "POSIX Character Classes" in perlrecharclass). Thus, this modifier
879 doesn't mean you can't use Unicode, it means that to get Unicode
880 matching you must explicitly use a construct (\f(CW\*(C`\ep{}\*(C'\fR, \f(CW\*(C`\eP{}\*(C'\fR) that
881 signals Unicode.
882 .PP
883 As you would expect, this modifier causes, for example, \f(CW\*(C`\eD\*(C'\fR to mean
884 the same thing as \f(CW\*(C`[^0\-9]\*(C'\fR; in fact, all non-ASCII characters match
885 \&\f(CW\*(C`\eD\*(C'\fR, \f(CW\*(C`\eS\*(C'\fR, and \f(CW\*(C`\eW\*(C'\fR. \f(CW\*(C`\eb\*(C'\fR still means to match at the boundary
886 between \f(CW\*(C`\ew\*(C'\fR and \f(CW\*(C`\eW\*(C'\fR, using the \f(CW\*(C`/a\*(C'\fR definitions of them (similarly
887 for \f(CW\*(C`\eB\*(C'\fR).
888 .PP
889 Otherwise, \f(CW\*(C`/a\*(C'\fR behaves like the \f(CW\*(C`/u\*(C'\fR modifier, in that
890 case-insensitive matching uses Unicode rules; for example, "k" will
891 match the Unicode \f(CW\*(C`\eN{KELVIN SIGN}\*(C'\fR under \f(CW\*(C`/i\*(C'\fR matching, and code
892 points in the Latin1 range, above ASCII will have Unicode rules when it
893 comes to case-insensitive matching.
894 .PP
895 To forbid ASCII/non\-ASCII matches (like "k" with \f(CW\*(C`\eN{KELVIN SIGN}\*(C'\fR),
896 specify the \f(CW"a"\fR twice, for example \f(CW\*(C`/aai\*(C'\fR or \f(CW\*(C`/aia\*(C'\fR. (The first
897 occurrence of \f(CW"a"\fR restricts the \f(CW\*(C`\ed\*(C'\fR, \fIetc\fR., and the second occurrence
898 adds the \f(CW\*(C`/i\*(C'\fR restrictions.) But, note that code points outside the
899 ASCII range will use Unicode rules for \f(CW\*(C`/i\*(C'\fR matching, so the modifier
900 doesn't really restrict things to just ASCII; it just forbids the
901 intermixing of ASCII and non-ASCII.
902 .PP
903 To summarize, this modifier provides protection for applications that
904 don't wish to be exposed to all of Unicode. Specifying it twice
905 gives added protection.
906 .PP
907 This modifier may be specified to be the default by \f(CW\*(C`use re \*(Aq/a\*(Aq\*(C'\fR
908 or \f(CW\*(C`use re \*(Aq/aa\*(Aq\*(C'\fR. If you do so, you may actually have occasion to use
909 the \f(CW\*(C`/u\*(C'\fR modifier explicitly if there are a few regular expressions
910 where you do want full Unicode rules (but even here, it's best if
911 everything were under feature \f(CW"unicode_strings"\fR, along with the
912 \&\f(CW\*(C`use re \*(Aq/aa\*(Aq\*(C'\fR). Also see "Which character set modifier is in
913 effect?".
914 .IX Xref " a aa"
915 .PP
916 Which character set modifier is in effect?
917 .IX Subsection "Which character set modifier is in effect?"
918 .PP
919 Which of these modifiers is in effect at any given point in a regular
920 expression depends on a fairly complex set of interactions. These have
921 been designed so that in general you don't have to worry about it, but
922 this section gives the gory details. As
923 explained below in "Extended Patterns" it is possible to explicitly
924 specify modifiers that apply only to portions of a regular expression.
925 The innermost always has priority over any outer ones, and one applying
926 to the whole expression has priority over any of the default settings that are
927 described in the remainder of this section.
928 .PP
929 The \f(CW\*(C`use re \*(Aq/foo\*(Aq\*(C'\fR pragma can be used to set
930 default modifiers (including these) for regular expressions compiled
931 within its scope. This pragma has precedence over the other pragmas
932 listed below that also change the defaults. Note that the /x modifier does
933 NOT affect \f(CW\*(C`split STR\*(C'\fR patterns.
934 .PP
935 Otherwise, \f(CW\*(C`use locale\*(C'\fR sets the default modifier to \f(CW\*(C`/l\*(C'\fR;
936 and \f(CW\*(C`use feature \*(Aqunicode_strings\*(C'\fR, or
937 \&\f(CW\*(C`use v5.12\*(C'\fR (or higher) set the default to
938 \&\f(CW\*(C`/u\*(C'\fR when not in the same scope as either \f(CW\*(C`use locale\*(C'\fR
939 or \f(CW\*(C`use bytes\*(C'\fR.
940 (\f(CW\*(C`use locale \*(Aq:not_characters\*(Aq\*(C'\fR also
941 sets the default to \f(CW\*(C`/u\*(C'\fR, overriding any plain \f(CW\*(C`use locale\*(C'\fR.)
942 Unlike the mechanisms mentioned above, these
943 affect operations besides regular expressions pattern matching, and so
944 give more consistent results with other operators, including using
945 \&\f(CW\*(C`\eU\*(C'\fR, \f(CW\*(C`\el\*(C'\fR, \fIetc\fR. in substitution replacements.
946 .PP
947 If none of the above apply, for backwards compatibility reasons, the
948 \&\f(CW\*(C`/d\*(C'\fR modifier is the one in effect by default. As this can lead to
949 unexpected results, it is best to specify which other rule set should be
950 used.
951 .PP
952 Character set modifier behavior prior to Perl 5.14
953 .IX Subsection "Character set modifier behavior prior to Perl 5.14"
954 .PP
955 Prior to 5.14, there were no explicit modifiers, but \f(CW\*(C`/l\*(C'\fR was implied
956 for regexes compiled within the scope of \f(CW\*(C`use locale\*(C'\fR, and \f(CW\*(C`/d\*(C'\fR was
957 implied otherwise. However, interpolating a regex into a larger regex
958 would ignore the original compilation in favor of whatever was in effect
959 at the time of the second compilation. There were a number of
960 inconsistencies (bugs) with the \f(CW\*(C`/d\*(C'\fR modifier, where Unicode rules
961 would be used when inappropriate, and vice versa. \f(CW\*(C`\ep{}\*(C'\fR did not imply
962 Unicode rules, and neither did all occurrences of \f(CW\*(C`\eN{}\*(C'\fR, until 5.12.
963 .SS "Regular Expressions"
964 .IX Subsection "Regular Expressions"
965 \fIQuantifiers\fR
966 .IX Subsection "Quantifiers"
967 .PP
968 Quantifiers are used when a particular portion of a pattern needs to
969 match a certain number (or numbers) of times. If there isn't a
970 quantifier the number of times to match is exactly one. The following
971 standard quantifiers are recognized:
972 .IX Xref "metacharacter quantifier * + ? {n} {n,} {n,m}"
973 .PP
974 .Vb 7
975 \& * Match 0 or more times
976 \& + Match 1 or more times
977 \& ? Match 1 or 0 times
978 \& {n} Match exactly n times
979 \& {n,} Match at least n times
980 \& {,n} Match at most n times
981 \& {n,m} Match at least n but not more than m times
982 .Ve
983 .PP
984 (If a non-escaped curly bracket occurs in a context other than one of
985 the quantifiers listed above, where it does not form part of a
986 backslashed sequence like \f(CW\*(C`\ex{...}\*(C'\fR, it is either a fatal syntax error,
987 or treated as a regular character, generally with a deprecation warning
988 raised. To escape it, you can precede it with a backslash (\f(CW"\e{"\fR) or
989 enclose it within square brackets (\f(CW"[{]"\fR).
990 This change will allow for future syntax extensions (like making the
991 lower bound of a quantifier optional), and better error checking of
992 quantifiers).
993 .PP
994 The \f(CW"*"\fR quantifier is equivalent to \f(CW\*(C`{0,}\*(C'\fR, the \f(CW"+"\fR
995 quantifier to \f(CW\*(C`{1,}\*(C'\fR, and the \f(CW"?"\fR quantifier to \f(CW\*(C`{0,1}\*(C'\fR. \fIn\fR and \fIm\fR are limited
996 to non-negative integral values less than a preset limit defined when perl is built.
997 This is usually 65534 on the most common platforms. The actual limit can
998 be seen in the error message generated by code such as this:
999 .PP
1000 .Vb 1
1001 \& $_ **= $_ , / {$_} / for 2 .. 42;
1002 .Ve
1003 .PP
1004 By default, a quantified subpattern is "greedy", that is, it will match as
1005 many times as possible (given a particular starting location) while still
1006 allowing the rest of the pattern to match. If you want it to match the
1007 minimum number of times possible, follow the quantifier with a \f(CW"?"\fR. Note
1008 that the meanings don't change, just the "greediness":
1009 .IX Xref "metacharacter greedy greediness ? *? +? ?? {n}? {n,}? {,n}? {n,m}?"
1010 .PP
1011 .Vb 7
1012 \& *? Match 0 or more times, not greedily
1013 \& +? Match 1 or more times, not greedily
1014 \& ?? Match 0 or 1 time, not greedily
1015 \& {n}? Match exactly n times, not greedily (redundant)
1016 \& {n,}? Match at least n times, not greedily
1017 \& {,n}? Match at most n times, not greedily
1018 \& {n,m}? Match at least n but not more than m times, not greedily
1019 .Ve
1020 .PP
1021 Normally when a quantified subpattern does not allow the rest of the
1022 overall pattern to match, Perl will backtrack. However, this behaviour is
1023 sometimes undesirable. Thus Perl provides the "possessive" quantifier form
1024 as well.
1025 .PP
1026 .Vb 7
1027 \& *+ Match 0 or more times and give nothing back
1028 \& ++ Match 1 or more times and give nothing back
1029 \& ?+ Match 0 or 1 time and give nothing back
1030 \& {n}+ Match exactly n times and give nothing back (redundant)
1031 \& {n,}+ Match at least n times and give nothing back
1032 \& {,n}+ Match at most n times and give nothing back
1033 \& {n,m}+ Match at least n but not more than m times and give nothing back
1034 .Ve
1035 .PP
1036 For instance,
1037 .PP
1038 .Vb 1
1039 \& \*(Aqaaaa\*(Aq =~ /a++a/
1040 .Ve
1041 .PP
1042 will never match, as the \f(CW\*(C`a++\*(C'\fR will gobble up all the \f(CW"a"\fR's in the
1043 string and won't leave any for the remaining part of the pattern. This
1044 feature can be extremely useful to give perl hints about where it
1045 shouldn't backtrack. For instance, the typical "match a double-quoted
1046 string" problem can be most efficiently performed when written as:
1047 .PP
1048 .Vb 1
1049 \& /"(?:[^"\e\e]++|\e\e.)*+"/
1050 .Ve
1051 .PP
1052 as we know that if the final quote does not match, backtracking will not
1053 help. See the independent subexpression
1054 \&\f(CW"(?>\fR\f(CIpattern\fR\f(CW)"\fR for more details;
1055 possessive quantifiers are just syntactic sugar for that construct. For
1056 instance the above example could also be written as follows:
1057 .PP
1058 .Vb 1
1059 \& /"(?>(?:(?>[^"\e\e]+)|\e\e.)*)"/
1060 .Ve
1061 .PP
1062 Note that the possessive quantifier modifier can not be combined
1063 with the non-greedy modifier. This is because it would make no sense.
1064 Consider the follow equivalency table:
1065 .PP
1066 .Vb 5
1067 \& Illegal Legal
1068 \& \-\-\-\-\-\-\-\-\-\-\-\- \-\-\-\-\-\-
1069 \& X??+ X{0}
1070 \& X+?+ X{1}
1071 \& X{min,max}?+ X{min}
1072 .Ve
1073 .PP
1074 \fIEscape sequences\fR
1075 .IX Subsection "Escape sequences"
1076 .PP
1077 Because patterns are processed as double-quoted strings, the following
1078 also work:
1079 .PP
1080 .Vb 10
1081 \& \et tab (HT, TAB)
1082 \& \en newline (LF, NL)
1083 \& \er return (CR)
1084 \& \ef form feed (FF)
1085 \& \ea alarm (bell) (BEL)
1086 \& \ee escape (think troff) (ESC)
1087 \& \ecK control char (example: VT)
1088 \& \ex{}, \ex00 character whose ordinal is the given hexadecimal number
1089 \& \eN{name} named Unicode character or character sequence
1090 \& \eN{U+263D} Unicode character (example: FIRST QUARTER MOON)
1091 \& \eo{}, \e000 character whose ordinal is the given octal number
1092 \& \el lowercase next char (think vi)
1093 \& \eu uppercase next char (think vi)
1094 \& \eL lowercase until \eE (think vi)
1095 \& \eU uppercase until \eE (think vi)
1096 \& \eQ quote (disable) pattern metacharacters until \eE
1097 \& \eE end either case modification or quoted section, think vi
1098 .Ve
1099 .PP
1100 Details are in "Quote and Quote-like Operators" in perlop.
1101 .PP
1102 \fICharacter Classes and other Special Escapes\fR
1103 .IX Subsection "Character Classes and other Special Escapes"
1104 .PP
1105 In addition, Perl defines the following:
1106 .IX Xref "\\g \\k \\K backreference"
1107 .PP
1108 .Vb 10
1109 \& Sequence Note Description
1110 \& [...] [1] Match a character according to the rules of the
1111 \& bracketed character class defined by the "...".
1112 \& Example: [a\-z] matches "a" or "b" or "c" ... or "z"
1113 \& [[:...:]] [2] Match a character according to the rules of the POSIX
1114 \& character class "..." within the outer bracketed
1115 \& character class. Example: [[:upper:]] matches any
1116 \& uppercase character.
1117 \& (?[...]) [8] Extended bracketed character class
1118 \& \ew [3] Match a "word" character (alphanumeric plus "_", plus
1119 \& other connector punctuation chars plus Unicode
1120 \& marks)
1121 \& \eW [3] Match a non\-"word" character
1122 \& \es [3] Match a whitespace character
1123 \& \eS [3] Match a non\-whitespace character
1124 \& \ed [3] Match a decimal digit character
1125 \& \eD [3] Match a non\-digit character
1126 \& \epP [3] Match P, named property. Use \ep{Prop} for longer names
1127 \& \ePP [3] Match non\-P
1128 \& \eX [4] Match Unicode "eXtended grapheme cluster"
1129 \& \e1 [5] Backreference to a specific capture group or buffer.
1130 \& \*(Aq1\*(Aq may actually be any positive integer.
1131 \& \eg1 [5] Backreference to a specific or previous group,
1132 \& \eg{\-1} [5] The number may be negative indicating a relative
1133 \& previous group and may optionally be wrapped in
1134 \& curly brackets for safer parsing.
1135 \& \eg{name} [5] Named backreference
1136 \& \ek<name> [5] Named backreference
1137 \& \ek\*(Aqname\*(Aq [5] Named backreference
1138 \& \ek{name} [5] Named backreference
1139 \& \eK [6] Keep the stuff left of the \eK, don\*(Aqt include it in $&
1140 \& \eN [7] Any character but \en. Not affected by /s modifier
1141 \& \ev [3] Vertical whitespace
1142 \& \eV [3] Not vertical whitespace
1143 \& \eh [3] Horizontal whitespace
1144 \& \eH [3] Not horizontal whitespace
1145 \& \eR [4] Linebreak
1146 .Ve
1147 .IP [1] 4
1148 .IX Item "[1]"
1149 See "Bracketed Character Classes" in perlrecharclass for details.
1150 .IP [2] 4
1151 .IX Item "[2]"
1152 See "POSIX Character Classes" in perlrecharclass for details.
1153 .IP [3] 4
1154 .IX Item "[3]"
1155 See "Unicode Character Properties" in perlunicode for details
1156 .IP [4] 4
1157 .IX Item "[4]"
1158 See "Misc" in perlrebackslash for details.
1159 .IP [5] 4
1160 .IX Item "[5]"
1161 See "Capture groups" below for details.
1162 .IP [6] 4
1163 .IX Item "[6]"
1164 See "Extended Patterns" below for details.
1165 .IP [7] 4
1166 .IX Item "[7]"
1167 Note that \f(CW\*(C`\eN\*(C'\fR has two meanings. When of the form \f(CW\*(C`\eN{\fR\f(CINAME\fR\f(CW}\*(C'\fR, it
1168 matches the character or character sequence whose name is \fINAME\fR; and
1169 similarly
1170 when of the form \f(CW\*(C`\eN{U+\fR\f(CIhex\fR\f(CW}\*(C'\fR, it matches the character whose Unicode
1171 code point is \fIhex\fR. Otherwise it matches any character but \f(CW\*(C`\en\*(C'\fR.
1172 .IP [8] 4
1173 .IX Item "[8]"
1174 See "Extended Bracketed Character Classes" in perlrecharclass for details.
1175 .PP
1176 \fIAssertions\fR
1177 .IX Subsection "Assertions"
1178 .PP
1179 Besides \f(CW"^"\fR and \f(CW"$"\fR, Perl defines the following
1180 zero-width assertions:
1181 .IX Xref "zero-width assertion assertion regex, zero-width assertion regexp, zero-width assertion regular expression, zero-width assertion \\b \\B \\A \\Z \\z \\G"
1182 .PP
1183 .Vb 9
1184 \& \eb{} Match at Unicode boundary of specified type
1185 \& \eB{} Match where corresponding \eb{} doesn\*(Aqt match
1186 \& \eb Match a \ew\eW or \eW\ew boundary
1187 \& \eB Match except at a \ew\eW or \eW\ew boundary
1188 \& \eA Match only at beginning of string
1189 \& \eZ Match only at end of string, or before newline at the end
1190 \& \ez Match only at end of string
1191 \& \eG Match only at pos() (e.g. at the end\-of\-match position
1192 \& of prior m//g)
1193 .Ve
1194 .PP
1195 A Unicode boundary (\f(CW\*(C`\eb{}\*(C'\fR), available starting in v5.22, is a spot
1196 between two characters, or before the first character in the string, or
1197 after the final character in the string where certain criteria defined
1198 by Unicode are met. See "\eb{}, \eb, \eB{}, \eB" in perlrebackslash for
1199 details.
1200 .PP
1201 A word boundary (\f(CW\*(C`\eb\*(C'\fR) is a spot between two characters
1202 that has a \f(CW\*(C`\ew\*(C'\fR on one side of it and a \f(CW\*(C`\eW\*(C'\fR on the other side
1203 of it (in either order), counting the imaginary characters off the
1204 beginning and end of the string as matching a \f(CW\*(C`\eW\*(C'\fR. (Within
1205 character classes \f(CW\*(C`\eb\*(C'\fR represents backspace rather than a word
1206 boundary, just as it normally does in any double-quoted string.)
1207 The \f(CW\*(C`\eA\*(C'\fR and \f(CW\*(C`\eZ\*(C'\fR are just like \f(CW"^"\fR and \f(CW"$"\fR, except that they
1208 won't match multiple times when the \f(CW\*(C`/m\*(C'\fR modifier is used, while
1209 \&\f(CW"^"\fR and \f(CW"$"\fR will match at every internal line boundary. To match
1210 the actual end of the string and not ignore an optional trailing
1211 newline, use \f(CW\*(C`\ez\*(C'\fR.
1212 .IX Xref "\\b \\A \\Z \\z m"
1213 .PP
1214 The \f(CW\*(C`\eG\*(C'\fR assertion can be used to chain global matches (using
1215 \&\f(CW\*(C`m//g\*(C'\fR), as described in "Regexp Quote-Like Operators" in perlop.
1216 It is also useful when writing \f(CW\*(C`lex\*(C'\fR\-like scanners, when you have
1217 several patterns that you want to match against consequent substrings
1218 of your string; see the previous reference. The actual location
1219 where \f(CW\*(C`\eG\*(C'\fR will match can also be influenced by using \f(CWpos()\fR as
1220 an lvalue: see "pos" in perlfunc. Note that the rule for zero-length
1221 matches (see "Repeated Patterns Matching a Zero-length Substring")
1222 is modified somewhat, in that contents to the left of \f(CW\*(C`\eG\*(C'\fR are
1223 not counted when determining the length of the match. Thus the following
1224 will not match forever:
1225 .IX Xref "\\G"
1226 .PP
1227 .Vb 5
1228 \& my $string = \*(AqABC\*(Aq;
1229 \& pos($string) = 1;
1230 \& while ($string =~ /(.\eG)/g) {
1231 \& print $1;
1232 \& }
1233 .Ve
1234 .PP
1235 It will print 'A' and then terminate, as it considers the match to
1236 be zero-width, and thus will not match at the same position twice in a
1237 row.
