This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlrebackslash: Fix a couple of nits.
[perl5.git] / pod / perlrebackslash.pod
CommitLineData
8a118206
RGS
1=head1 NAME
2
3perlrebackslash - Perl Regular Expression Backslash Sequences and Escapes
4
5=head1 DESCRIPTION
6
7The top level documentation about Perl regular expressions
8is found in L<perlre>.
9
10This document describes all backslash and escape sequences. After
11explaining the role of the backslash, it lists all the sequences that have
12a special meaning in Perl regular expressions (in alphabetical order),
13then describes each of them.
14
15Most sequences are described in detail in different documents; the primary
16purpose of this document is to have a quick reference guide describing all
17backslash and escape sequences.
18
8a118206
RGS
19=head2 The backslash
20
21In a regular expression, the backslash can perform one of two tasks:
22it either takes away the special meaning of the character following it
23(for instance, C<\|> matches a vertical bar, it's not an alternation),
24or it is the start of a backslash or escape sequence.
25
26The rules determining what it is are quite simple: if the character
df225385 27following the backslash is an ASCII punctuation (non-word) character (that is,
b6538e4f
TC
28anything that is not a letter, digit, or underscore), then the backslash just
29takes away any special meaning of the character following it.
df225385
KW
30
31If the character following the backslash is an ASCII letter or an ASCII digit,
32then the sequence may be special; if so, it's listed below. A few letters have
6b46370c
KW
33not been used yet, so escaping them with a backslash doesn't change them to be
34special. A future version of Perl may assign a special meaning to them, so if
b6538e4f 35you have warnings turned on, Perl issues a warning if you use such a
6b46370c 36sequence. [1].
8a118206 37
e2cb52ee 38It is however guaranteed that backslash or escape sequences never have a
8a118206
RGS
39punctuation character following the backslash, not now, and not in a future
40version of Perl 5. So it is safe to put a backslash in front of a non-word
41character.
42
43Note that the backslash itself is special; if you want to match a backslash,
44you have to escape the backslash with a backslash: C</\\/> matches a single
45backslash.
46
47=over 4
48
49=item [1]
50
b6538e4f 51There is one exception. If you use an alphanumeric character as the
8a118206 52delimiter of your pattern (which you probably shouldn't do for readability
b6538e4f 53reasons), you have to escape the delimiter if you want to match
8a118206
RGS
54it. Perl won't warn then. See also L<perlop/Gory details of parsing
55quoted constructs>.
56
57=back
58
59
60=head2 All the sequences and escapes
61
df225385
KW
62Those not usable within a bracketed character class (like C<[\da-z]>) are marked
63as C<Not in [].>
64
f0a2b745 65 \000 Octal escape sequence. See also \o{}.
df225385 66 \1 Absolute backreference. Not in [].
8a118206 67 \a Alarm or bell.
df225385 68 \A Beginning of string. Not in [].
64935bc6 69 \b{}, \b Boundary. (\b is a backspace in []).
bc1d9728 70 \B{}, \B Not a boundary. Not in [].
f321be7e 71 \cX Control-X.
05a0cace
KW
72 \d Match any digit character.
73 \D Match any character that isn't a digit.
8a118206 74 \e Escape character.
df225385 75 \E Turn off \Q, \L and \U processing. Not in [].
8a118206 76 \f Form feed.
628253b8 77 \F Foldcase till \E. Not in [].
f321be7e
SK
78 \g{}, \g1 Named, absolute or relative backreference.
79 Not in [].
df225385 80 \G Pos assertion. Not in [].
05a0cace
KW
81 \h Match any horizontal whitespace character.
82 \H Match any character that isn't horizontal whitespace.
df225385
KW
83 \k{}, \k<>, \k'' Named backreference. Not in [].
84 \K Keep the stuff left of \K. Not in [].
85 \l Lowercase next character. Not in [].
86 \L Lowercase till \E. Not in [].
8a118206 87 \n (Logical) newline character.
05a0cace 88 \N Match any character but newline. Not in [].
fb121860 89 \N{} Named or numbered (Unicode) character or sequence.
f0a2b745 90 \o{} Octal escape sequence.
05a0cace
KW
91 \p{}, \pP Match any character with the given Unicode property.