1238 .PP
1239 It is worth noting that \f(CW\*(C`\eG\*(C'\fR improperly used can result in an infinite
1240 loop. Take care when using patterns that include \f(CW\*(C`\eG\*(C'\fR in an alternation.
1241 .PP
1242 Note also that \f(CW\*(C`s///\*(C'\fR will refuse to overwrite part of a substitution
1243 that has already been replaced; so for example this will stop after the
1244 first iteration, rather than iterating its way backwards through the
1245 string:
1246 .PP
1247 .Vb 4
1248 \& $_ = "123456789";
1249 \& pos = 6;
1250 \& s/.(?=.\eG)/X/g;
1251 \& print; # prints 1234X6789, not XXXXX6789
1252 .Ve
1253 .PP
1254 \fICapture groups\fR
1255 .IX Subsection "Capture groups"
1256 .PP
1257 The grouping construct \f(CW\*(C`( ... )\*(C'\fR creates capture groups (also referred to as
1258 capture buffers). To refer to the current contents of a group later on, within
1259 the same pattern, use \f(CW\*(C`\eg1\*(C'\fR (or \f(CW\*(C`\eg{1}\*(C'\fR) for the first, \f(CW\*(C`\eg2\*(C'\fR (or \f(CW\*(C`\eg{2}\*(C'\fR)
1260 for the second, and so on.
1261 This is called a \fIbackreference\fR.
1262
1263
1264
1265
1266
1267
1268
1269
1270 There is no limit to the number of captured substrings that you may use.
1271 Groups are numbered with the leftmost open parenthesis being number 1, \fIetc\fR. If
1272 a group did not match, the associated backreference won't match either. (This
1273 can happen if the group is optional, or in a different branch of an
1274 alternation.)
1275 You can omit the \f(CW"g"\fR, and write \f(CW"\e1"\fR, \fIetc\fR, but there are some issues with
1276 this form, described below.
1277 .IX Xref "regex, capture buffer regexp, capture buffer regex, capture group regexp, capture group regular expression, capture buffer backreference regular expression, capture group backreference \\g{1} \\g{-1} \\g{name} relative backreference named backreference named capture buffer regular expression, named capture buffer named capture group regular expression, named capture group %+ $+{name} \\k<name>"
1278 .PP
1279 You can also refer to capture groups relatively, by using a negative number, so
1280 that \f(CW\*(C`\eg\-1\*(C'\fR and \f(CW\*(C`\eg{\-1}\*(C'\fR both refer to the immediately preceding capture
1281 group, and \f(CW\*(C`\eg\-2\*(C'\fR and \f(CW\*(C`\eg{\-2}\*(C'\fR both refer to the group before it. For
1282 example:
1283 .PP
1284 .Vb 8
1285 \& /
1286 \& (Y) # group 1
1287 \& ( # group 2
1288 \& (X) # group 3
1289 \& \eg{\-1} # backref to group 3
1290 \& \eg{\-3} # backref to group 1
1291 \& )
1292 \& /x
1293 .Ve
1294 .PP
1295 would match the same as \f(CW\*(C`/(Y) ( (X) \eg3 \eg1 )/x\*(C'\fR. This allows you to
1296 interpolate regexes into larger regexes and not have to worry about the
1297 capture groups being renumbered.
1298 .PP
1299 You can dispense with numbers altogether and create named capture groups.
1300 The notation is \f(CW\*(C`(?<\fR\f(CIname\fR\f(CW>...)\*(C'\fR to declare and \f(CW\*(C`\eg{\fR\f(CIname\fR\f(CW}\*(C'\fR to
1301 reference. (To be compatible with .Net regular expressions, \f(CW\*(C`\eg{\fR\f(CIname\fR\f(CW}\*(C'\fR may
1302 also be written as \f(CW\*(C`\ek{\fR\f(CIname\fR\f(CW}\*(C'\fR, \f(CW\*(C`\ek<\fR\f(CIname\fR\f(CW>\*(C'\fR or \f(CW\*(C`\ek\*(Aq\fR\f(CIname\fR\f(CW\*(Aq\*(C'\fR.)
1303 \&\fIname\fR must not begin with a number, nor contain hyphens.
1304 When different groups within the same pattern have the same name, any reference
1305 to that name assumes the leftmost defined group. Named groups count in
1306 absolute and relative numbering, and so can also be referred to by those
1307 numbers.
1308 (It's possible to do things with named capture groups that would otherwise
1309 require \f(CW\*(C`(??{})\*(C'\fR.)
1310 .PP
1311 Capture group contents are dynamically scoped and available to you outside the
1312 pattern until the end of the enclosing block or until the next successful
1313 match in the same scope, whichever comes first.
1314 See "Compound Statements" in perlsyn and
1315 "Scoping Rules of Regex Variables" in perlvar for more details.
1316 .PP
1317 You can access the contents of a capture group by absolute number (using
1318 \&\f(CW"$1"\fR instead of \f(CW"\eg1"\fR, \fIetc\fR); or by name via the \f(CW\*(C`%+\*(C'\fR hash,
1319 using \f(CW"$+{\fR\f(CIname\fR\f(CW}"\fR.
1320 .PP
1321 Braces are required in referring to named capture groups, but are optional for
1322 absolute or relative numbered ones. Braces are safer when creating a regex by
1323 concatenating smaller strings. For example if you have \f(CW\*(C`qr/$x$y/\*(C'\fR, and \f(CW$x\fR
1324 contained \f(CW"\eg1"\fR, and \f(CW$y\fR contained \f(CW"37"\fR, you would get \f(CW\*(C`/\eg137/\*(C'\fR which
1325 is probably not what you intended.
1326 .PP
1327 If you use braces, you may also optionally add any number of blank
1328 (space or tab) characters within but adjacent to the braces, like
1329 \&\f(CW\*(C`\eg{\ \-1\ }\*(C'\fR, or \f(CW\*(C`\ek{\ \fR\f(CIname\fR\f(CW\ }\*(C'\fR.
1330 .PP
1331 The \f(CW\*(C`\eg\*(C'\fR and \f(CW\*(C`\ek\*(C'\fR notations were introduced in Perl 5.10.0. Prior to that
1332 there were no named nor relative numbered capture groups. Absolute numbered
1333 groups were referred to using \f(CW\*(C`\e1\*(C'\fR,
1334 \&\f(CW\*(C`\e2\*(C'\fR, \fIetc\fR., and this notation is still
1335 accepted (and likely always will be). But it leads to some ambiguities if
1336 there are more than 9 capture groups, as \f(CW\*(C`\e10\*(C'\fR could mean either the tenth
1337 capture group, or the character whose ordinal in octal is 010 (a backspace in
1338 ASCII). Perl resolves this ambiguity by interpreting \f(CW\*(C`\e10\*(C'\fR as a backreference
1339 only if at least 10 left parentheses have opened before it. Likewise \f(CW\*(C`\e11\*(C'\fR is
1340 a backreference only if at least 11 left parentheses have opened before it.
1341 And so on. \f(CW\*(C`\e1\*(C'\fR through \f(CW\*(C`\e9\*(C'\fR are always interpreted as backreferences.
1342 There are several examples below that illustrate these perils. You can avoid
1343 the ambiguity by always using \f(CW\*(C`\eg{}\*(C'\fR or \f(CW\*(C`\eg\*(C'\fR if you mean capturing groups;
1344 and for octal constants always using \f(CW\*(C`\eo{}\*(C'\fR, or for \f(CW\*(C`\e077\*(C'\fR and below, using 3
1345 digits padded with leading zeros, since a leading zero implies an octal
1346 constant.
1347 .PP
1348 The \f(CW\*(C`\e\fR\f(CIdigit\fR\f(CW\*(C'\fR notation also works in certain circumstances outside
1349 the pattern. See "Warning on \e1 Instead of \f(CW$1\fR" below for details.
1350 .PP
1351 Examples:
1352 .PP
1353 .Vb 1
1354 \& s/^([^ ]*) *([^ ]*)/$2 $1/; # swap first two words
1355 \&
1356 \& /(.)\eg1/ # find first doubled char
1357 \& and print "\*(Aq$1\*(Aq is the first doubled character\en";
1358 \&
1359 \& /(?<char>.)\ek<char>/ # ... a different way
1360 \& and print "\*(Aq$+{char}\*(Aq is the first doubled character\en";
1361 \&
1362 \& /(?\*(Aqchar\*(Aq.)\eg1/ # ... mix and match
1363 \& and print "\*(Aq$1\*(Aq is the first doubled character\en";
1364 \&
1365 \& if (/Time: (..):(..):(..)/) { # parse out values
1366 \& $hours = $1;
1367 \& $minutes = $2;
1368 \& $seconds = $3;
1369 \& }
1370 \&
1371 \& /(.)(.)(.)(.)(.)(.)(.)(.)(.)\eg10/ # \eg10 is a backreference
1372 \& /(.)(.)(.)(.)(.)(.)(.)(.)(.)\e10/ # \e10 is octal
1373 \& /((.)(.)(.)(.)(.)(.)(.)(.)(.))\e10/ # \e10 is a backreference
1374 \& /((.)(.)(.)(.)(.)(.)(.)(.)(.))\e010/ # \e010 is octal
1375 \&
1376 \& $x = \*(Aq(.)\e1\*(Aq; # Creates problems when concatenated.
1377 \& $y = \*(Aq(.)\eg{1}\*(Aq; # Avoids the problems.
1378 \& "aa" =~ /${x}/; # True
1379 \& "aa" =~ /${y}/; # True
1380 \& "aa0" =~ /${x}0/; # False!
1381 \& "aa0" =~ /${y}0/; # True
1382 \& "aa\ex08" =~ /${x}0/; # True!
1383 \& "aa\ex08" =~ /${y}0/; # False
1384 .Ve
1385 .PP
1386 Several special variables also refer back to portions of the previous
1387 match. \f(CW$+\fR returns whatever the last bracket match matched.
1388 \&\f(CW$&\fR returns the entire matched string. (At one point \f(CW$0\fR did
1389 also, but now it returns the name of the program.) \f(CW\*(C`$\`\*(C'\fR returns
1390 everything before the matched string. \f(CW\*(C`$\*(Aq\*(C'\fR returns everything
1391 after the matched string. And \f(CW$^N\fR contains whatever was matched by
1392 the most-recently closed group (submatch). \f(CW$^N\fR can be used in
1393 extended patterns (see below), for example to assign a submatch to a
1394 variable.
1395 .IX Xref "$+ $^N $& $` $'"
1396 .PP
1397 These special variables, like the \f(CW\*(C`%+\*(C'\fR hash and the numbered match variables
1398 (\f(CW$1\fR, \f(CW$2\fR, \f(CW$3\fR, \fIetc\fR.) are dynamically scoped
1399 until the end of the enclosing block or until the next successful
1400 match, whichever comes first. (See "Compound Statements" in perlsyn.)
1401 .IX Xref "$+ $^N $& $` $' $1 $2 $3 $4 $5 $6 $7 $8 $9 @{^CAPTURE}"
1402 .PP
1403 The \f(CW\*(C`@{^CAPTURE}\*(C'\fR array may be used to access ALL of the capture buffers
1404 as an array without needing to know how many there are. For instance
1405 .PP
1406 .Vb 1
1407 \& $string=~/$pattern/ and @captured = @{^CAPTURE};
1408 .Ve
1409 .PP
1410 will place a copy of each capture variable, \f(CW$1\fR, \f(CW$2\fR etc, into the
1411 \&\f(CW@captured\fR array.
1412 .PP
1413 Be aware that when interpolating a subscript of the \f(CW\*(C`@{^CAPTURE}\*(C'\fR
1414 array you must use demarcated curly brace notation:
1415 .PP
1416 .Vb 1
1417 \& print "${^CAPTURE[0]}";
1418 .Ve
1419 .PP
1420 See "Demarcated variable names using braces" in perldata for more on
1421 this notation.
1422 .PP
1423 \&\fBNOTE\fR: Failed matches in Perl do not reset the match variables,
1424 which makes it easier to write code that tests for a series of more
1425 specific cases and remembers the best match.
1426 .PP
1427 \&\fBWARNING\fR: If your code is to run on Perl 5.16 or earlier,
1428 beware that once Perl sees that you need one of \f(CW$&\fR, \f(CW\*(C`$\`\*(C'\fR, or
1429 \&\f(CW\*(C`$\*(Aq\*(C'\fR anywhere in the program, it has to provide them for every
1430 pattern match. This may substantially slow your program.
1431 .PP
1432 Perl uses the same mechanism to produce \f(CW$1\fR, \f(CW$2\fR, \fIetc\fR, so you also
1433 pay a price for each pattern that contains capturing parentheses.
1434 (To avoid this cost while retaining the grouping behaviour, use the
1435 extended regular expression \f(CW\*(C`(?: ... )\*(C'\fR instead.) But if you never
1436 use \f(CW$&\fR, \f(CW\*(C`$\`\*(C'\fR or \f(CW\*(C`$\*(Aq\*(C'\fR, then patterns \fIwithout\fR capturing
1437 parentheses will not be penalized. So avoid \f(CW$&\fR, \f(CW\*(C`$\*(Aq\*(C'\fR, and \f(CW\*(C`$\`\*(C'\fR
1438 if you can, but if you can't (and some algorithms really appreciate
1439 them), once you've used them once, use them at will, because you've
1440 already paid the price.
1441 .IX Xref "$& $` $'"
1442 .PP
1443 Perl 5.16 introduced a slightly more efficient mechanism that notes
1444 separately whether each of \f(CW\*(C`$\`\*(C'\fR, \f(CW$&\fR, and \f(CW\*(C`$\*(Aq\*(C'\fR have been seen, and
1445 thus may only need to copy part of the string. Perl 5.20 introduced a
1446 much more efficient copy-on-write mechanism which eliminates any slowdown.
1447 .PP
1448 As another workaround for this problem, Perl 5.10.0 introduced \f(CW\*(C`${^PREMATCH}\*(C'\fR,
1449 \&\f(CW\*(C`${^MATCH}\*(C'\fR and \f(CW\*(C`${^POSTMATCH}\*(C'\fR, which are equivalent to \f(CW\*(C`$\`\*(C'\fR, \f(CW$&\fR
1450 and \f(CW\*(C`$\*(Aq\*(C'\fR, \fBexcept\fR that they are only guaranteed to be defined after a
1451 successful match that was executed with the \f(CW\*(C`/p\*(C'\fR (preserve) modifier.
1452 The use of these variables incurs no global performance penalty, unlike
1453 their punctuation character equivalents, however at the trade-off that you
1454 have to tell perl when you want to use them.
1455 .IX Xref " p p modifier"
1456 .SS "Quoting metacharacters"
1457 .IX Subsection "Quoting metacharacters"
1458 Backslashed metacharacters in Perl are alphanumeric, such as \f(CW\*(C`\eb\*(C'\fR,
1459 \&\f(CW\*(C`\ew\*(C'\fR, \f(CW\*(C`\en\*(C'\fR. Unlike some other regular expression languages, there
1460 are no backslashed symbols that aren't alphanumeric. So anything
1461 that looks like \f(CW\*(C`\e\e\*(C'\fR, \f(CW\*(C`\e(\*(C'\fR, \f(CW\*(C`\e)\*(C'\fR, \f(CW\*(C`\e[\*(C'\fR, \f(CW\*(C`\e]\*(C'\fR, \f(CW\*(C`\e{\*(C'\fR, or \f(CW\*(C`\e}\*(C'\fR is
1462 always
1463 interpreted as a literal character, not a metacharacter. This was
1464 once used in a common idiom to disable or quote the special meanings
1465 of regular expression metacharacters in a string that you want to
1466 use for a pattern. Simply quote all non\-"word" characters:
1467 .PP
1468 .Vb 1
1469 \& $pattern =~ s/(\eW)/\e\e$1/g;
1470 .Ve
1471 .PP
1472 (If \f(CW\*(C`use locale\*(C'\fR is set, then this depends on the current locale.)
1473 Today it is more common to use the \f(CWquotemeta()\fR
1474 function or the \f(CW\*(C`\eQ\*(C'\fR metaquoting escape sequence to disable all
1475 metacharacters' special meanings like this:
1476 .PP
1477 .Vb 1
1478 \& /$unquoted\eQ$quoted\eE$unquoted/
1479 .Ve
1480 .PP
1481 Beware that if you put literal backslashes (those not inside
1482 interpolated variables) between \f(CW\*(C`\eQ\*(C'\fR and \f(CW\*(C`\eE\*(C'\fR, double-quotish
1483 backslash interpolation may lead to confusing results. If you
1484 \&\fIneed\fR to use literal backslashes within \f(CW\*(C`\eQ...\eE\*(C'\fR,
1485 consult "Gory details of parsing quoted constructs" in perlop.
1486 .PP
1487 \&\f(CWquotemeta()\fR and \f(CW\*(C`\eQ\*(C'\fR are fully described in "quotemeta" in perlfunc.
1488 .SS "Extended Patterns"
1489 .IX Subsection "Extended Patterns"
1490 Perl also defines a consistent extension syntax for features not
1491 found in standard tools like \fBawk\fR and
1492 \&\fBlex\fR. The syntax for most of these is a
1493 pair of parentheses with a question mark as the first thing within
1494 the parentheses. The character after the question mark indicates
1495 the extension.
1496 .PP
1497 A question mark was chosen for this and for the minimal-matching
1498 construct because 1) question marks are rare in older regular
1499 expressions, and 2) whenever you see one, you should stop and
1500 "question" exactly what is going on. That's psychology....
1501 .ie n .IP """(?#\fItext\fR)""" 4
1502 .el .IP \f(CW(?#\fR\f(CItext\fR\f(CW)\fR 4
1503 .IX Xref "(?#)"
1504 .IX Item "(?#text)"
1505 A comment. The \fItext\fR is ignored.
1506 Note that Perl closes
1507 the comment as soon as it sees a \f(CW")"\fR, so there is no way to put a literal
1508 \&\f(CW")"\fR in the comment. The pattern's closing delimiter must be escaped by
1509 a backslash if it appears in the comment.
1510 .Sp
1511 See "/x" for another way to have comments in patterns.
1512 .Sp
1513 Note that a comment can go just about anywhere, except in the middle of
1514 an escape sequence. Examples:
1515 .Sp
1516 .Vb 1
1517 \& qr/foo(?#comment)bar/\*(Aq # Matches \*(Aqfoobar\*(Aq
1518 \&
1519 \& # The pattern below matches \*(Aqabcd\*(Aq, \*(Aqabccd\*(Aq, or \*(Aqabcccd\*(Aq
1520 \& qr/abc(?#comment between literal and its quantifier){1,3}d/
1521 \&
1522 \& # The pattern below generates a syntax error, because the \*(Aq\ep\*(Aq must
1523 \& # be followed immediately by a \*(Aq{\*(Aq.
1524 \& qr/\ep(?#comment between \ep and its property name){Any}/
1525 \&
1526 \& # The pattern below generates a syntax error, because the initial
1527 \& # \*(Aq\e(\*(Aq is a literal opening parenthesis, and so there is nothing
1528 \& # for the closing \*(Aq)\*(Aq to match
1529 \& qr/\e(?#the backslash means this isn\*(Aqt a comment)p{Any}/
1530 \&
1531 \& # Comments can be used to fold long patterns into multiple lines
1532 \& qr/First part of a long regex(?#
1533 \& )remaining part/
1534 .Ve
1535 .ie n .IP """(?adlupimnsx\-imnsx)""" 4
1536 .el .IP \f(CW(?adlupimnsx\-imnsx)\fR 4
1537 .IX Item "(?adlupimnsx-imnsx)"
1538 .PD 0
1539 .ie n .IP """(?^alupimnsx)""" 4
1540 .el .IP \f(CW(?^alupimnsx)\fR 4
1541 .IX Xref "(?) (?^)"
1542 .IX Item "(?^alupimnsx)"
1543 .PD
1544 Zero or more embedded pattern-match modifiers, to be turned on (or
1545 turned off if preceded by \f(CW"\-"\fR) for the remainder of the pattern or
1546 the remainder of the enclosing pattern group (if any).
1547 .Sp
1548 This is particularly useful for dynamically-generated patterns,
1549 such as those read in from a
1550 configuration file, taken from an argument, or specified in a table
1551 somewhere. Consider the case where some patterns want to be
1552 case-sensitive and some do not: The case-insensitive ones merely need to
1553 include \f(CW\*(C`(?i)\*(C'\fR at the front of the pattern. For example:
1554 .Sp
1555 .Vb 2
1556 \& $pattern = "foobar";
1557 \& if ( /$pattern/i ) { }
1558 \&
1559 \& # more flexible:
1560 \&
1561 \& $pattern = "(?i)foobar";
1562 \& if ( /$pattern/ ) { }
1563 .Ve
1564 .Sp
1565 These modifiers are restored at the end of the enclosing group. For example,
1566 .Sp
1567 .Vb 1
1568 \& ( (?i) blah ) \es+ \eg1
1569 .Ve
1570 .Sp
1571 will match \f(CW\*(C`blah\*(C'\fR in any case, some spaces, and an exact (\fIincluding the case\fR!)
1572 repetition of the previous word, assuming the \f(CW\*(C`/x\*(C'\fR modifier, and no \f(CW\*(C`/i\*(C'\fR
1573 modifier outside this group.
1574 .Sp
1575 These modifiers do not carry over into named subpatterns called in the
1576 enclosing group. In other words, a pattern such as \f(CW\*(C`((?i)(?&\fR\f(CINAME\fR\f(CW))\*(C'\fR does not
1577 change the case-sensitivity of the \fINAME\fR pattern.
1578 .Sp
1579 A modifier is overridden by later occurrences of this construct in the
1580 same scope containing the same modifier, so that
1581 .Sp
1582 .Vb 1
1583 \& /((?im)foo(?\-m)bar)/
1584 .Ve
1585 .Sp
1586 matches all of \f(CW\*(C`foobar\*(C'\fR case insensitively, but uses \f(CW\*(C`/m\*(C'\fR rules for
1587 only the \f(CW\*(C`foo\*(C'\fR portion. The \f(CW"a"\fR flag overrides \f(CW\*(C`aa\*(C'\fR as well;
1588 likewise \f(CW\*(C`aa\*(C'\fR overrides \f(CW"a"\fR. The same goes for \f(CW"x"\fR and \f(CW\*(C`xx\*(C'\fR.
1589 Hence, in
1590 .Sp
1591 .Vb 1
1592 \& /(?\-x)foo/xx
1593 .Ve
1594 .Sp
1595 both \f(CW\*(C`/x\*(C'\fR and \f(CW\*(C`/xx\*(C'\fR are turned off during matching \f(CW\*(C`foo\*(C'\fR. And in
1596 .Sp
1597 .Vb 1
1598 \& /(?x)foo/x
1599 .Ve
1600 .Sp
1601 \&\f(CW\*(C`/x\*(C'\fR but NOT \f(CW\*(C`/xx\*(C'\fR is turned on for matching \f(CW\*(C`foo\*(C'\fR. (One might
1602 mistakenly think that since the inner \f(CW\*(C`(?x)\*(C'\fR is already in the scope of
1603 \&\f(CW\*(C`/x\*(C'\fR, that the result would effectively be the sum of them, yielding
1604 \&\f(CW\*(C`/xx\*(C'\fR. It doesn't work that way.) Similarly, doing something like
1605 \&\f(CW\*(C`(?xx\-x)foo\*(C'\fR turns off all \f(CW"x"\fR behavior for matching \f(CW\*(C`foo\*(C'\fR, it is not
1606 that you subtract 1 \f(CW"x"\fR from 2 to get 1 \f(CW"x"\fR remaining.
1607 .Sp
1608 Any of these modifiers can be set to apply globally to all regular
1609 expressions compiled within the scope of a \f(CW\*(C`use re\*(C'\fR. See
1610 "'/flags' mode" in re.
1611 .Sp
1612 Starting in Perl 5.14, a \f(CW"^"\fR (caret or circumflex accent) immediately
1613 after the \f(CW"?"\fR is a shorthand equivalent to \f(CW\*(C`d\-imnsx\*(C'\fR. Flags (except
1614 \&\f(CW"d"\fR) may follow the caret to override it.
1615 But a minus sign is not legal with it.
1616 .Sp
1617 Note that the \f(CW"a"\fR, \f(CW"d"\fR, \f(CW"l"\fR, \f(CW"p"\fR, and \f(CW"u"\fR modifiers are special in
1618 that they can only be enabled, not disabled, and the \f(CW"a"\fR, \f(CW"d"\fR, \f(CW"l"\fR, and
1619 \&\f(CW"u"\fR modifiers are mutually exclusive: specifying one de-specifies the
1620 others, and a maximum of one (or two \f(CW"a"\fR's) may appear in the
1621 construct. Thus, for
1622 example, \f(CW\*(C`(?\-p)\*(C'\fR will warn when compiled under \f(CW\*(C`use warnings\*(C'\fR;
1623 \&\f(CW\*(C`(?\-d:...)\*(C'\fR and \f(CW\*(C`(?dl:...)\*(C'\fR are fatal errors.
1624 .Sp
1625 Note also that the \f(CW"p"\fR modifier is special in that its presence
1626 anywhere in a pattern has a global effect.
1627 .Sp
1628 Having zero modifiers makes this a no-op (so why did you specify it,
1629 unless it's generated code), and starting in v5.30, warns under \f(CW\*(C`use
1630 re \*(Aqstrict\*(Aq\*(C'\fR.
1631 .ie n .IP """(?:\fIpattern\fR)""" 4
1632 .el .IP \f(CW(?:\fR\f(CIpattern\fR\f(CW)\fR 4
1633 .IX Xref "(?:)"
1634 .IX Item "(?:pattern)"
1635 .PD 0
1636 .ie n .IP """(?adluimnsx\-imnsx:\fIpattern\fR)""" 4
1637 .el .IP \f(CW(?adluimnsx\-imnsx:\fR\f(CIpattern\fR\f(CW)\fR 4
1638 .IX Item "(?adluimnsx-imnsx:pattern)"
1639 .ie n .IP """(?^aluimnsx:\fIpattern\fR)""" 4
1640 .el .IP \f(CW(?^aluimnsx:\fR\f(CIpattern\fR\f(CW)\fR 4
1641 .IX Xref "(?^:)"
1642 .IX Item "(?^aluimnsx:pattern)"
1643 .PD
1644 This is for clustering, not capturing; it groups subexpressions like
1645 \&\f(CW"()"\fR, but doesn't make backreferences as \f(CW"()"\fR does. So
1646 .Sp
1647 .Vb 1
1648 \& @fields = split(/\eb(?:a|b|c)\eb/)
1649 .Ve
1650 .Sp
1651 matches the same field delimiters as
1652 .Sp
1653 .Vb 1
1654 \& @fields = split(/\eb(a|b|c)\eb/)
1655 .Ve
1656 .Sp
1657 but doesn't spit out the delimiters themselves as extra fields (even though
1658 that's the behaviour of "split" in perlfunc when its pattern contains capturing
1659 groups). It's also cheaper not to capture
1660 characters if you don't need to.
1661 .Sp
1662 Any letters between \f(CW"?"\fR and \f(CW":"\fR act as flags modifiers as with
1663 \&\f(CW\*(C`(?adluimnsx\-imnsx)\*(C'\fR. For example,
1664 .Sp
1665 .Vb 1
1666 \& /(?s\-i:more.*than).*million/i
1667 .Ve
1668 .Sp
1669 is equivalent to the more verbose
1670 .Sp
1671 .Vb 1
1672 \& /(?:(?s\-i)more.*than).*million/i
1673 .Ve
1674 .Sp
1675 Note that any \f(CW\*(C`()\*(C'\fR constructs enclosed within this one will still
1676 capture unless the \f(CW\*(C`/n\*(C'\fR modifier is in effect.
1677 .Sp
1678 Like the "(?adlupimnsx\-imnsx)" construct, \f(CW\*(C`aa\*(C'\fR and \f(CW"a"\fR override each
1679 other, as do \f(CW\*(C`xx\*(C'\fR and \f(CW"x"\fR. They are not additive. So, doing
1680 something like \f(CW\*(C`(?xx\-x:foo)\*(C'\fR turns off all \f(CW"x"\fR behavior for matching
1681 \&\f(CW\*(C`foo\*(C'\fR.
1682 .Sp
1683 Starting in Perl 5.14, a \f(CW"^"\fR (caret or circumflex accent) immediately
1684 after the \f(CW"?"\fR is a shorthand equivalent to \f(CW\*(C`d\-imnsx\*(C'\fR. Any positive
1685 flags (except \f(CW"d"\fR) may follow the caret, so
1686 .Sp
1687 .Vb 1
1688 \& (?^x:foo)
1689 .Ve
1690 .Sp
1691 is equivalent to
1692 .Sp
1693 .Vb 1
1694 \& (?x\-imns:foo)
1695 .Ve
1696 .Sp
1697 The caret tells Perl that this cluster doesn't inherit the flags of any
1698 surrounding pattern, but uses the system defaults (\f(CW\*(C`d\-imnsx\*(C'\fR),
1699 modified by any flags specified.
1700 .Sp
1701 The caret allows for simpler stringification of compiled regular
1702 expressions. These look like
1703 .Sp
1704 .Vb 1
1705 \& (?^:pattern)
1706 .Ve
1707 .Sp
1708 with any non-default flags appearing between the caret and the colon.
1709 A test that looks at such stringification thus doesn't need to have the
1710 system default flags hard-coded in it, just the caret. If new flags are
1711 added to Perl, the meaning of the caret's expansion will change to include
1712 the default for those flags, so the test will still work, unchanged.
1713 .Sp
1714 Specifying a negative flag after the caret is an error, as the flag is
1715 redundant.
1716 .Sp
1717 Mnemonic for \f(CW\*(C`(?^...)\*(C'\fR: A fresh beginning since the usual use of a caret is
1718 to match at the beginning.
1719 .ie n .IP """(?|\fIpattern\fR)""" 4
1720 .el .IP \f(CW(?|\fR\f(CIpattern\fR\f(CW)\fR 4
1721 .IX Xref "(?|) Branch reset"
1722 .IX Item "(?|pattern)"
1723 This is the "branch reset" pattern, which has the special property
1724 that the capture groups are numbered from the same starting point
1725 in each alternation branch. It is available starting from perl 5.10.0.
1726 .Sp
1727 Capture groups are numbered from left to right, but inside this
1728 construct the numbering is restarted for each branch.
1729 .Sp
1730 The numbering within each branch will be as normal, and any groups
1731 following this construct will be numbered as though the construct
1732 contained only one branch, that being the one with the most capture
1733 groups in it.
1734 .Sp
1735 This construct is useful when you want to capture one of a
1736 number of alternative matches.
1737 .Sp
1738 Consider the following pattern. The numbers underneath show in
1739 which group the captured content will be stored.
1740 .Sp
1741 .Vb 3
1742 \& # before \-\-\-\-\-\-\-\-\-\-\-\-\-\-\-branch\-reset\-\-\-\-\-\-\-\-\-\-\- after
1743 \& / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x
1744 \& # 1 2 2 3 2 3 4
1745 .Ve
1746 .Sp
1747 Be careful when using the branch reset pattern in combination with
1748 named captures. Named captures are implemented as being aliases to
1749 numbered groups holding the captures, and that interferes with the
1750 implementation of the branch reset pattern. If you are using named
1751 captures in a branch reset pattern, it's best to use the same names,
1752 in the same order, in each of the alternations:
1753 .Sp
1754 .Vb 2
1755 \& /(?| (?<a> x ) (?<b> y )
1756 \& | (?<a> z ) (?<b> w )) /x
1757 .Ve
1758 .Sp
1759 Not doing so may lead to surprises:
1760 .Sp
1761 .Vb 3
1762 \& "12" =~ /(?| (?<a> \ed+ ) | (?<b> \eD+))/x;
1763 \& say $+{a}; # Prints \*(Aq12\*(Aq
1764 \& say $+{b}; # *Also* prints \*(Aq12\*(Aq.
1765 .Ve
1766 .Sp
1767 The problem here is that both the group named \f(CW\*(C`a\*(C'\fR and the group
1768 named \f(CW\*(C`b\*(C'\fR are aliases for the group belonging to \f(CW$1\fR.
1769 .IP "Lookaround Assertions" 4
1770 .IX Xref "look-around assertion lookaround assertion look-around lookaround"
1771 .IX Item "Lookaround Assertions"
1772 Lookaround assertions are zero-width patterns which match a specific
1773 pattern without including it in \f(CW$&\fR. Positive assertions match when
1774 their subpattern matches, negative assertions match when their subpattern
1775 fails. Lookbehind matches text up to the current match position,
1776 lookahead matches text following the current match position.
1777 .RS 4
1778 .ie n .IP """(?=\fIpattern\fR)""" 4
1779 .el .IP \f(CW(?=\fR\f(CIpattern\fR\f(CW)\fR 4
1780 .IX Item "(?=pattern)"
1781 .PD 0
1782 .ie n .IP """(*pla:\fIpattern\fR)""" 4
1783 .el .IP \f(CW(*pla:\fR\f(CIpattern\fR\f(CW)\fR 4
1784 .IX Item "(*pla:pattern)"
1785 .ie n .IP """(*positive_lookahead:\fIpattern\fR)""" 4
1786 .el .IP \f(CW(*positive_lookahead:\fR\f(CIpattern\fR\f(CW)\fR 4
1787 .IX Xref "(?=) (*pla (*positive_lookahead look-ahead, positive lookahead, positive"
1788 .IX Item "(*positive_lookahead:pattern)"
1789 .PD
1790 A zero-width positive lookahead assertion. For example, \f(CW\*(C`/\ew+(?=\et)/\*(C'\fR
1791 matches a word followed by a tab, without including the tab in \f(CW$&\fR.
1792 .ie n .IP """(?!\fIpattern\fR)""" 4
1793 .el .IP \f(CW(?!\fR\f(CIpattern\fR\f(CW)\fR 4
1794 .IX Item "(?!pattern)"
1795 .PD 0
1796 .ie n .IP """(*nla:\fIpattern\fR)""" 4
1797 .el .IP \f(CW(*nla:\fR\f(CIpattern\fR\f(CW)\fR 4
1798 .IX Item "(*nla:pattern)"
1799 .ie n .IP """(*negative_lookahead:\fIpattern\fR)""" 4
1800 .el .IP \f(CW(*negative_lookahead:\fR\f(CIpattern\fR\f(CW)\fR 4
1801 .IX Xref "(?!) (*nla (*negative_lookahead look-ahead, negative lookahead, negative"
1802 .IX Item "(*negative_lookahead:pattern)"
1803 .PD
1804 A zero-width negative lookahead assertion. For example \f(CW\*(C`/foo(?!bar)/\*(C'\fR
1805 matches any occurrence of "foo" that isn't followed by "bar". Note
1806 however that lookahead and lookbehind are NOT the same thing. You cannot
1807 use this for lookbehind.
1808 .Sp
1809 If you are looking for a "bar" that isn't preceded by a "foo", \f(CW\*(C`/(?!foo)bar/\*(C'\fR
1810 will not do what you want. That's because the \f(CW\*(C`(?!foo)\*(C'\fR is just saying that
1811 the next thing cannot be "foo"\-\-and it's not, it's a "bar", so "foobar" will
1812 match. Use lookbehind instead (see below).
1813 .ie n .IP """(?<=\fIpattern\fR)""" 4
1814 .el .IP \f(CW(?<=\fR\f(CIpattern\fR\f(CW)\fR 4
1815 .IX Item "(?<=pattern)"
1816 .PD 0
1817 .ie n .IP """\eK""" 4
1818 .el .IP \f(CW\eK\fR 4
1819 .IX Item "K"
1820 .ie n .IP """(*plb:\fIpattern\fR)""" 4
1821 .el .IP \f(CW(*plb:\fR\f(CIpattern\fR\f(CW)\fR 4
1822 .IX Item "(*plb:pattern)"
1823 .ie n .IP """(*positive_lookbehind:\fIpattern\fR)""" 4
1824 .el .IP \f(CW(*positive_lookbehind:\fR\f(CIpattern\fR\f(CW)\fR 4
1825 .IX Xref "(?<=) (*plb (*positive_lookbehind look-behind, positive lookbehind, positive \\K"
1826 .IX Item "(*positive_lookbehind:pattern)"
1827 .PD
1828 A zero-width positive lookbehind assertion. For example, \f(CW\*(C`/(?<=\et)\ew+/\*(C'\fR
1829 matches a word that follows a tab, without including the tab in \f(CW$&\fR.
1830 .Sp
1831 Prior to Perl 5.30, it worked only for fixed-width lookbehind, but
1832 starting in that release, it can handle variable lengths from 1 to 255
1833 characters as an experimental feature. The feature is enabled
1834 automatically if you use a variable length positive lookbehind assertion.
1835 .Sp
1836 In Perl 5.35.10 the scope of the experimental nature of this construct
1837 has been reduced, and experimental warnings will only be produced when
1838 the construct contains capturing parenthesis. The warnings will be
1839 raised at pattern compilation time, unless turned off, in the
1840 \&\f(CW\*(C`experimental::vlb\*(C'\fR category. This is to warn you that the exact
1841 contents of capturing buffers in a variable length positive lookbehind
1842 is not well defined and is subject to change in a future release of perl.
1843 .Sp
1844 Currently if you use capture buffers inside of a positive variable length
1845 lookbehind the result will be the longest and thus leftmost match possible.
1846 This means that
1847 .Sp
1848 .Vb 4
1849 \& "aax" =~ /(?=x)(?<=(a|aa))/
1850 \& "aax" =~ /(?=x)(?<=(aa|a))/
1851 \& "aax" =~ /(?=x)(?<=(a{1,2}?)/
1852 \& "aax" =~ /(?=x)(?<=(a{1,2})/
1853 .Ve
1854 .Sp
1855 will all result in \f(CW$1\fR containing \f(CW"aa"\fR. It is possible in a future
1856 release of perl we will change this behavior.
1857 .Sp
1858 There is a special form of this construct, called \f(CW\*(C`\eK\*(C'\fR
1859 (available since Perl 5.10.0), which causes the
1860 regex engine to "keep" everything it had matched prior to the \f(CW\*(C`\eK\*(C'\fR and
1861 not include it in \f(CW$&\fR. This effectively provides non-experimental
1862 variable-length lookbehind of any length.
1863 .Sp
1864 And, there is a technique that can be used to handle variable length
1865 lookbehinds on earlier releases, and longer than 255 characters. It is
1866 described in
1867 <http://www.drregex.com/2019/02/variable\-length\-lookbehinds\-actually.html>.
1868 .Sp
1869 Note that under \f(CW\*(C`/i\*(C'\fR, a few single characters match two or three other
1870 characters. This makes them variable length, and the 255 length applies
1871 to the maximum number of characters in the match. For
1872 example \f(CW\*(C`qr/\eN{LATIN SMALL LETTER SHARP S}/i\*(C'\fR matches the sequence
1873 \&\f(CW"ss"\fR. Your lookbehind assertion could contain 127 Sharp S
1874 characters under \f(CW\*(C`/i\*(C'\fR, but adding a 128th would generate a compilation
1875 error, as that could match 256 \f(CW"s"\fR characters in a row.
1876 .Sp
1877 The use of \f(CW\*(C`\eK\*(C'\fR inside of another lookaround assertion
1878 is allowed, but the behaviour is currently not well defined.
1879 .Sp
1880 For various reasons \f(CW\*(C`\eK\*(C'\fR may be significantly more efficient than the
1881 equivalent \f(CW\*(C`(?<=...)\*(C'\fR construct, and it is especially useful in
1882 situations where you want to efficiently remove something following
1883 something else in a string. For instance
1884 .Sp
1885 .Vb 1
1886 \& s/(foo)bar/$1/g;
1887 .Ve
1888 .Sp
1889 can be rewritten as the much more efficient
1890 .Sp
1891 .Vb 1
1892 \& s/foo\eKbar//g;
1893 .Ve
1894 .Sp
1895 Use of the non-greedy modifier \f(CW"?"\fR may not give you the expected
1896 results if it is within a capturing group within the construct.
1897 .ie n .IP """(?<!\fIpattern\fR)""" 4
1898 .el .IP \f(CW(?<!\fR\f(CIpattern\fR\f(CW)\fR 4
1899 .IX Item "(?<!pattern)"
1900 .PD 0
1901 .ie n .IP """(*nlb:\fIpattern\fR)""" 4
1902 .el .IP \f(CW(*nlb:\fR\f(CIpattern\fR\f(CW)\fR 4
1903 .IX Item "(*nlb:pattern)"
1904 .ie n .IP """(*negative_lookbehind:\fIpattern\fR)""" 4
1905 .el .IP \f(CW(*negative_lookbehind:\fR\f(CIpattern\fR\f(CW)\fR 4
1906 .IX Xref "(?<!) (*nlb (*negative_lookbehind look-behind, negative lookbehind, negative"
1907 .IX Item "(*negative_lookbehind:pattern)"
1908 .PD
1909 A zero-width negative lookbehind assertion. For example \f(CW\*(C`/(?<!bar)foo/\*(C'\fR
1910 matches any occurrence of "foo" that does not follow "bar".
1911 .Sp
1912 Prior to Perl 5.30, it worked only for fixed-width lookbehind, but
1913 starting in that release, it can handle variable lengths from 1 to 255
1914 characters as an experimental feature. The feature is enabled
1915 automatically if you use a variable length negative lookbehind assertion.
1916 .Sp
1917 In Perl 5.35.10 the scope of the experimental nature of this construct
1918 has been reduced, and experimental warnings will only be produced when
1919 the construct contains capturing parentheses. The warnings will be
1920 raised at pattern compilation time, unless turned off, in the
1921 \&\f(CW\*(C`experimental::vlb\*(C'\fR category. This is to warn you that the exact
1922 contents of capturing buffers in a variable length negative lookbehind
1923 is not well defined and is subject to change in a future release of perl.
1924 .Sp
1925 Currently if you use capture buffers inside of a negative variable length
1926 lookbehind the result may not be what you expect, for instance:
1927 .Sp
1928 .Vb 1
1929 \& say "axfoo"=~/(?=foo)(?<!(a|ax)(?{ say $1 }))/ ? "y" : "n";
1930 .Ve
1931 .Sp
1932 will output the following:
1933 .Sp
1934 .Vb 2
1935 \& a
1936 \& no
1937 .Ve
1938 .Sp
1939 which does not make sense as this should print out "ax" as the "a" does
1940 not line up at the correct place. Another example would be:
1941 .Sp
1942 .Vb 1
1943 \& say "yes: \*(Aq$1\-$2\*(Aq" if "aayfoo"=~/(?=foo)(?<!(a|aa)(a|aa)x)/;
1944 .Ve
1945 .Sp
1946 will output the following:
1947 .Sp
1948 .Vb 1
1949 \& yes: \*(Aqaa\-a\*(Aq
1950 .Ve
1951 .Sp
1952 It is possible in a future release of perl we will change this behavior
1953 so both of these examples produced more reasonable output.
1954 .Sp
1955 Note that we are confident that the construct will match and reject
1956 patterns appropriately, the undefined behavior strictly relates to the
1957 value of the capture buffer during or after matching.
1958 .Sp
1959 There is a technique that can be used to handle variable length
1960 lookbehind on earlier releases, and longer than 255 characters. It is
1961 described in
1962 <http://www.drregex.com/2019/02/variable\-length\-lookbehinds\-actually.html>.
1963 .Sp
1964 Note that under \f(CW\*(C`/i\*(C'\fR, a few single characters match two or three other
1965 characters. This makes them variable length, and the 255 length applies
1966 to the maximum number of characters in the match. For
1967 example \f(CW\*(C`qr/\eN{LATIN SMALL LETTER SHARP S}/i\*(C'\fR matches the sequence
1968 \&\f(CW"ss"\fR. Your lookbehind assertion could contain 127 Sharp S
1969 characters under \f(CW\*(C`/i\*(C'\fR, but adding a 128th would generate a compilation
1970 error, as that could match 256 \f(CW"s"\fR characters in a row.
1971 .Sp
1972 Use of the non-greedy modifier \f(CW"?"\fR may not give you the expected
1973 results if it is within a capturing group within the construct.
1974 .RE
1975 .RS 4
1976 .RE
1977 .ie n .IP """(?<\fINAME\fR>\fIpattern\fR)""" 4
1978 .el .IP \f(CW(?<\fR\f(CINAME\fR\f(CW>\fR\f(CIpattern\fR\f(CW)\fR 4
1979 .IX Item "(?<NAME>pattern)"
1980 .PD 0
1981 .ie n .IP """(?\*(Aq\fINAME\fR\*(Aq\fIpattern\fR)""" 4
1982 .el .IP \f(CW(?\*(Aq\fR\f(CINAME\fR\f(CW\*(Aq\fR\f(CIpattern\fR\f(CW)\fR 4
1983 .IX Xref "(?<NAME>) (?'NAME') named capture capture"
1984 .IX Item "(?NAMEpattern)"
1985 .PD
1986 A named capture group. Identical in every respect to normal capturing
1987 parentheses \f(CW\*(C`()\*(C'\fR but for the additional fact that the group
1988 can be referred to by name in various regular expression
1989 constructs (like \f(CW\*(C`\eg{\fR\f(CINAME\fR\f(CW}\*(C'\fR) and can be accessed by name
1990 after a successful match via \f(CW\*(C`%+\*(C'\fR or \f(CW\*(C`%\-\*(C'\fR. See perlvar
1991 for more details on the \f(CW\*(C`%+\*(C'\fR and \f(CW\*(C`%\-\*(C'\fR hashes.
1992 .Sp
1993 If multiple distinct capture groups have the same name, then
1994 \&\f(CW$+{\fR\f(CINAME\fR\f(CW}\fR will refer to the leftmost defined group in the match.
1995 .Sp
1996 The forms \f(CW\*(C`(?\*(Aq\fR\f(CINAME\fR\f(CW\*(Aq\fR\f(CIpattern\fR\f(CW)\*(C'\fR and \f(CW\*(C`(?<\fR\f(CINAME\fR\f(CW>\fR\f(CIpattern\fR\f(CW)\*(C'\fR
1997 are equivalent.
1998 .Sp
1999 \&\fBNOTE:\fR While the notation of this construct is the same as the similar
2000 function in .NET regexes, the behavior is not. In Perl the groups are
2001 numbered sequentially regardless of being named or not. Thus in the
2002 pattern
2003 .Sp
2004 .Vb 1
2005 \& /(x)(?<foo>y)(z)/
2006 .Ve
2007 .Sp
2008 \&\f(CW$+{foo}\fR will be the same as \f(CW$2\fR, and \f(CW$3\fR will contain 'z' instead of
2009 the opposite which is what a .NET regex hacker might expect.
2010 .Sp
2011 Currently \fINAME\fR is restricted to simple identifiers only.
2012 In other words, it must match \f(CW\*(C`/^[_A\-Za\-z][_A\-Za\-z0\-9]*\ez/\*(C'\fR or
2013 its Unicode extension (see utf8),
2014 though it isn't extended by the locale (see perllocale).
2015 .Sp
2016 \&\fBNOTE:\fR In order to make things easier for programmers with experience
2017 with the Python or PCRE regex engines, the pattern \f(CW\*(C`(?P<\fR\f(CINAME\fR\f(CW>\fR\f(CIpattern\fR\f(CW)\*(C'\fR
2018 may be used instead of \f(CW\*(C`(?<\fR\f(CINAME\fR\f(CW>\fR\f(CIpattern\fR\f(CW)\*(C'\fR; however this form does not
2019 support the use of single quotes as a delimiter for the name.
2020 .ie n .IP """\ek<\fINAME\fR>""" 4
2021 .el .IP \f(CW\ek<\fR\f(CINAME\fR\f(CW>\fR 4
2022 .IX Item "k<NAME>"
2023 .PD 0
2024 .ie n .IP """\ek\*(Aq\fINAME\fR\*(Aq""" 4
2025 .el .IP \f(CW\ek\*(Aq\fR\f(CINAME\fR\f(CW\*(Aq\fR 4
2026 .IX Item "kNAME"
2027 .ie n .IP """\ek{\fINAME\fR}""" 4
2028 .el .IP \f(CW\ek{\fR\f(CINAME\fR\f(CW}\fR 4
2029 .IX Item "k{NAME}"
2030 .PD
2031 Named backreference. Similar to numeric backreferences, except that
2032 the group is designated by name and not number. If multiple groups
2033 have the same name then it refers to the leftmost defined group in
2034 the current match.
2035 .Sp
2036 It is an error to refer to a name not defined by a \f(CW\*(C`(?<\fR\f(CINAME\fR\f(CW>)\*(C'\fR
2037 earlier in the pattern.
2038 .Sp
2039 All three forms are equivalent, although with \f(CW\*(C`\ek{ \fR\f(CINAME\fR\f(CW }\*(C'\fR,
2040 you may optionally have blanks within but adjacent to the braces, as
2041 shown.
2042 .Sp
2043 \&\fBNOTE:\fR In order to make things easier for programmers with experience
2044 with the Python or PCRE regex engines, the pattern \f(CW\*(C`(?P=\fR\f(CINAME\fR\f(CW)\*(C'\fR
2045 may be used instead of \f(CW\*(C`\ek<\fR\f(CINAME\fR\f(CW>\*(C'\fR.
2046 .ie n .IP """(?{ \fIcode\fR })""" 4
2047 .el .IP "\f(CW(?{ \fR\f(CIcode\fR\f(CW })\fR" 4
2048 .IX Xref "(?{}) regex, code in regexp, code in regular expression, code in"
2049 .IX Item "(?{ code })"
2050 \&\fBWARNING\fR: Using this feature safely requires that you understand its
2051 limitations. Code executed that has side effects may not perform identically
2052 from version to version due to the effect of future optimisations in the regex
2053 engine. For more information on this, see "Embedded Code Execution
2054 Frequency".
2055 .Sp
2056 This zero-width assertion executes any embedded Perl code. It always
2057 succeeds, and its return value is set as \f(CW$^R\fR.
2058 .Sp
2059 In literal patterns, the code is parsed at the same time as the
2060 surrounding code. While within the pattern, control is passed temporarily
2061 back to the perl parser, until the logically-balancing closing brace is
2062 encountered. This is similar to the way that an array index expression in
2063 a literal string is handled, for example
2064 .Sp
2065 .Vb 1
2066 \& "abc$array[ 1 + f(\*(Aq[\*(Aq) + g()]def"
2067 .Ve
2068 .Sp
2069 In particular, braces do not need to be balanced:
2070 .Sp
2071 .Vb 1
2072 \& s/abc(?{ f(\*(Aq{\*(Aq); })/def/
2073 .Ve
2074 .Sp
2075 Even in a pattern that is interpolated and compiled at run-time, literal
2076 code blocks will be compiled once, at perl compile time; the following
2077 prints "ABCD":
2078 .Sp
2079 .Vb 5
2080 \& print "D";
2081 \& my $qr = qr/(?{ BEGIN { print "A" } })/;
2082 \& my $foo = "foo";
2083 \& /$foo$qr(?{ BEGIN { print "B" } })/;
2084 \& BEGIN { print "C" }
2085 .Ve
2086 .Sp
2087 In patterns where the text of the code is derived from run-time
2088 information rather than appearing literally in a source code /pattern/,
2089 the code is compiled at the same time that the pattern is compiled, and
2090 for reasons of security, \f(CW\*(C`use re \*(Aqeval\*(Aq\*(C'\fR must be in scope. This is to
2091 stop user-supplied patterns containing code snippets from being
2092 executable.
2093 .Sp
2094 In situations where you need to enable this with \f(CW\*(C`use re \*(Aqeval\*(Aq\*(C'\fR, you should
2095 also have taint checking enabled, if your perl supports it.
2096 Better yet, use the carefully constrained evaluation within a Safe compartment.
2097 See perlsec for details about both these mechanisms.
2098 .Sp
2099 From the viewpoint of parsing, lexical variable scope and closures,
2100 .Sp
2101 .Vb 1
2102 \& /AAA(?{ BBB })CCC/
2103 .Ve
2104 .Sp
2105 behaves approximately like
2106 .Sp
2107 .Vb 1
2108 \& /AAA/ && do { BBB } && /CCC/
2109 .Ve
2110 .Sp
2111 Similarly,
2112 .Sp
2113 .Vb 1
2114 \& qr/AAA(?{ BBB })CCC/
2115 .Ve
2116 .Sp
2117 behaves approximately like
2118 .Sp
2119 .Vb 1
2120 \& sub { /AAA/ && do { BBB } && /CCC/ }
2121 .Ve
2122 .Sp
2123 In particular:
2124 .Sp
2125 .Vb 3
2126 \& { my $i = 1; $r = qr/(?{ print $i })/ }
2127 \& my $i = 2;
2128 \& /$r/; # prints "1"
2129 .Ve
2130 .Sp
2131 Inside a \f(CW\*(C`(?{...})\*(C'\fR block, \f(CW$_\fR refers to the string the regular
2132 expression is matching against. You can also use \f(CWpos()\fR to know what is
2133 the current position of matching within this string.
2134 .Sp
2135 The code block introduces a new scope from the perspective of lexical
2136 variable declarations, but \fBnot\fR from the perspective of \f(CW\*(C`local\*(C'\fR and
2137 similar localizing behaviours. So later code blocks within the same
2138 pattern will still see the values which were localized in earlier blocks.
2139 These accumulated localizations are undone either at the end of a
2140 successful match, or if the assertion is backtracked (compare
2141 "Backtracking"). For example,
2142 .Sp
2143 .Vb 10
2144 \& $_ = \*(Aqa\*(Aq x 8;
2145 \& m<
2146 \& (?{ $cnt = 0 }) # Initialize $cnt.
2147 \& (
2148 \& a
2149 \& (?{
2150 \& local $cnt = $cnt + 1; # Update $cnt,
2151 \& # backtracking\-safe.
2152 \& })
2153 \& )*
2154 \& aaaa
2155 \& (?{ $res = $cnt }) # On success copy to
2156 \& # non\-localized location.
2157 \& >x;
2158 .Ve
2159 .Sp
2160 will initially increment \f(CW$cnt\fR up to 8; then during backtracking, its
2161 value will be unwound back to 4, which is the value assigned to \f(CW$res\fR.
2162 At the end of the regex execution, \f(CW$cnt\fR will be wound back to its initial
2163 value of 0.
2164 .Sp
2165 This assertion may be used as the condition in a
2166 .Sp
2167 .Vb 1
2168 \& (?(condition)yes\-pattern|no\-pattern)
2169 .Ve
2170 .Sp
2171 switch. If \fInot\fR used in this way, the result of evaluation of \fIcode\fR
2172 is put into the special variable \f(CW$^R\fR. This happens immediately, so
2173 \&\f(CW$^R\fR can be used from other \f(CW\*(C`(?{ \fR\f(CIcode\fR\f(CW })\*(C'\fR assertions inside the same
2174 regular expression.
2175 .Sp
2176 The assignment to \f(CW$^R\fR above is properly localized, so the old
2177 value of \f(CW$^R\fR is restored if the assertion is backtracked; compare
2178 "Backtracking".
2179 .Sp
2180 Note that the special variable \f(CW$^N\fR is particularly useful with code
2181 blocks to capture the results of submatches in variables without having to
2182 keep track of the number of nested parentheses. For example:
2183 .Sp
2184 .Vb 3
2185 \& $_ = "The brown fox jumps over the lazy dog";
2186 \& /the (\eS+)(?{ $color = $^N }) (\eS+)(?{ $animal = $^N })/i;
2187 \& print "color = $color, animal = $animal\en";
2188 .Ve
2189 .Sp
2190 The use of this construct disables some optimisations globally in the
2191 pattern, and the pattern may execute much slower as a consequence.
2192 Use a \f(CW\*(C`*\*(C'\fR instead of the \f(CW\*(C`?\*(C'\fR block to create an optimistic form of
2193 this construct. \f(CW\*(C`(*{ ... })\*(C'\fR should not disable any optimisations.
2194 .ie n .IP """(*{ \fIcode\fR })""" 4
2195 .el .IP "\f(CW(*{ \fR\f(CIcode\fR\f(CW })\fR" 4
2196 .IX Xref "(*{}) regex, optimistic code"
2197 .IX Item "(*{ code })"
2198 This is *exactly* the same as \f(CW\*(C`(?{ \fR\f(CIcode\fR\f(CW })\*(C'\fR with the exception
2199 that it does not disable \fBany\fR optimisations at all in the regex engine.
2200 How often it is executed may vary from perl release to perl release.
2201 In a failing match it may not even be executed at all.
2202 .ie n .IP """(??{ \fIcode\fR })""" 4
2203 .el .IP "\f(CW(??{ \fR\f(CIcode\fR\f(CW })\fR" 4
2204 .IX Xref "(??{}) regex, postponed regexp, postponed regular expression, postponed"
2205 .IX Item "(??{ code })"
2206 \&\fBWARNING\fR: Using this feature safely requires that you understand its
2207 limitations. Code executed that has side effects may not perform
2208 identically from version to version due to the effect of future
2209 optimisations in the regex engine. For more information on this, see
2210 "Embedded Code Execution Frequency".
2211 .Sp
2212 This is a "postponed" regular subexpression. It behaves in \fIexactly\fR the
2213 same way as a \f(CW\*(C`(?{ \fR\f(CIcode\fR\f(CW })\*(C'\fR code block as described above, except that
2214 its return value, rather than being assigned to \f(CW$^R\fR, is treated as a
2215 pattern, compiled if it's a string (or used as-is if its a qr// object),
2216 then matched as if it were inserted instead of this construct.
2217 .Sp
2218 During the matching of this sub-pattern, it has its own set of
2219 captures which are valid during the sub-match, but are discarded once
2220 control returns to the main pattern. For example, the following matches,
2221 with the inner pattern capturing "B" and matching "BB", while the outer
2222 pattern captures "A";
2223 .Sp
2224 .Vb 3
2225 \& my $inner = \*(Aq(.)\e1\*(Aq;
2226 \& "ABBA" =~ /^(.)(??{ $inner })\e1/;
2227 \& print $1; # prints "A";
2228 .Ve
2229 .Sp
2230 Note that this means that there is no way for the inner pattern to refer
2231 to a capture group defined outside. (The code block itself can use \f(CW$1\fR,
2232 \&\fIetc\fR., to refer to the enclosing pattern's capture groups.) Thus, although
2233 .Sp
2234 .Vb 1
2235 \& (\*(Aqa\*(Aq x 100)=~/(??{\*(Aq(.)\*(Aq x 100})/
2236 .Ve
2237 .Sp
2238 \&\fIwill\fR match, it will \fInot\fR set \f(CW$1\fR on exit.
2239 .Sp
2240 The following pattern matches a parenthesized group:
2241 .Sp
2242 .Vb 9
2243 \& $re = qr{
2244 \& \e(
2245 \& (?:
2246 \& (?> [^()]+ ) # Non\-parens without backtracking
2247 \& |
2248 \& (??{ $re }) # Group with matching parens
2249 \& )*
2250 \& \e)
2251 \& }x;
2252 .Ve
2253 .Sp
2254 See also
2255 \&\f(CW\*(C`(?\fR\f(CIPARNO\fR\f(CW)\*(C'\fR
2256 for a different, more efficient way to accomplish
2257 the same task.
2258 .Sp
2259 Executing a postponed regular expression too many times without
2260 consuming any input string will also result in a fatal error. The depth
2261 at which that happens is compiled into perl, so it can be changed with a
2262 custom build.
2263 .Sp
2264 The use of this construct disables some optimisations globally in the pattern,
2265 and the pattern may execute much slower as a consequence.
2266 .ie n .IP """(?\fIPARNO\fR)"" ""(?\-\fIPARNO\fR)"" ""(?+\fIPARNO\fR)"" ""(?R)"" ""(?0)""" 4
2267 .el .IP "\f(CW(?\fR\f(CIPARNO\fR\f(CW)\fR \f(CW(?\-\fR\f(CIPARNO\fR\f(CW)\fR \f(CW(?+\fR\f(CIPARNO\fR\f(CW)\fR \f(CW(?R)\fR \f(CW(?0)\fR" 4
2268 .IX Xref "(?PARNO) (?1) (?R) (?0) (?-1) (?+1) (?-PARNO) (?+PARNO) regex, recursive regexp, recursive regular expression, recursive regex, relative recursion GOSUB GOSTART"
2269 .IX Item "(?PARNO) (?-PARNO) (?+PARNO) (?R) (?0)"
2270 Recursive subpattern. Treat the contents of a given capture buffer in the
2271 current pattern as an independent subpattern and attempt to match it at
2272 the current position in the string. Information about capture state from
2273 the caller for things like backreferences is available to the subpattern,
2274 but capture buffers set by the subpattern are not visible to the caller.
2275 .Sp
2276 Similar to \f(CW\*(C`(??{ \fR\f(CIcode\fR\f(CW })\*(C'\fR except that it does not involve executing any
2277 code or potentially compiling a returned pattern string; instead it treats
2278 the part of the current pattern contained within a specified capture group
2279 as an independent pattern that must match at the current position. Also
2280 different is the treatment of capture buffers, unlike \f(CW\*(C`(??{ \fR\f(CIcode\fR\f(CW })\*(C'\fR
2281 recursive patterns have access to their caller's match state, so one can
2282 use backreferences safely.
2283 .Sp
2284 \&\fIPARNO\fR is a sequence of digits (not starting with 0) whose value reflects
2285 the paren-number of the capture group to recurse to. \f(CW\*(C`(?R)\*(C'\fR recurses to
2286 the beginning of the whole pattern. \f(CW\*(C`(?0)\*(C'\fR is an alternate syntax for
2287 \&\f(CW\*(C`(?R)\*(C'\fR. If \fIPARNO\fR is preceded by a plus or minus sign then it is assumed
2288 to be relative, with negative numbers indicating preceding capture groups
2289 and positive ones following. Thus \f(CW\*(C`(?\-1)\*(C'\fR refers to the most recently
2290 declared group, and \f(CW\*(C`(?+1)\*(C'\fR indicates the next group to be declared.
2291 Note that the counting for relative recursion differs from that of
2292 relative backreferences, in that with recursion unclosed groups \fBare\fR
2293 included.
2294 .Sp
2295 The following pattern matches a function \f(CWfoo()\fR which may contain
2296 balanced parentheses as the argument.
2297 .Sp
2298 .Vb 10
2299 \& $re = qr{ ( # paren group 1 (full function)
2300 \& foo
2301 \& ( # paren group 2 (parens)
2302 \& \e(
2303 \& ( # paren group 3 (contents of parens)
2304 \& (?:
2305 \& (?> [^()]+ ) # Non\-parens without backtracking
2306 \& |
2307 \& (?2) # Recurse to start of paren group 2
2308 \& )*
2309 \& )
2310 \& \e)
2311 \& )
2312 \& )
2313 \& }x;
2314 .Ve
2315 .Sp
2316 If the pattern was used as follows
2317 .Sp
2318 .Vb 4
2319 \& \*(Aqfoo(bar(baz)+baz(bop))\*(Aq=~/$re/
2320 \& and print "\e$1 = $1\en",
2321 \& "\e$2 = $2\en",
2322 \& "\e$3 = $3\en";
2323 .Ve
2324 .Sp
2325 the output produced should be the following:
2326 .Sp
2327 .Vb 3
2328 \& $1 = foo(bar(baz)+baz(bop))
2329 \& $2 = (bar(baz)+baz(bop))
2330 \& $3 = bar(baz)+baz(bop)
2331 .Ve
2332 .Sp
2333 If there is no corresponding capture group defined, then it is a
2334 fatal error. Recursing deeply without consuming any input string will
2335 also result in a fatal error. The depth at which that happens is
2336 compiled into perl, so it can be changed with a custom build.
2337 .Sp
2338 The following shows how using negative indexing can make it
2339 easier to embed recursive patterns inside of a \f(CW\*(C`qr//\*(C'\fR construct
2340 for later use:
2341 .Sp
2342 .Vb 4
2343 \& my $parens = qr/(\e((?:[^()]++|(?\-1))*+\e))/;
2344 \& if (/foo $parens \es+ \e+ \es+ bar $parens/x) {
2345 \& # do something here...
2346 \& }
2347 .Ve
2348 .Sp
2349 \&\fBNote\fR that this pattern does not behave the same way as the equivalent
2350 PCRE or Python construct of the same form. In Perl you can backtrack into
2351 a recursed group, in PCRE and Python the recursed into group is treated
2352 as atomic. Also, modifiers are resolved at compile time, so constructs
2353 like \f(CW\*(C`(?i:(?1))\*(C'\fR or \f(CW\*(C`(?:(?i)(?1))\*(C'\fR do not affect how the sub-pattern will
2354 be processed.
2355 .ie n .IP """(?&\fINAME\fR)""" 4
2356 .el .IP \f(CW(?&\fR\f(CINAME\fR\f(CW)\fR 4
2357 .IX Xref "(?&NAME)"
2358 .IX Item "(?&NAME)"
2359 Recurse to a named subpattern. Identical to \f(CW\*(C`(?\fR\f(CIPARNO\fR\f(CW)\*(C'\fR except that the
2360 parenthesis to recurse to is determined by name. If multiple parentheses have
2361 the same name, then it recurses to the leftmost.
2362 .Sp
2363 It is an error to refer to a name that is not declared somewhere in the
2364 pattern.
2365 .Sp
2366 \&\fBNOTE:\fR In order to make things easier for programmers with experience
2367 with the Python or PCRE regex engines the pattern \f(CW\*(C`(?P>\fR\f(CINAME\fR\f(CW)\*(C'\fR
2368 may be used instead of \f(CW\*(C`(?&\fR\f(CINAME\fR\f(CW)\*(C'\fR.
2369 .ie n .IP """(?(\fIcondition\fR)\fIyes\-pattern\fR|\fIno\-pattern\fR)""" 4
2370 .el .IP \f(CW(?(\fR\f(CIcondition\fR\f(CW)\fR\f(CIyes\-pattern\fR\f(CW|\fR\f(CIno\-pattern\fR\f(CW)\fR 4
2371 .IX Xref "(?()"
2372 .IX Item "(?(condition)yes-pattern|no-pattern)"
2373 .PD 0
2374 .ie n .IP """(?(\fIcondition\fR)\fIyes\-pattern\fR)""" 4
2375 .el .IP \f(CW(?(\fR\f(CIcondition\fR\f(CW)\fR\f(CIyes\-pattern\fR\f(CW)\fR 4
2376 .IX Item "(?(condition)yes-pattern)"
2377 .PD
2378 Conditional expression. Matches \fIyes-pattern\fR if \fIcondition\fR yields
2379 a true value, matches \fIno-pattern\fR otherwise. A missing pattern always
2380 matches.
2381 .Sp
2382 \&\f(CW\*(C`(\fR\f(CIcondition\fR\f(CW)\*(C'\fR should be one of:
2383 .RS 4
2384 .IP "an integer in parentheses" 4
2385 .IX Item "an integer in parentheses"
2386 (which is valid if the corresponding pair of parentheses
2387 matched);
2388 .IP "a lookahead/lookbehind/evaluate zero-width assertion;" 4
2389 .IX Item "a lookahead/lookbehind/evaluate zero-width assertion;"
2390 .PD 0
2391 .IP "a name in angle brackets or single quotes" 4
2392 .IX Item "a name in angle brackets or single quotes"
2393 .PD
2394 (which is valid if a group with the given name matched);
2395 .ie n .IP "the special symbol ""(R)""" 4
2396 .el .IP "the special symbol \f(CW(R)\fR" 4
2397 .IX Item "the special symbol (R)"
2398 (true when evaluated inside of recursion or eval). Additionally the
2399 \&\f(CW"R"\fR may be
2400 followed by a number, (which will be true when evaluated when recursing
2401 inside of the appropriate group), or by \f(CW\*(C`&\fR\f(CINAME\fR\f(CW\*(C'\fR, in which case it will
2402 be true only when evaluated during recursion in the named group.
2403 .RE
2404 .RS 4
2405 .Sp
2406 Here's a summary of the possible predicates:
2407 .ie n .IP """(1)"" ""(2)"" ..." 4
2408 .el .IP "\f(CW(1)\fR \f(CW(2)\fR ..." 4
2409 .IX Item "(1) (2) ..."
2410 Checks if the numbered capturing group has matched something.
2411 Full syntax: \f(CW\*(C`(?(1)then|else)\*(C'\fR
2412 .ie n .IP """(<\fINAME\fR>)"" ""(\*(Aq\fINAME\fR\*(Aq)""" 4
2413 .el .IP "\f(CW(<\fR\f(CINAME\fR\f(CW>)\fR \f(CW(\*(Aq\fR\f(CINAME\fR\f(CW\*(Aq)\fR" 4
2414 .IX Item "(<NAME>) (NAME)"
2415 Checks if a group with the given name has matched something.
2416 Full syntax: \f(CW\*(C`(?(<name>)then|else)\*(C'\fR
2417 .ie n .IP """(?=...)"" ""(?!...)"" ""(?<=...)"" ""(?<!...)""" 4
2418 .el .IP "\f(CW(?=...)\fR \f(CW(?!...)\fR \f(CW(?<=...)\fR \f(CW(?<!...)\fR" 4
2419 .IX Item "(?=...) (?!...) (?<=...) (?<!...)"
2420 Checks whether the pattern matches (or does not match, for the \f(CW"!"\fR
2421 variants).
2422 Full syntax: \f(CW\*(C`(?(?=\fR\f(CIlookahead\fR\f(CW)\fR\f(CIthen\fR\f(CW|\fR\f(CIelse\fR\f(CW)\*(C'\fR
2423 .ie n .IP """(?{ \fICODE\fR })""" 4
2424 .el .IP "\f(CW(?{ \fR\f(CICODE\fR\f(CW })\fR" 4
2425 .IX Item "(?{ CODE })"
2426 Treats the return value of the code block as the condition.
2427 Full syntax: \f(CW\*(C`(?(?{ \fR\f(CICODE\fR\f(CW })\fR\f(CIthen\fR\f(CW|\fR\f(CIelse\fR\f(CW)\*(C'\fR
2428 .Sp
2429 Note use of this construct may globally affect the performance
2430 of the pattern. Consider using \f(CW\*(C`(*{ \fR\f(CICODE\fR\f(CW })\*(C'\fR
2431 .ie n .IP """(*{ \fICODE\fR })""" 4
2432 .el .IP "\f(CW(*{ \fR\f(CICODE\fR\f(CW })\fR" 4
2433 .IX Item "(*{ CODE })"
2434 Treats the return value of the code block as the condition.
2435 Full syntax: \f(CW\*(C`(?(*{ \fR\f(CICODE\fR\f(CW })\fR\f(CIthen\fR\f(CW|\fR\f(CIelse\fR\f(CW)\*(C'\fR
2436 .ie n .IP """(R)""" 4
2437 .el .IP \f(CW(R)\fR 4
2438 .IX Item "(R)"
2439 Checks if the expression has been evaluated inside of recursion.
2440 Full syntax: \f(CW\*(C`(?(R)\fR\f(CIthen\fR\f(CW|\fR\f(CIelse\fR\f(CW)\*(C'\fR
2441 .ie n .IP """(R1)"" ""(R2)"" ..." 4
2442 .el .IP "\f(CW(R1)\fR \f(CW(R2)\fR ..." 4
2443 .IX Item "(R1) (R2) ..."
2444 Checks if the expression has been evaluated while executing directly
2445 inside of the n\-th capture group. This check is the regex equivalent of
2446 .Sp
2447 .Vb 1
2448 \& if ((caller(0))[3] eq \*(Aqsubname\*(Aq) { ... }
2449 .Ve
2450 .Sp
2451 In other words, it does not check the full recursion stack.
2452 .Sp
2453 Full syntax: \f(CW\*(C`(?(R1)\fR\f(CIthen\fR\f(CW|\fR\f(CIelse\fR\f(CW)\*(C'\fR
2454 .ie n .IP """(R&\fINAME\fR)""" 4
2455 .el .IP \f(CW(R&\fR\f(CINAME\fR\f(CW)\fR 4
2456 .IX Item "(R&NAME)"
2457 Similar to \f(CW\*(C`(R1)\*(C'\fR, this predicate checks to see if we're executing
2458 directly inside of the leftmost group with a given name (this is the same
2459 logic used by \f(CW\*(C`(?&\fR\f(CINAME\fR\f(CW)\*(C'\fR to disambiguate). It does not check the full
2460 stack, but only the name of the innermost active recursion.
2461 Full syntax: \f(CW\*(C`(?(R&\fR\f(CIname\fR\f(CW)\fR\f(CIthen\fR\f(CW|\fR\f(CIelse\fR\f(CW)\*(C'\fR
2462 .ie n .IP """(DEFINE)""" 4
2463 .el .IP \f(CW(DEFINE)\fR 4
2464 .IX Item "(DEFINE)"
2465 In this case, the yes-pattern is never directly executed, and no
2466 no-pattern is allowed. Similar in spirit to \f(CW\*(C`(?{0})\*(C'\fR but more efficient.
2467 See below for details.
2468 Full syntax: \f(CW\*(C`(?(DEFINE)\fR\f(CIdefinitions\fR\f(CW...)\*(C'\fR
2469 .RE
2470 .RS 4
2471 .Sp
2472 For example:
2473 .Sp
2474 .Vb 4
2475 \& m{ ( \e( )?
2476 \& [^()]+
2477 \& (?(1) \e) )
2478 \& }x
2479 .Ve
2480 .Sp
2481 matches a chunk of non-parentheses, possibly included in parentheses
2482 themselves.
2483 .Sp
2484 A special form is the \f(CW\*(C`(DEFINE)\*(C'\fR predicate, which never executes its
2485 yes-pattern directly, and does not allow a no-pattern. This allows one to
2486 define subpatterns which will be executed only by the recursion mechanism.
2487 This way, you can define a set of regular expression rules that can be
2488 bundled into any pattern you choose.
2489 .Sp
2490 It is recommended that for this usage you put the DEFINE block at the
2491 end of the pattern, and that you name any subpatterns defined within it.
2492 .Sp
2493 Also, it's worth noting that patterns defined this way probably will
2494 not be as efficient, as the optimizer is not very clever about
2495 handling them.
2496 .Sp
2497 An example of how this might be used is as follows:
2498 .Sp
2499 .Vb 5
2500 \& /(?<NAME>(?&NAME_PAT))(?<ADDR>(?&ADDRESS_PAT))
2501 \& (?(DEFINE)
2502 \& (?<NAME_PAT>....)
2503 \& (?<ADDRESS_PAT>....)
2504 \& )/x
2505 .Ve
2506 .Sp
2507 Note that capture groups matched inside of recursion are not accessible
2508 after the recursion returns, so the extra layer of capturing groups is
2509 necessary. Thus \f(CW$+{NAME_PAT}\fR would not be defined even though
2510 \&\f(CW$+{NAME}\fR would be.
2511 .Sp
2512 Finally, keep in mind that subpatterns created inside a DEFINE block
2513 count towards the absolute and relative number of captures, so this:
2514 .Sp
2515 .Vb 5
2516 \& my @captures = "a" =~ /(.) # First capture
2517 \& (?(DEFINE)
2518 \& (?<EXAMPLE> 1 ) # Second capture
2519 \& )/x;
2520 \& say scalar @captures;
2521 .Ve
2522 .Sp
2523 Will output 2, not 1. This is particularly important if you intend to
2524 compile the definitions with the \f(CW\*(C`qr//\*(C'\fR operator, and later
2525 interpolate them in another pattern.
2526 .RE
2527 .ie n .IP """(?>\fIpattern\fR)""" 4
2528 .el .IP \f(CW(?>\fR\f(CIpattern\fR\f(CW)\fR 4
2529 .IX Item "(?>pattern)"
2530 .PD 0
2531 .ie n .IP """(*atomic:\fIpattern\fR)""" 4
2532 .el .IP \f(CW(*atomic:\fR\f(CIpattern\fR\f(CW)\fR 4
2533 .IX Xref "(?>pattern) (*atomic backtrack backtracking atomic possessive"
2534 .IX Item "(*atomic:pattern)"
2535 .PD
2536 An "independent" subexpression, one which matches the substring
2537 that a standalone \fIpattern\fR would match if anchored at the given
2538 position, and it matches \fInothing other than this substring\fR. This
2539 construct is useful for optimizations of what would otherwise be
2540 "eternal" matches, because it will not backtrack (see "Backtracking").
2541 It may also be useful in places where the "grab all you can, and do not
2542 give anything back" semantic is desirable.
2543 .Sp
2544 For example: \f(CW\*(C`^(?>a*)ab\*(C'\fR will never match, since \f(CW\*(C`(?>a*)\*(C'\fR
2545 (anchored at the beginning of string, as above) will match \fIall\fR
2546 characters \f(CW"a"\fR at the beginning of string, leaving no \f(CW"a"\fR for
2547 \&\f(CW\*(C`ab\*(C'\fR to match. In contrast, \f(CW\*(C`a*ab\*(C'\fR will match the same as \f(CW\*(C`a+b\*(C'\fR,
2548 since the match of the subgroup \f(CW\*(C`a*\*(C'\fR is influenced by the following
2549 group \f(CW\*(C`ab\*(C'\fR (see "Backtracking"). In particular, \f(CW\*(C`a*\*(C'\fR inside
2550 \&\f(CW\*(C`a*ab\*(C'\fR will match fewer characters than a standalone \f(CW\*(C`a*\*(C'\fR, since
2551 this makes the tail match.
2552 .Sp
2553 \&\f(CW\*(C`(?>\fR\f(CIpattern\fR\f(CW)\*(C'\fR does not disable backtracking altogether once it has
2554 matched. It is still possible to backtrack past the construct, but not
2555 into it. So \f(CW\*(C`((?>a*)|(?>b*))ar\*(C'\fR will still match "bar".
2556 .Sp
2557 An effect similar to \f(CW\*(C`(?>\fR\f(CIpattern\fR\f(CW)\*(C'\fR may be achieved by writing
2558 \&\f(CW\*(C`(?=(\fR\f(CIpattern\fR\f(CW))\eg{\-1}\*(C'\fR. This matches the same substring as a standalone
2559 \&\f(CW\*(C`a+\*(C'\fR, and the following \f(CW\*(C`\eg{\-1}\*(C'\fR eats the matched string; it therefore
2560 makes a zero-length assertion into an analogue of \f(CW\*(C`(?>...)\*(C'\fR.
2561 (The difference between these two constructs is that the second one
2562 uses a capturing group, thus shifting ordinals of backreferences
2563 in the rest of a regular expression.)
2564 .Sp
2565 Consider this pattern:
2566 .Sp
2567 .Vb 8
2568 \& m{ \e(
2569 \& (
2570 \& [^()]+ # x+
2571 \& |
2572 \& \e( [^()]* \e)
2573 \& )+
2574 \& \e)
2575 \& }x
2576 .Ve
2577 .Sp
2578 That will efficiently match a nonempty group with matching parentheses
2579 two levels deep or less. However, if there is no such group, it
2580 will take virtually forever on a long string. That's because there
2581 are so many different ways to split a long string into several
2582 substrings. This is what \f(CW\*(C`(.+)+\*(C'\fR is doing, and \f(CW\*(C`(.+)+\*(C'\fR is similar
2583 to a subpattern of the above pattern. Consider how the pattern
2584 above detects no-match on \f(CW\*(C`((()aaaaaaaaaaaaaaaaaa\*(C'\fR in several
2585 seconds, but that each extra letter doubles this time. This
2586 exponential performance will make it appear that your program has
2587 hung. However, a tiny change to this pattern
2588 .Sp
2589 .Vb 8
2590 \& m{ \e(
2591 \& (
2592 \& (?> [^()]+ ) # change x+ above to (?> x+ )
2593 \& |
2594 \& \e( [^()]* \e)
2595 \& )+
2596 \& \e)
2597 \& }x
2598 .Ve
2599 .Sp
2600 which uses \f(CW\*(C`(?>...)\*(C'\fR matches exactly when the one above does (verifying
2601 this yourself would be a productive exercise), but finishes in a fourth
2602 the time when used on a similar string with 1000000 \f(CW"a"\fRs. Be aware,
2603 however, that, when this construct is followed by a
2604 quantifier, it currently triggers a warning message under
2605 the \f(CW\*(C`use warnings\*(C'\fR pragma or \fB\-w\fR switch saying it
2606 \&\f(CW"matches null string many times in regex"\fR.
2607 .Sp
2608 On simple groups, such as the pattern \f(CW\*(C`(?> [^()]+ )\*(C'\fR, a comparable
2609 effect may be achieved by negative lookahead, as in \f(CW\*(C`[^()]+ (?! [^()] )\*(C'\fR.
2610 This was only 4 times slower on a string with 1000000 \f(CW"a"\fRs.
2611 .Sp
2612 The "grab all you can, and do not give anything back" semantic is desirable
2613 in many situations where on the first sight a simple \f(CW\*(C`()*\*(C'\fR looks like
2614 the correct solution. Suppose we parse text with comments being delimited
2615 by \f(CW"#"\fR followed by some optional (horizontal) whitespace. Contrary to
2616 its appearance, \f(CW\*(C`#[ \et]*\*(C'\fR \fIis not\fR the correct subexpression to match
2617 the comment delimiter, because it may "give up" some whitespace if
2618 the remainder of the pattern can be made to match that way. The correct
2619 answer is either one of these:
2620 .Sp
2621 .Vb 2
2622 \& (?>#[ \et]*)
2623 \& #[ \et]*(?![ \et])
2624 .Ve
2625 .Sp
2626 For example, to grab non-empty comments into \f(CW$1\fR, one should use either
2627 one of these:
2628 .Sp
2629 .Vb 2
2630 \& / (?> \e# [ \et]* ) ( .+ ) /x;
2631 \& / \e# [ \et]* ( [^ \et] .* ) /x;
2632 .Ve
2633 .Sp
2634 Which one you pick depends on which of these expressions better reflects
2635 the above specification of comments.
2636 .Sp
2637 In some literature this construct is called "atomic matching" or
2638 "possessive matching".
2639 .Sp
2640 Possessive quantifiers are equivalent to putting the item they are applied
2641 to inside of one of these constructs. The following equivalences apply:
2642 .Sp
2643 .Vb 6
2644 \& Quantifier Form Bracketing Form
2645 \& \-\-\-\-\-\-\-\-\-\-\-\-\-\-\- \-\-\-\-\-\-\-\-\-\-\-\-\-\-\-
2646 \& PAT*+ (?>PAT*)
2647 \& PAT++ (?>PAT+)
2648 \& PAT?+ (?>PAT?)
2649 \& PAT{min,max}+ (?>PAT{min,max})
2650 .Ve
2651 .Sp
2652 Nested \f(CW\*(C`(?>...)\*(C'\fR constructs are not no-ops, even if at first glance
2653 they might seem to be. This is because the nested \f(CW\*(C`(?>...)\*(C'\fR can
2654 restrict internal backtracking that otherwise might occur. For example,
2655 .Sp
2656 .Vb 1
2657 \& "abc" =~ /(?>a[bc]*c)/
2658 .Ve
2659 .Sp
2660 matches, but
2661 .Sp
2662 .Vb 1
2663 \& "abc" =~ /(?>a(?>[bc]*)c)/
2664 .Ve
2665 .Sp
2666 does not.
2667 .ie n .IP """(?[ ])""" 4
2668 .el .IP "\f(CW(?[ ])\fR" 4
2669 .IX Item "(?[ ])"
2670 See "Extended Bracketed Character Classes" in perlrecharclass.
2671 .SS Backtracking
2672 .IX Xref "backtrack backtracking"
2673 .IX Subsection "Backtracking"
2674 NOTE: This section presents an abstract approximation of regular
2675 expression behavior. For a more rigorous (and complicated) view of
2676 the rules involved in selecting a match among possible alternatives,
2677 see "Combining RE Pieces".
2678 .PP
2679 A fundamental feature of regular expression matching involves the
2680 notion called \fIbacktracking\fR, which is currently used (when needed)
2681 by all regular non-possessive expression quantifiers, namely \f(CW"*"\fR,
2682 \&\f(CW\*(C`*?\*(C'\fR, \f(CW"+"\fR, \f(CW\*(C`+?\*(C'\fR, \f(CW\*(C`{n,m}\*(C'\fR, and \f(CW\*(C`{n,m}?\*(C'\fR. Backtracking is often
2683 optimized internally, but the general principle outlined here is valid.
2684 .PP
2685 For a regular expression to match, the \fIentire\fR regular expression must
2686 match, not just part of it. So if the beginning of a pattern containing a
2687 quantifier succeeds in a way that causes later parts in the pattern to
2688 fail, the matching engine backs up and recalculates the beginning
2689 part\-\-that's why it's called backtracking.
2690 .PP
2691 Here is an example of backtracking: Let's say you want to find the
2692 word following "foo" in the string "Food is on the foo table.":
2693 .PP
2694 .Vb 4
2695 \& $_ = "Food is on the foo table.";
2696 \& if ( /\eb(foo)\es+(\ew+)/i ) {
2697 \& print "$2 follows $1.\en";
2698 \& }
2699 .Ve
2700 .PP
2701 When the match runs, the first part of the regular expression (\f(CW\*(C`\eb(foo)\*(C'\fR)
2702 finds a possible match right at the beginning of the string, and loads up
2703 \&\f(CW$1\fR with "Foo". However, as soon as the matching engine sees that there's
2704 no whitespace following the "Foo" that it had saved in \f(CW$1\fR, it realizes its
2705 mistake and starts over again one character after where it had the
2706 tentative match. This time it goes all the way until the next occurrence
2707 of "foo". The complete regular expression matches this time, and you get
2708 the expected output of "table follows foo."
2709 .PP
2710 Sometimes minimal matching can help a lot. Imagine you'd like to match
2711 everything between "foo" and "bar". Initially, you write something
2712 like this:
2713 .PP
2714 .Vb 4
2715 \& $_ = "The food is under the bar in the barn.";
2716 \& if ( /foo(.*)bar/ ) {
2717 \& print "got <$1>\en";
2718 \& }
2719 .Ve
2720 .PP
2721 Which perhaps unexpectedly yields:
2722 .PP
2723 .Vb 1
2724 \& got <d is under the bar in the >
2725 .Ve
2726 .PP
2727 That's because \f(CW\*(C`.*\*(C'\fR was greedy, so you get everything between the
2728 \&\fIfirst\fR "foo" and the \fIlast\fR "bar". Here it's more effective
2729 to use minimal matching to make sure you get the text between a "foo"
2730 and the first "bar" thereafter.
2731 .PP
2732 .Vb 2
2733 \& if ( /foo(.*?)bar/ ) { print "got <$1>\en" }
2734 \& got <d is under the >
2735 .Ve
2736 .PP
2737 Here's another example. Let's say you'd like to match a number at the end
2738 of a string, and you also want to keep the preceding part of the match.
2739 So you write this:
2740 .PP
2741 .Vb 4
2742 \& $_ = "I have 2 numbers: 53147";
2743 \& if ( /(.*)(\ed*)/ ) { # Wrong!
2744 \& print "Beginning is <$1>, number is <$2>.\en";
2745 \& }
2746 .Ve
2747 .PP
2748 That won't work at all, because \f(CW\*(C`.*\*(C'\fR was greedy and gobbled up the
2749 whole string. As \f(CW\*(C`\ed*\*(C'\fR can match on an empty string the complete
2750 regular expression matched successfully.
2751 .PP
2752 .Vb 1
2753 \& Beginning is <I have 2 numbers: 53147>, number is <>.
2754 .Ve
2755 .PP
2756 Here are some variants, most of which don't work:
2757 .PP
2758 .Vb 11
2759 \& $_ = "I have 2 numbers: 53147";
2760 \& @pats = qw{
2761 \& (.*)(\ed*)
2762 \& (.*)(\ed+)
2763 \& (.*?)(\ed*)
2764 \& (.*?)(\ed+)
2765 \& (.*)(\ed+)$
2766 \& (.*?)(\ed+)$
2767 \& (.*)\eb(\ed+)$
2768 \& (.*\eD)(\ed+)$
2769 \& };
2770 \&
2771 \& for $pat (@pats) {
2772 \& printf "%\-12s ", $pat;
2773 \& if ( /$pat/ ) {
2774 \& print "<$1> <$2>\en";
2775 \& } else {
2776 \& print "FAIL\en";
2777 \& }
2778 \& }
2779 .Ve
2780 .PP
2781 That will print out:
2782 .PP
2783 .Vb 8
2784 \& (.*)(\ed*) <I have 2 numbers: 53147> <>
2785 \& (.*)(\ed+) <I have 2 numbers: 5314> <7>
2786 \& (.*?)(\ed*) <> <>
2787 \& (.*?)(\ed+) <I have > <2>
2788 \& (.*)(\ed+)$ <I have 2 numbers: 5314> <7>
2789 \& (.*?)(\ed+)$ <I have 2 numbers: > <53147>
2790 \& (.*)\eb(\ed+)$ <I have 2 numbers: > <53147>
2791 \& (.*\eD)(\ed+)$ <I have 2 numbers: > <53147>
2792 .Ve
2793 .PP
2794 As you see, this can be a bit tricky. It's important to realize that a
2795 regular expression is merely a set of assertions that gives a definition
2796 of success. There may be 0, 1, or several different ways that the
2797 definition might succeed against a particular string. And if there are
2798 multiple ways it might succeed, you need to understand backtracking to
2799 know which variety of success you will achieve.
2800 .PP
2801 When using lookahead assertions and negations, this can all get even
2802 trickier. Imagine you'd like to find a sequence of non-digits not
2803 followed by "123". You might try to write that as
2804 .PP
2805 .Vb 4
2806 \& $_ = "ABC123";
2807 \& if ( /^\eD*(?!123)/ ) { # Wrong!
2808 \& print "Yup, no 123 in $_\en";
2809 \& }
2810 .Ve
2811 .PP
2812 But that isn't going to match; at least, not the way you're hoping. It
2813 claims that there is no 123 in the string. Here's a clearer picture of
2814 why that pattern matches, contrary to popular expectations:
2815 .PP
2816 .Vb 2
2817 \& $x = \*(AqABC123\*(Aq;
2818 \& $y = \*(AqABC445\*(Aq;
2819 \&
2820 \& print "1: got $1\en" if $x =~ /^(ABC)(?!123)/;
2821 \& print "2: got $1\en" if $y =~ /^(ABC)(?!123)/;
2822 \&
2823 \& print "3: got $1\en" if $x =~ /^(\eD*)(?!123)/;
2824 \& print "4: got $1\en" if $y =~ /^(\eD*)(?!123)/;
2825 .Ve
2826 .PP
2827 This prints
2828 .PP
2829 .Vb 3
2830 \& 2: got ABC
2831 \& 3: got AB
2832 \& 4: got ABC
2833 .Ve
2834 .PP
2835 You might have expected test 3 to fail because it seems to a more
2836 general purpose version of test 1. The important difference between
2837 them is that test 3 contains a quantifier (\f(CW\*(C`\eD*\*(C'\fR) and so can use
2838 backtracking, whereas test 1 will not. What's happening is
2839 that you've asked "Is it true that at the start of \f(CW$x\fR, following 0 or more
2840 non-digits, you have something that's not 123?" If the pattern matcher had
2841 let \f(CW\*(C`\eD*\*(C'\fR expand to "ABC", this would have caused the whole pattern to
2842 fail.
2843 .PP
2844 The search engine will initially match \f(CW\*(C`\eD*\*(C'\fR with "ABC". Then it will
2845 try to match \f(CW\*(C`(?!123)\*(C'\fR with "123", which fails. But because
2846 a quantifier (\f(CW\*(C`\eD*\*(C'\fR) has been used in the regular expression, the
2847 search engine can backtrack and retry the match differently
2848 in the hope of matching the complete regular expression.
2849 .PP
2850 The pattern really, \fIreally\fR wants to succeed, so it uses the
2851 standard pattern back-off-and-retry and lets \f(CW\*(C`\eD*\*(C'\fR expand to just "AB" this
2852 time. Now there's indeed something following "AB" that is not
2853 "123". It's "C123", which suffices.
2854 .PP
2855 We can deal with this by using both an assertion and a negation.
2856 We'll say that the first part in \f(CW$1\fR must be followed both by a digit
2857 and by something that's not "123". Remember that the lookaheads
2858 are zero-width expressions\-\-they only look, but don't consume any
2859 of the string in their match. So rewriting this way produces what
2860 you'd expect; that is, case 5 will fail, but case 6 succeeds:
2861 .PP
2862 .Vb 2
2863 \& print "5: got $1\en" if $x =~ /^(\eD*)(?=\ed)(?!123)/;
2864 \& print "6: got $1\en" if $y =~ /^(\eD*)(?=\ed)(?!123)/;
2865 \&
2866 \& 6: got ABC
2867 .Ve
2868 .PP
2869 In other words, the two zero-width assertions next to each other work as though
2870 they're ANDed together, just as you'd use any built-in assertions: \f(CW\*(C`/^$/\*(C'\fR
2871 matches only if you're at the beginning of the line AND the end of the
2872 line simultaneously. The deeper underlying truth is that juxtaposition in
2873 regular expressions always means AND, except when you write an explicit OR
2874 using the vertical bar. \f(CW\*(C`/ab/\*(C'\fR means match "a" AND (then) match "b",
2875 although the attempted matches are made at different positions because "a"
2876 is not a zero-width assertion, but a one-width assertion.
2877 .PP
2878 \&\fBWARNING\fR: Particularly complicated regular expressions can take
2879 exponential time to solve because of the immense number of possible
2880 ways they can use backtracking to try for a match. For example, without
2881 internal optimizations done by the regular expression engine, this will
2882 take a painfully long time to run:
2883 .PP
2884 .Vb 1
2885 \& \*(Aqaaaaaaaaaaaa\*(Aq =~ /((a{0,5}){0,5})*[c]/
2886 .Ve
2887 .PP
2888 And if you used \f(CW"*"\fR's in the internal groups instead of limiting them
2889 to 0 through 5 matches, then it would take forever\-\-or until you ran
2890 out of stack space. Moreover, these internal optimizations are not
2891 always applicable. For example, if you put \f(CW\*(C`{0,5}\*(C'\fR instead of \f(CW"*"\fR
2892 on the external group, no current optimization is applicable, and the
2893 match takes a long time to finish.
2894 .PP
2895 A powerful tool for optimizing such beasts is what is known as an
2896 "independent group",
2897 which does not backtrack (see \f(CW"(?>pattern)"\fR). Note also that
2898 zero-length lookahead/lookbehind assertions will not backtrack to make
2899 the tail match, since they are in "logical" context: only
2900 whether they match is considered relevant. For an example
2901 where side-effects of lookahead \fImight\fR have influenced the
2902 following match, see \f(CW"(?>pattern)"\fR.
2903 .SS "Script Runs"
2904 .IX Xref "(*script_run:...) (sr:...) (*atomic_script_run:...) (asr:...)"
2905 .IX Subsection "Script Runs"
2906 A script run is basically a sequence of characters, all from the same
2907 Unicode script (see "Scripts" in perlunicode), such as Latin or Greek. In
2908 most places a single word would never be written in multiple scripts,
2909 unless it is a spoofing attack. An infamous example, is
2910 .PP
2911 .Vb 1
2912 \& paypal.com
2913 .Ve
2914 .PP
2915 Those letters could all be Latin (as in the example just above), or they
2916 could be all Cyrillic (except for the dot), or they could be a mixture
2917 of the two. In the case of an internet address the \f(CW\*(C`.com\*(C'\fR would be in
2918 Latin, And any Cyrillic ones would cause it to be a mixture, not a
2919 script run. Someone clicking on such a link would not be directed to
2920 the real Paypal website, but an attacker would craft a look-alike one to
2921 attempt to gather sensitive information from the person.
2922 .PP
2923 Starting in Perl 5.28, it is now easy to detect strings that aren't
2924 script runs. Simply enclose just about any pattern like either of
2925 these:
2926 .PP
2927 .Vb 2
2928 \& (*script_run:pattern)
2929 \& (*sr:pattern)
2930 .Ve
2931 .PP
2932 What happens is that after \fIpattern\fR succeeds in matching, it is
2933 subjected to the additional criterion that every character in it must be
2934 from the same script (see exceptions below). If this isn't true,
2935 backtracking occurs until something all in the same script is found that
2936 matches, or all possibilities are exhausted. This can cause a lot of
2937 backtracking, but generally, only malicious input will result in this,
2938 though the slow down could cause a denial of service attack. If your
2939 needs permit, it is best to make the pattern atomic to cut down on the
2940 amount of backtracking. This is so likely to be what you want, that
2941 instead of writing this:
2942 .PP
2943 .Vb 1
2944 \& (*script_run:(?>pattern))
2945 .Ve
2946 .PP
2947 you can write either of these:
2948 .PP
2949 .Vb 2
2950 \& (*atomic_script_run:pattern)
2951 \& (*asr:pattern)
2952 .Ve
2953 .PP
2954 (See \f(CW"(?>\fR\f(CIpattern\fR\f(CW)"\fR.)
2955 .PP
2956 In Taiwan, Japan, and Korea, it is common for text to have a mixture of
2957 characters from their native scripts and base Chinese. Perl follows
2958 Unicode's UTS 39 (<https://unicode.org/reports/tr39/>) Unicode Security
2959 Mechanisms in allowing such mixtures. For example, the Japanese scripts
2960 Katakana and Hiragana are commonly mixed together in practice, along
2961 with some Chinese characters, and hence are treated as being in a single
2962 script run by Perl.
2963 .PP
2964 The rules used for matching decimal digits are slightly stricter. Many
2965 scripts have their own sets of digits equivalent to the Western \f(CW0\fR
2966 through \f(CW9\fR ones. A few, such as Arabic, have more than one set. For
2967 a string to be considered a script run, all digits in it must come from
2968 the same set of ten, as determined by the first digit encountered.
2969 As an example,
2970 .PP
2971 .Vb 1
2972 \& qr/(*script_run: \ed+ \eb )/x
2973 .Ve
2974 .PP
2975 guarantees that the digits matched will all be from the same set of 10.
2976 You won't get a look-alike digit from a different script that has a
2977 different value than what it appears to be.
2978 .PP
2979 Unicode has three pseudo scripts that are handled specially.
2980 .PP
2981 "Unknown" is applied to code points whose meaning has yet to be
2982 determined. Perl currently will match as a script run, any single
2983 character string consisting of one of these code points. But any string
2984 longer than one code point containing one of these will not be
2985 considered a script run.
2986 .PP
2987 "Inherited" is applied to characters that modify another, such as an
2988 accent of some type. These are considered to be in the script of the
2989 master character, and so never cause a script run to not match.
2990 .PP
2991 The other one is "Common". This consists of mostly punctuation, emoji,
2992 characters used in mathematics and music, the ASCII digits \f(CW0\fR
2993 through \f(CW9\fR, and full-width forms of these digits. These characters
2994 can appear intermixed in text in many of the world's scripts. These
2995 also don't cause a script run to not match. But like other scripts, all
2996 digits in a run must come from the same set of 10.
2997 .PP
2998 This construct is non-capturing. You can add parentheses to \fIpattern\fR
2999 to capture, if desired. You will have to do this if you plan to use
3000 "(*ACCEPT) (*ACCEPT:arg)" and not have it bypass the script run
3001 checking.
3002 .PP
3003 The \f(CW\*(C`Script_Extensions\*(C'\fR property as modified by UTS 39
3004 (<https://unicode.org/reports/tr39/>) is used as the basis for this
3005 feature.
3006 .PP
3007 To summarize,
3008 .IP \(bu 4
3009 All length 0 or length 1 sequences are script runs.
3010 .IP \(bu 4
3011 A longer sequence is a script run if and only if \fBall\fR of the following
3012 conditions are met:
3013 .Sp
3014
3015 .RS 4
3016 .IP 1. 4
3017 No code point in the sequence has the \f(CW\*(C`Script_Extension\*(C'\fR property of
3018 \&\f(CW\*(C`Unknown\*(C'\fR.
3019 .Sp
3020 This currently means that all code points in the sequence have been
3021 assigned by Unicode to be characters that aren't private use nor
3022 surrogate code points.
3023 .IP 2. 4
3024 All characters in the sequence come from the Common script and/or the
3025 Inherited script and/or a single other script.
3026 .Sp
3027 The script of a character is determined by the \f(CW\*(C`Script_Extensions\*(C'\fR
3028 property as modified by UTS 39 (<https://unicode.org/reports/tr39/>), as
3029 described above.
3030 .IP 3. 4
3031 All decimal digits in the sequence come from the same block of 10
3032 consecutive digits.
3033 .RE
3034 .RS 4
3035 .RE
3036 .SS "Special Backtracking Control Verbs"
3037 .IX Subsection "Special Backtracking Control Verbs"
3038 These special patterns are generally of the form \f(CW\*(C`(*\fR\f(CIVERB\fR\f(CW:\fR\f(CIarg\fR\f(CW)\*(C'\fR. Unless
3039 otherwise stated the \fIarg\fR argument is optional; in some cases, it is
3040 mandatory.
3041 .PP
3042 Any pattern containing a special backtracking verb that allows an argument
3043 has the special behaviour that when executed it sets the current package's
3044 \&\f(CW$REGERROR\fR and \f(CW$REGMARK\fR variables. When doing so the following
3045 rules apply:
3046 .PP
3047 On failure, the \f(CW$REGERROR\fR variable will be set to the \fIarg\fR value of the
3048 verb pattern, if the verb was involved in the failure of the match. If the
3049 \&\fIarg\fR part of the pattern was omitted, then \f(CW$REGERROR\fR will be set to the
3050 name of the last \f(CW\*(C`(*MARK:\fR\f(CINAME\fR\f(CW)\*(C'\fR pattern executed, or to TRUE if there was
3051 none. Also, the \f(CW$REGMARK\fR variable will be set to FALSE.
3052 .PP
3053 On a successful match, the \f(CW$REGERROR\fR variable will be set to FALSE, and
3054 the \f(CW$REGMARK\fR variable will be set to the name of the last
3055 \&\f(CW\*(C`(*MARK:\fR\f(CINAME\fR\f(CW)\*(C'\fR pattern executed. See the explanation for the
3056 \&\f(CW\*(C`(*MARK:\fR\f(CINAME\fR\f(CW)\*(C'\fR verb below for more details.
3057 .PP
3058 \&\fBNOTE:\fR \f(CW$REGERROR\fR and \f(CW$REGMARK\fR are not magic variables like \f(CW$1\fR
3059 and most other regex-related variables. They are not local to a scope, nor
3060 readonly, but instead are volatile package variables similar to \f(CW$AUTOLOAD\fR.
3061 They are set in the package containing the code that \fIexecuted\fR the regex
3062 (rather than the one that compiled it, where those differ). If necessary, you
3063 can use \f(CW\*(C`local\*(C'\fR to localize changes to these variables to a specific scope
3064 before executing a regex.
3065 .PP
3066 If a pattern does not contain a special backtracking verb that allows an
3067 argument, then \f(CW$REGERROR\fR and \f(CW$REGMARK\fR are not touched at all.
3068 .IP Verbs 3
3069 .IX Item "Verbs"
3070 .RS 3
3071 .PD 0
3072 .ie n .IP """(*PRUNE)"" ""(*PRUNE:\fINAME\fR)""" 4
3073 .el .IP "\f(CW(*PRUNE)\fR \f(CW(*PRUNE:\fR\f(CINAME\fR\f(CW)\fR" 4
3074 .IX Xref "(*PRUNE) (*PRUNE:NAME)"
3075 .IX Item "(*PRUNE) (*PRUNE:NAME)"
3076 .PD
3077 This zero-width pattern prunes the backtracking tree at the current point
3078 when backtracked into on failure. Consider the pattern \f(CW\*(C`/\fR\f(CIA\fR\f(CW (*PRUNE) \fR\f(CIB\fR\f(CW/\*(C'\fR,
3079 where \fIA\fR and \fIB\fR are complex patterns. Until the \f(CW\*(C`(*PRUNE)\*(C'\fR verb is reached,
3080 \&\fIA\fR may backtrack as necessary to match. Once it is reached, matching
3081 continues in \fIB\fR, which may also backtrack as necessary; however, should B
3082 not match, then no further backtracking will take place, and the pattern
3083 will fail outright at the current starting position.
3084 .Sp
3085 The following example counts all the possible matching strings in a
3086 pattern (without actually matching any of them).
3087 .Sp
3088 .Vb 2
3089 \& \*(Aqaaab\*(Aq =~ /a+b?(?{print "$&\en"; $count++})(*FAIL)/;
3090 \& print "Count=$count\en";
3091 .Ve
3092 .Sp
3093 which produces:
3094 .Sp
3095 .Vb 10
3096 \& aaab
3097 \& aaa
3098 \& aa
3099 \& a
3100 \& aab
3101 \& aa
3102 \& a
3103 \& ab
3104 \& a
3105 \& Count=9
3106 .Ve
3107 .Sp
3108 If we add a \f(CW\*(C`(*PRUNE)\*(C'\fR before the count like the following
3109 .Sp
3110 .Vb 2
3111 \& \*(Aqaaab\*(Aq =~ /a+b?(*PRUNE)(?{print "$&\en"; $count++})(*FAIL)/;
3112 \& print "Count=$count\en";
3113 .Ve
3114 .Sp
3115 we prevent backtracking and find the count of the longest matching string
3116 at each matching starting point like so:
3117 .Sp
3118 .Vb 4
3119 \& aaab
3120 \& aab
3121 \& ab
3122 \& Count=3
3123 .Ve
3124 .Sp
3125 Any number of \f(CW\*(C`(*PRUNE)\*(C'\fR assertions may be used in a pattern.
3126 .Sp
3127 See also \f(CW"(?>\fR\f(CIpattern\fR\f(CW)"\fR and possessive quantifiers for
3128 other ways to
3129 control backtracking. In some cases, the use of \f(CW\*(C`(*PRUNE)\*(C'\fR can be
3130 replaced with a \f(CW\*(C`(?>pattern)\*(C'\fR with no functional difference; however,
3131 \&\f(CW\*(C`(*PRUNE)\*(C'\fR can be used to handle cases that cannot be expressed using a
3132 \&\f(CW\*(C`(?>pattern)\*(C'\fR alone.
3133 .ie n .IP """(*SKIP)"" ""(*SKIP:\fINAME\fR)""" 4
3134 .el .IP "\f(CW(*SKIP)\fR \f(CW(*SKIP:\fR\f(CINAME\fR\f(CW)\fR" 4
3135 .IX Xref "(*SKIP)"
3136 .IX Item "(*SKIP) (*SKIP:NAME)"
3137 This zero-width pattern is similar to \f(CW\*(C`(*PRUNE)\*(C'\fR, except that on
3138 failure it also signifies that whatever text that was matched leading up
3139 to the \f(CW\*(C`(*SKIP)\*(C'\fR pattern being executed cannot be part of \fIany\fR match
3140 of this pattern. This effectively means that the regex engine "skips" forward
3141 to this position on failure and tries to match again, (assuming that
3142 there is sufficient room to match).
3143 .Sp
3144 The name of the \f(CW\*(C`(*SKIP:\fR\f(CINAME\fR\f(CW)\*(C'\fR pattern has special significance. If a
3145 \&\f(CW\*(C`(*MARK:\fR\f(CINAME\fR\f(CW)\*(C'\fR was encountered while matching, then it is that position
3146 which is used as the "skip point". If no \f(CW\*(C`(*MARK)\*(C'\fR of that name was
3147 encountered, then the \f(CW\*(C`(*SKIP)\*(C'\fR operator has no effect. When used
3148 without a name the "skip point" is where the match point was when
3149 executing the \f(CW\*(C`(*SKIP)\*(C'\fR pattern.
3150 .Sp
3151 Compare the following to the examples in \f(CW\*(C`(*PRUNE)\*(C'\fR; note the string
3152 is twice as long:
3153 .Sp
3154 .Vb 2
3155 \& \*(Aqaaabaaab\*(Aq =~ /a+b?(*SKIP)(?{print "$&\en"; $count++})(*FAIL)/;
3156 \& print "Count=$count\en";
3157 .Ve
3158 .Sp
3159 outputs
3160 .Sp
3161 .Vb 3
3162 \& aaab
3163 \& aaab
3164 \& Count=2
3165 .Ve
3166 .Sp
3167 Once the 'aaab' at the start of the string has matched, and the \f(CW\*(C`(*SKIP)\*(C'\fR
3168 executed, the next starting point will be where the cursor was when the
3169 \&\f(CW\*(C`(*SKIP)\*(C'\fR was executed.
3170 .ie n .IP """(*MARK:\fINAME\fR)"" ""(*:\fINAME\fR)""" 4
3171 .el .IP "\f(CW(*MARK:\fR\f(CINAME\fR\f(CW)\fR \f(CW(*:\fR\f(CINAME\fR\f(CW)\fR" 4
3172 .IX Xref "(*MARK) (*MARK:NAME) (*:NAME)"
3173 .IX Item "(*MARK:NAME) (*:NAME)"
3174 This zero-width pattern can be used to mark the point reached in a string
3175 when a certain part of the pattern has been successfully matched. This
3176 mark may be given a name. A later \f(CW\*(C`(*SKIP)\*(C'\fR pattern will then skip
3177 forward to that point if backtracked into on failure. Any number of
3178 \&\f(CW\*(C`(*MARK)\*(C'\fR patterns are allowed, and the \fINAME\fR portion may be duplicated.
3179 .Sp
3180 In addition to interacting with the \f(CW\*(C`(*SKIP)\*(C'\fR pattern, \f(CW\*(C`(*MARK:\fR\f(CINAME\fR\f(CW)\*(C'\fR
3181 can be used to "label" a pattern branch, so that after matching, the
3182 program can determine which branches of the pattern were involved in the
3183 match.
3184 .Sp
3185 When a match is successful, the \f(CW$REGMARK\fR variable will be set to the
3186 name of the most recently executed \f(CW\*(C`(*MARK:\fR\f(CINAME\fR\f(CW)\*(C'\fR that was involved
3187 in the match.
3188 .Sp
3189 This can be used to determine which branch of a pattern was matched
3190 without using a separate capture group for each branch, which in turn
3191 can result in a performance improvement, as perl cannot optimize
3192 \&\f(CW\*(C`/(?:(x)|(y)|(z))/\*(C'\fR as efficiently as something like
3193 \&\f(CW\*(C`/(?:x(*MARK:x)|y(*MARK:y)|z(*MARK:z))/\*(C'\fR.
3194 .Sp
3195 When a match has failed, and unless another verb has been involved in
3196 failing the match and has provided its own name to use, the \f(CW$REGERROR\fR
3197 variable will be set to the name of the most recently executed
3198 \&\f(CW\*(C`(*MARK:\fR\f(CINAME\fR\f(CW)\*(C'\fR.
3199 .Sp
3200 See "(*SKIP)" for more details.
3201 .Sp
3202 As a shortcut \f(CW\*(C`(*MARK:\fR\f(CINAME\fR\f(CW)\*(C'\fR can be written \f(CW\*(C`(*:\fR\f(CINAME\fR\f(CW)\*(C'\fR.
3203 .ie n .IP """(*THEN)"" ""(*THEN:\fINAME\fR)""" 4
3204 .el .IP "\f(CW(*THEN)\fR \f(CW(*THEN:\fR\f(CINAME\fR\f(CW)\fR" 4
3205 .IX Item "(*THEN) (*THEN:NAME)"
3206 This is similar to the "cut group" operator \f(CW\*(C`::\*(C'\fR from Raku. Like
3207 \&\f(CW\*(C`(*PRUNE)\*(C'\fR, this verb always matches, and when backtracked into on
3208 failure, it causes the regex engine to try the next alternation in the
3209 innermost enclosing group (capturing or otherwise) that has alternations.
3210 The two branches of a \f(CW\*(C`(?(\fR\f(CIcondition\fR\f(CW)\fR\f(CIyes\-pattern\fR\f(CW|\fR\f(CIno\-pattern\fR\f(CW)\*(C'\fR do not
3211 count as an alternation, as far as \f(CW\*(C`(*THEN)\*(C'\fR is concerned.
3212 .Sp
3213 Its name comes from the observation that this operation combined with the
3214 alternation operator (\f(CW"|"\fR) can be used to create what is essentially a
3215 pattern-based if/then/else block:
3216 .Sp
3217 .Vb 1
3218 \& ( COND (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ )
3219 .Ve
3220 .Sp
3221 Note that if this operator is used and NOT inside of an alternation then
3222 it acts exactly like the \f(CW\*(C`(*PRUNE)\*(C'\fR operator.
3223 .Sp
3224 .Vb 1
3225 \& / A (*PRUNE) B /
3226 .Ve
3227 .Sp
3228 is the same as
3229 .Sp
3230 .Vb 1
3231 \& / A (*THEN) B /
3232 .Ve
3233 .Sp
3234 but
3235 .Sp
3236 .Vb 1
3237 \& / ( A (*THEN) B | C ) /
3238 .Ve
3239 .Sp
3240 is not the same as
3241 .Sp
3242 .Vb 1
3243 \& / ( A (*PRUNE) B | C ) /
3244 .Ve
3245 .Sp
3246 as after matching the \fIA\fR but failing on the \fIB\fR the \f(CW\*(C`(*THEN)\*(C'\fR verb will
3247 backtrack and try \fIC\fR; but the \f(CW\*(C`(*PRUNE)\*(C'\fR verb will simply fail.
3248 .ie n .IP """(*COMMIT)"" ""(*COMMIT:\fIarg\fR)""" 4
3249 .el .IP "\f(CW(*COMMIT)\fR \f(CW(*COMMIT:\fR\f(CIarg\fR\f(CW)\fR" 4
3250 .IX Xref "(*COMMIT)"
3251 .IX Item "(*COMMIT) (*COMMIT:arg)"
3252 This is the Raku "commit pattern" \f(CW\*(C`<commit>\*(C'\fR or \f(CW\*(C`:::\*(C'\fR. It's a
3253 zero-width pattern similar to \f(CW\*(C`(*SKIP)\*(C'\fR, except that when backtracked
3254 into on failure it causes the match to fail outright. No further attempts
3255 to find a valid match by advancing the start pointer will occur again.
3256 For example,
3257 .Sp
3258 .Vb 2
3259 \& \*(Aqaaabaaab\*(Aq =~ /a+b?(*COMMIT)(?{print "$&\en"; $count++})(*FAIL)/;
3260 \& print "Count=$count\en";
3261 .Ve
3262 .Sp
3263 outputs
3264 .Sp
3265 .Vb 2
3266 \& aaab
3267 \& Count=1
3268 .Ve
3269 .Sp
3270 In other words, once the \f(CW\*(C`(*COMMIT)\*(C'\fR has been entered, and if the pattern
3271 does not match, the regex engine will not try any further matching on the
3272 rest of the string.
3273 .ie n .IP """(*FAIL)"" ""(*F)"" ""(*FAIL:\fIarg\fR)""" 4
3274 .el .IP "\f(CW(*FAIL)\fR \f(CW(*F)\fR \f(CW(*FAIL:\fR\f(CIarg\fR\f(CW)\fR" 4
3275 .IX Xref "(*FAIL) (*F)"
3276 .IX Item "(*FAIL) (*F) (*FAIL:arg)"
3277 This pattern matches nothing and always fails. It can be used to force the
3278 engine to backtrack. It is equivalent to \f(CW\*(C`(?!)\*(C'\fR, but easier to read. In
3279 fact, \f(CW\*(C`(?!)\*(C'\fR gets optimised into \f(CW\*(C`(*FAIL)\*(C'\fR internally. You can provide
3280 an argument so that if the match fails because of this \f(CW\*(C`FAIL\*(C'\fR directive
3281 the argument can be obtained from \f(CW$REGERROR\fR.
3282 .Sp
3283 It is probably useful only when combined with \f(CW\*(C`(?{})\*(C'\fR or \f(CW\*(C`(??{})\*(C'\fR.
3284 .ie n .IP """(*ACCEPT)"" ""(*ACCEPT:\fIarg\fR)""" 4
3285 .el .IP "\f(CW(*ACCEPT)\fR \f(CW(*ACCEPT:\fR\f(CIarg\fR\f(CW)\fR" 4
3286 .IX Xref "(*ACCEPT)"
3287 .IX Item "(*ACCEPT) (*ACCEPT:arg)"
3288 This pattern matches nothing and causes the end of successful matching at
3289 the point at which the \f(CW\*(C`(*ACCEPT)\*(C'\fR pattern was encountered, regardless of
3290 whether there is actually more to match in the string. When inside of a
3291 nested pattern, such as recursion, or in a subpattern dynamically generated
3292 via \f(CW\*(C`(??{})\*(C'\fR, only the innermost pattern is ended immediately.
3293 .Sp
3294 If the \f(CW\*(C`(*ACCEPT)\*(C'\fR is inside of capturing groups then the groups are
3295 marked as ended at the point at which the \f(CW\*(C`(*ACCEPT)\*(C'\fR was encountered.
3296 For instance:
3297 .Sp
3298 .Vb 1
3299 \& \*(AqAB\*(Aq =~ /(A (A|B(*ACCEPT)|C) D)(E)/x;
3300 .Ve
3301 .Sp
3302 will match, and \f(CW$1\fR will be \f(CW\*(C`AB\*(C'\fR and \f(CW$2\fR will be \f(CW"B"\fR, \f(CW$3\fR will not
3303 be set. If another branch in the inner parentheses was matched, such as in the
3304 string 'ACDE', then the \f(CW"D"\fR and \f(CW"E"\fR would have to be matched as well.
3305 .Sp
3306 You can provide an argument, which will be available in the var
3307 \&\f(CW$REGMARK\fR after the match completes.
3308 .RE
3309 .RS 3
3310 .RE
3311 .ie n .SS "Warning on ""\e1"" Instead of $1"
3312 .el .SS "Warning on \f(CW\e1\fP Instead of \f(CW$1\fP"
3313 .IX Subsection "Warning on 1 Instead of $1"
3314 Some people get too used to writing things like:
3315 .PP
3316 .Vb 1
3317 \& $pattern =~ s/(\eW)/\e\e\e1/g;
3318 .Ve
3319 .PP
3320 This is grandfathered (for \e1 to \e9) for the RHS of a substitute to avoid
3321 shocking the
3322 \&\fBsed\fR addicts, but it's a dirty habit to get into. That's because in
3323 PerlThink, the righthand side of an \f(CW\*(C`s///\*(C'\fR is a double-quoted string. \f(CW\*(C`\e1\*(C'\fR in
3324 the usual double-quoted string means a control-A. The customary Unix
3325 meaning of \f(CW\*(C`\e1\*(C'\fR is kludged in for \f(CW\*(C`s///\*(C'\fR. However, if you get into the habit
3326 of doing that, you get yourself into trouble if you then add an \f(CW\*(C`/e\*(C'\fR
3327 modifier.
3328 .PP
3329 .Vb 1
3330 \& s/(\ed+)/ \e1 + 1 /eg; # causes warning under \-w
3331 .Ve
3332 .PP
3333 Or if you try to do
3334 .PP
3335 .Vb 1
3336 \& s/(\ed+)/\e1000/;
3337 .Ve
3338 .PP
3339 You can't disambiguate that by saying \f(CW\*(C`\e{1}000\*(C'\fR, whereas you can fix it with
3340 \&\f(CW\*(C`${1}000\*(C'\fR. The operation of interpolation should not be confused
3341 with the operation of matching a backreference. Certainly they mean two
3342 different things on the \fIleft\fR side of the \f(CW\*(C`s///\*(C'\fR.
3343 .SS "Repeated Patterns Matching a Zero-length Substring"
3344 .IX Subsection "Repeated Patterns Matching a Zero-length Substring"
3345 \&\fBWARNING\fR: Difficult material (and prose) ahead. This section needs a rewrite.
3346 .PP
3347 Regular expressions provide a terse and powerful programming language. As
3348 with most other power tools, power comes together with the ability
3349 to wreak havoc.
3350 .PP
3351 A common abuse of this power stems from the ability to make infinite
3352 loops using regular expressions, with something as innocuous as:
3353 .PP
3354 .Vb 1
3355 \& \*(Aqfoo\*(Aq =~ m{ ( o? )* }x;
3356 .Ve
3357 .PP
3358 The \f(CW\*(C`o?\*(C'\fR matches at the beginning of "\f(CW\*(C`foo\*(C'\fR", and since the position
3359 in the string is not moved by the match, \f(CW\*(C`o?\*(C'\fR would match again and again
3360 because of the \f(CW"*"\fR quantifier. Another common way to create a similar cycle
3361 is with the looping modifier \f(CW\*(C`/g\*(C'\fR:
3362 .PP
3363 .Vb 1
3364 \& @matches = ( \*(Aqfoo\*(Aq =~ m{ o? }xg );
3365 .Ve
3366 .PP
3367 or
3368 .PP
3369 .Vb 1
3370 \& print "match: <$&>\en" while \*(Aqfoo\*(Aq =~ m{ o? }xg;
3371 .Ve
3372 .PP
3373 or the loop implied by \f(CWsplit()\fR.
3374 .PP
3375 However, long experience has shown that many programming tasks may
3376 be significantly simplified by using repeated subexpressions that
3377 may match zero-length substrings. Here's a simple example being:
3378 .PP
3379 .Vb 2
3380 \& @chars = split //, $string; # // is not magic in split
3381 \& ($whitewashed = $string) =~ s/()/ /g; # parens avoid magic s// /
3382 .Ve
3383 .PP
3384 Thus Perl allows such constructs, by \fIforcefully breaking
3385 the infinite loop\fR. The rules for this are different for lower-level
3386 loops given by the greedy quantifiers \f(CW\*(C`*+{}\*(C'\fR, and for higher-level
3387 ones like the \f(CW\*(C`/g\*(C'\fR modifier or \f(CWsplit()\fR operator.
3388 .PP
3389 The lower-level loops are \fIinterrupted\fR (that is, the loop is
3390 broken) when Perl detects that a repeated expression matched a
3391 zero-length substring. Thus
3392 .PP
3393 .Vb 1
3394 \& m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;
3395 .Ve
3396 .PP
3397 is made equivalent to
3398 .PP
3399 .Vb 1
3400 \& m{ (?: NON_ZERO_LENGTH )* (?: ZERO_LENGTH )? }x;
3401 .Ve
3402 .PP
3403 For example, this program
3404 .PP
3405 .Vb 12
3406 \& #!perl \-l
3407 \& "aaaaab" =~ /
3408 \& (?:
3409 \& a # non\-zero
3410 \& | # or
3411 \& (?{print "hello"}) # print hello whenever this
3412 \& # branch is tried
3413 \& (?=(b)) # zero\-width assertion
3414 \& )* # any number of times
3415 \& /x;
3416 \& print $&;
3417 \& print $1;
3418 .Ve
3419 .PP
3420 prints
3421 .PP
3422 .Vb 3
3423 \& hello
3424 \& aaaaa
3425 \& b
3426 .Ve
3427 .PP
3428 Notice that "hello" is only printed once, as when Perl sees that the sixth
3429 iteration of the outermost \f(CW\*(C`(?:)*\*(C'\fR matches a zero-length string, it stops
3430 the \f(CW"*"\fR.
3431 .PP
3432 The higher-level loops preserve an additional state between iterations:
3433 whether the last match was zero-length. To break the loop, the following
3434 match after a zero-length match is prohibited to have a length of zero.
3435 This prohibition interacts with backtracking (see "Backtracking"),
3436 and so the \fIsecond best\fR match is chosen if the \fIbest\fR match is of
3437 zero length.
3438 .PP
3439 For example:
3440 .PP
3441 .Vb 2
3442 \& $_ = \*(Aqbar\*(Aq;
3443 \& s/\ew??/<$&>/g;
3444 .Ve
3445 .PP
3446 results in \f(CW\*(C`<><b><><a><><r><>\*(C'\fR. At each position of the string the best
3447 match given by non-greedy \f(CW\*(C`??\*(C'\fR is the zero-length match, and the \fIsecond
3448 best\fR match is what is matched by \f(CW\*(C`\ew\*(C'\fR. Thus zero-length matches
3449 alternate with one-character-long matches.
3450 .PP
3451 Similarly, for repeated \f(CW\*(C`m/()/g\*(C'\fR the second-best match is the match at the
3452 position one notch further in the string.
3453 .PP
3454 The additional state of being \fImatched with zero-length\fR is associated with
3455 the matched string, and is reset by each assignment to \f(CWpos()\fR.
3456 Zero-length matches at the end of the previous match are ignored
3457 during \f(CW\*(C`split\*(C'\fR.
3458 .SS "Combining RE Pieces"
3459 .IX Subsection "Combining RE Pieces"
3460 Each of the elementary pieces of regular expressions which were described
3461 before (such as \f(CW\*(C`ab\*(C'\fR or \f(CW\*(C`\eZ\*(C'\fR) could match at most one substring
3462 at the given position of the input string. However, in a typical regular
3463 expression these elementary pieces are combined into more complicated
3464 patterns using combining operators \f(CW\*(C`ST\*(C'\fR, \f(CW\*(C`S|T\*(C'\fR, \f(CW\*(C`S*\*(C'\fR \fIetc\fR.
3465 (in these examples \f(CW"S"\fR and \f(CW"T"\fR are regular subexpressions).
3466 .PP
3467 Such combinations can include alternatives, leading to a problem of choice:
3468 if we match a regular expression \f(CW\*(C`a|ab\*(C'\fR against \f(CW"abc"\fR, will it match
3469 substring \f(CW"a"\fR or \f(CW"ab"\fR? One way to describe which substring is
3470 actually matched is the concept of backtracking (see "Backtracking").
3471 However, this description is too low-level and makes you think
3472 in terms of a particular implementation.
3473 .PP
3474 Another description starts with notions of "better"/"worse". All the
3475 substrings which may be matched by the given regular expression can be
3476 sorted from the "best" match to the "worst" match, and it is the "best"
3477 match which is chosen. This substitutes the question of "what is chosen?"
3478 by the question of "which matches are better, and which are worse?".
3479 .PP
3480 Again, for elementary pieces there is no such question, since at most
3481 one match at a given position is possible. This section describes the
3482 notion of better/worse for combining operators. In the description
3483 below \f(CW"S"\fR and \f(CW"T"\fR are regular subexpressions.
3484 .ie n .IP """ST""" 4
3485 .el .IP \f(CWST\fR 4
3486 .IX Item "ST"
3487 Consider two possible matches, \f(CW\*(C`AB\*(C'\fR and \f(CW\*(C`A\*(AqB\*(Aq\*(C'\fR, \f(CW"A"\fR and \f(CW\*(C`A\*(Aq\*(C'\fR are
3488 substrings which can be matched by \f(CW"S"\fR, \f(CW"B"\fR and \f(CW\*(C`B\*(Aq\*(C'\fR are substrings
3489 which can be matched by \f(CW"T"\fR.
3490 .Sp
3491 If \f(CW"A"\fR is a better match for \f(CW"S"\fR than \f(CW\*(C`A\*(Aq\*(C'\fR, \f(CW\*(C`AB\*(C'\fR is a better
3492 match than \f(CW\*(C`A\*(AqB\*(Aq\*(C'\fR.
3493 .Sp
3494 If \f(CW"A"\fR and \f(CW\*(C`A\*(Aq\*(C'\fR coincide: \f(CW\*(C`AB\*(C'\fR is a better match than \f(CW\*(C`AB\*(Aq\*(C'\fR if
3495 \&\f(CW"B"\fR is a better match for \f(CW"T"\fR than \f(CW\*(C`B\*(Aq\*(C'\fR.
3496 .ie n .IP """S|T""" 4
3497 .el .IP \f(CWS|T\fR 4
3498 .IX Item "S|T"
3499 When \f(CW"S"\fR can match, it is a better match than when only \f(CW"T"\fR can match.
3500 .Sp
3501 Ordering of two matches for \f(CW"S"\fR is the same as for \f(CW"S"\fR. Similar for
3502 two matches for \f(CW"T"\fR.
3503 .ie n .IP """S{REPEAT_COUNT}""" 4
3504 .el .IP \f(CWS{REPEAT_COUNT}\fR 4
3505 .IX Item "S{REPEAT_COUNT}"
3506 Matches as \f(CW\*(C`SSS...S\*(C'\fR (repeated as many times as necessary).
3507 .ie n .IP """S{min,max}""" 4
3508 .el .IP \f(CWS{min,max}\fR 4
3509 .IX Item "S{min,max}"
3510 Matches as \f(CW\*(C`S{max}|S{max\-1}|...|S{min+1}|S{min}\*(C'\fR.
3511 .ie n .IP """S{min,max}?""" 4
3512 .el .IP \f(CWS{min,max}?\fR 4
3513 .IX Item "S{min,max}?"
3514 Matches as \f(CW\*(C`S{min}|S{min+1}|...|S{max\-1}|S{max}\*(C'\fR.
3515 .ie n .IP """S?"", ""S*"", ""S+""" 4
3516 .el .IP "\f(CWS?\fR, \f(CWS*\fR, \f(CWS+\fR" 4
3517 .IX Item "S?, S*, S+"
3518 Same as \f(CW\*(C`S{0,1}\*(C'\fR, \f(CW\*(C`S{0,BIG_NUMBER}\*(C'\fR, \f(CW\*(C`S{1,BIG_NUMBER}\*(C'\fR respectively.
3519 .ie n .IP """S??"", ""S*?"", ""S+?""" 4
3520 .el .IP "\f(CWS??\fR, \f(CWS*?\fR, \f(CWS+?\fR" 4
3521 .IX Item "S??, S*?, S+?"
3522 Same as \f(CW\*(C`S{0,1}?\*(C'\fR, \f(CW\*(C`S{0,BIG_NUMBER}?\*(C'\fR, \f(CW\*(C`S{1,BIG_NUMBER}?\*(C'\fR respectively.
3523 .ie n .IP """(?>S)""" 4
3524 .el .IP \f(CW(?>S)\fR 4
3525 .IX Item "(?>S)"
3526 Matches the best match for \f(CW"S"\fR and only that.
3527 .ie n .IP """(?=S)"", ""(?<=S)""" 4
3528 .el .IP "\f(CW(?=S)\fR, \f(CW(?<=S)\fR" 4
3529 .IX Item "(?=S), (?<=S)"
3530 Only the best match for \f(CW"S"\fR is considered. (This is important only if
3531 \&\f(CW"S"\fR has capturing parentheses, and backreferences are used somewhere
3532 else in the whole regular expression.)
3533 .ie n .IP """(?!S)"", ""(?<!S)""" 4
3534 .el .IP "\f(CW(?!S)\fR, \f(CW(?<!S)\fR" 4
3535 .IX Item "(?!S), (?<!S)"
3536 For this grouping operator there is no need to describe the ordering, since
3537 only whether or not \f(CW"S"\fR can match is important.
3538 .ie n .IP """(??{ \fIEXPR\fR })"", ""(?\fIPARNO\fR)""" 4
3539 .el .IP "\f(CW(??{ \fR\f(CIEXPR\fR\f(CW })\fR, \f(CW(?\fR\f(CIPARNO\fR\f(CW)\fR" 4
3540 .IX Item "(??{ EXPR }), (?PARNO)"
3541 The ordering is the same as for the regular expression which is
3542 the result of \fIEXPR\fR, or the pattern contained by capture group \fIPARNO\fR.
3543 .ie n .IP """(?(\fIcondition\fR)\fIyes\-pattern\fR|\fIno\-pattern\fR)""" 4
3544 .el .IP \f(CW(?(\fR\f(CIcondition\fR\f(CW)\fR\f(CIyes\-pattern\fR\f(CW|\fR\f(CIno\-pattern\fR\f(CW)\fR 4
3545 .IX Item "(?(condition)yes-pattern|no-pattern)"
3546 Recall that which of \fIyes-pattern\fR or \fIno-pattern\fR actually matches is
3547 already determined. The ordering of the matches is the same as for the
3548 chosen subexpression.
3549 .PP
3550 The above recipes describe the ordering of matches \fIat a given position\fR.
3551 One more rule is needed to understand how a match is determined for the
3552 whole regular expression: a match at an earlier position is always better
3553 than a match at a later position.
3554 .SS "Creating Custom RE Engines"
3555 .IX Subsection "Creating Custom RE Engines"
3556 As of Perl 5.10.0, one can create custom regular expression engines. This
3557 is not for the faint of heart, as they have to plug in at the C level. See
3558 perlreapi for more details.
3559 .PP
3560 As an alternative, overloaded constants (see overload) provide a simple
3561 way to extend the functionality of the RE engine, by substituting one
3562 pattern for another.
3563 .PP
3564 Suppose that we want to enable a new RE escape-sequence \f(CW\*(C`\eY|\*(C'\fR which
3565 matches at a boundary between whitespace characters and non-whitespace
3566 characters. Note that \f(CW\*(C`(?=\eS)(?<!\eS)|(?!\eS)(?<=\eS)\*(C'\fR matches exactly
3567 at these positions, so we want to have each \f(CW\*(C`\eY|\*(C'\fR in the place of the
3568 more complicated version. We can create a module \f(CW\*(C`customre\*(C'\fR to do
3569 this:
3570 .PP
3571 .Vb 2
3572 \& package customre;
3573 \& use overload;
3574 \&
3575 \& sub import {
3576 \& shift;
3577 \& die "No argument to customre::import allowed" if @_;
3578 \& overload::constant \*(Aqqr\*(Aq => \e&convert;
3579 \& }
3580 \&
3581 \& sub invalid { die "/$_[0]/: invalid escape \*(Aq\e\e$_[1]\*(Aq"}
3582 \&
3583 \& # We must also take care of not escaping the legitimate \e\eY|
3584 \& # sequence, hence the presence of \*(Aq\e\e\*(Aq in the conversion rules.
3585 \& my %rules = ( \*(Aq\e\e\*(Aq => \*(Aq\e\e\e\e\*(Aq,
3586 \& \*(AqY|\*(Aq => qr/(?=\eS)(?<!\eS)|(?!\eS)(?<=\eS)/ );
3587 \& sub convert {
3588 \& my $re = shift;
3589 \& $re =~ s{
3590 \& \e\e ( \e\e | Y . )
3591 \& }
3592 \& { $rules{$1} or invalid($re,$1) }sgex;
3593 \& return $re;
3594 \& }
3595 .Ve
3596 .PP
3597 Now \f(CW\*(C`use customre\*(C'\fR enables the new escape in constant regular
3598 expressions, \fIi.e.\fR, those without any runtime variable interpolations.
3599 As documented in overload, this conversion will work only over
3600 literal parts of regular expressions. For \f(CW\*(C`\eY|$re\eY|\*(C'\fR the variable
3601 part of this regular expression needs to be converted explicitly
3602 (but only if the special meaning of \f(CW\*(C`\eY|\*(C'\fR should be enabled inside \f(CW$re\fR):
3603 .PP
3604 .Vb 5
3605 \& use customre;
3606 \& $re = <>;
3607 \& chomp $re;
3608 \& $re = customre::convert $re;
3609 \& /\eY|$re\eY|/;
3610 .Ve
3611 .SS "Embedded Code Execution Frequency"
3612 .IX Subsection "Embedded Code Execution Frequency"
3613 The exact rules for how often \f(CW\*(C`(?{})\*(C'\fR and \f(CW\*(C`(??{})\*(C'\fR are executed in a pattern
3614 are unspecified, and this is even more true of \f(CW\*(C`(*{})\*(C'\fR.
3615 In the case of a successful match you can assume that they DWIM and
3616 will be executed in left to right order the appropriate number of times in the
3617 accepting path of the pattern as would any other meta-pattern. How non\-
3618 accepting pathways and match failures affect the number of times a pattern is
3619 executed is specifically unspecified and may vary depending on what
3620 optimizations can be applied to the pattern and is likely to change from
3621 version to version.
3622 .PP
3623 For instance in
3624 .PP
3625 .Vb 1
3626 \& "aaabcdeeeee"=~/a(?{print "a"})b(?{print "b"})cde/;
3627 .Ve
3628 .PP
3629 the exact number of times "a" or "b" are printed out is unspecified for
3630 failure, but you may assume they will be printed at least once during
3631 a successful match, additionally you may assume that if "b" is printed,
3632 it will be preceded by at least one "a".
3633 .PP
3634 In the case of branching constructs like the following:
3635 .PP
3636 .Vb 1
3637 \& /a(b|(?{ print "a" }))c(?{ print "c" })/;
3638 .Ve
3639 .PP
3640 you can assume that the input "ac" will output "ac", and that "abc"
3641 will output only "c".
3642 .PP
3643 When embedded code is quantified, successful matches will call the
3644 code once for each matched iteration of the quantifier. For
3645 example:
3646 .PP
3647 .Vb 1
3648 \& "good" =~ /g(?:o(?{print "o"}))*d/;
3649 .Ve
3650 .PP
3651 will output "o" twice.
3652 .PP
3653 For historical and consistency reasons the use of normal code blocks
3654 anywhere in a pattern will disable certain optimisations. As of 5.37.7
3655 you can use an "optimistic" codeblock, \f(CW\*(C`(*{ ... })\*(C'\fR as a replacement
3656 for \f(CW\*(C`(?{ ... })\*(C'\fR, if you do *not* wish to disable these optimisations.
3657 This may result in the code block being called less often than it might
3658 have been had they not been optimistic.
3659 .SS "PCRE/Python Support"
3660 .IX Subsection "PCRE/Python Support"
3661 As of Perl 5.10.0, Perl supports several Python/PCRE\-specific extensions
3662 to the regex syntax. While Perl programmers are encouraged to use the
3663 Perl-specific syntax, the following are also accepted:
3664 .ie n .IP """(?P<\fINAME\fR>\fIpattern\fR)""" 4
3665 .el .IP \f(CW(?P<\fR\f(CINAME\fR\f(CW>\fR\f(CIpattern\fR\f(CW)\fR 4
3666 .IX Item "(?P<NAME>pattern)"
3667 Define a named capture group. Equivalent to \f(CW\*(C`(?<\fR\f(CINAME\fR\f(CW>\fR\f(CIpattern\fR\f(CW)\*(C'\fR.
3668 .ie n .IP """(?P=\fINAME\fR)""" 4
3669 .el .IP \f(CW(?P=\fR\f(CINAME\fR\f(CW)\fR 4
3670 .IX Item "(?P=NAME)"
3671 Backreference to a named capture group. Equivalent to \f(CW\*(C`\eg{\fR\f(CINAME\fR\f(CW}\*(C'\fR.
3672 .ie n .IP """(?P>\fINAME\fR)""" 4
3673 .el .IP \f(CW(?P>\fR\f(CINAME\fR\f(CW)\fR 4
3674 .IX Item "(?P>NAME)"
3675 Subroutine call to a named capture group. Equivalent to \f(CW\*(C`(?&\fR\f(CINAME\fR\f(CW)\*(C'\fR.
3676 .SH BUGS
3677 .IX Header "BUGS"
3678 There are a number of issues with regard to case-insensitive matching
3679 in Unicode rules. See \f(CW"i"\fR under "Modifiers" above.
3680 .PP
3681 This document varies from difficult to understand to completely
3682 and utterly opaque. The wandering prose riddled with jargon is
3683 hard to fathom in several places.
3684 .PP
3685 This document needs a rewrite that separates the tutorial content
3686 from the reference content.
3687 .SH "SEE ALSO"
3688 .IX Header "SEE ALSO"
3689 The syntax of patterns used in Perl pattern matching evolved from those
3690 supplied in the Bell Labs Research Unix 8th Edition (Version 8) regex
3691 routines. (The code is actually derived (distantly) from Henry
3692 Spencer's freely redistributable reimplementation of those V8 routines.)
3693 .PP
3694 perlrequick.
3695 .PP
3696 perlretut.
3697 .PP
3698 "Regexp Quote-Like Operators" in perlop.
3699 .PP
3700 "Gory details of parsing quoted constructs" in perlop.
3701 .PP
3702 perlfaq6.
3703 .PP
3704 "pos" in perlfunc.
3705 .PP
3706 perllocale.
3707 .PP
3708 perlebcdic.
3709 .PP
3710 \&\fIMastering Regular Expressions\fR by Jeffrey Friedl, published
3711 by O'Reilly and Associates.