92 \P{}, \PP Match any character without the given property.
736fe711
KW
93 \Q Quote (disable) pattern metacharacters till \E. Not
94 in [].
8a118206 95 \r Return character.
df225385 96 \R Generic new line. Not in [].
05a0cace
KW
97 \s Match any whitespace character.
98 \S Match any character that isn't a whitespace.
8a118206 99 \t Tab character.
df225385
KW
100 \u Titlecase next character. Not in [].
101 \U Uppercase till \E. Not in [].
05a0cace
KW
102 \v Match any vertical whitespace character.
103 \V Match any character that isn't vertical whitespace
104 \w Match any word character.
105 \W Match any character that isn't a word character.
8a118206 106 \x{}, \x00 Hexadecimal escape sequence.
df225385
KW
107 \X Unicode "extended grapheme cluster". Not in [].
108 \z End of string. Not in [].
109 \Z End of string. Not in [].
8a118206
RGS
110
111=head2 Character Escapes
112
113=head3 Fixed characters
114
e2cb52ee 115A handful of characters have a dedicated I<character escape>. The following
58151fe4 116table shows them, along with their ASCII code points (in decimal and hex),
4948b50f
KW
117their ASCII name, the control escape on ASCII platforms and a short
118description. (For EBCDIC platforms, see L<perlebcdic/OPERATOR DIFFERENCES>.)
8a118206 119
4948b50f 120 Seq. Code Point ASCII Cntrl Description.
8a118206
RGS
121 Dec Hex
122 \a 7 07 BEL \cG alarm or bell
123 \b 8 08 BS \cH backspace [1]
124 \e 27 1B ESC \c[ escape character
125 \f 12 0C FF \cL form feed
126 \n 10 0A LF \cJ line feed [2]
127 \r 13 0D CR \cM carriage return
128 \t 9 09 TAB \cI tab
129
130=over 4
131
132=item [1]
133
301ba1af 134C<\b> is the backspace character only inside a character class. Outside a
64935bc6
KW
135character class, C<\b> alone is a word-character/non-word-character
136boundary, and C<\b{}> is some other type of boundary.
8a118206
RGS
137
138=item [2]
139
b6538e4f 140C<\n> matches a logical newline. Perl converts between C<\n> and your
f6993e9e 141OS's native newline character when reading from or writing to text files.
8a118206
RGS
142
143=back
144
145=head4 Example
146
147 $str =~ /\t/; # Matches if $str contains a (horizontal) tab.
148
149=head3 Control characters
150
151C<\c> is used to denote a control character; the character following C<\c>
4948b50f
KW
152determines the value of the construct. For example the value of C<\cA> is
153C<chr(1)>, and the value of C<\cb> is C<chr(2)>, etc.
154The gory details are in L<perlop/"Regexp Quote-Like Operators">. A complete
155list of what C<chr(1)>, etc. means for ASCII and EBCDIC platforms is in
156L<perlebcdic/OPERATOR DIFFERENCES>.
157
158Note that C<\c\> alone at the end of a regular expression (or doubled-quoted
159string) is not valid. The backslash must be followed by another character.
160That is, C<\c\I<X>> means C<chr(28) . 'I<X>'> for all characters I<X>.
161
162To write platform-independent code, you must use C<\N{I<NAME>}> instead, like
163C<\N{ESCAPE}> or C<\N{U+001B}>, see L<charnames>.
8a118206
RGS
164
165Mnemonic: I<c>ontrol character.
166
167=head4 Example
168
169 $str =~ /\cK/; # Matches if $str contains a vertical tab (control-K).
170
fb121860 171=head3 Named or numbered characters and character sequences
8a118206 172
17148a1a
KW
173Unicode characters have a Unicode name and numeric code point (ordinal)
174value. Use the
e526e8bb 175C<\N{}> construct to specify a character by either of these values.
fb121860 176Certain sequences of characters also have names.
e526e8bb 177
fb121860 178To specify by name, the name of the character or character sequence goes
fbb93542 179between the curly braces.
e526e8bb 180
b6538e4f
TC
181To specify a character by Unicode code point, use the form C<\N{U+I<code
182point>}>, where I<code point> is a number in hexadecimal that gives the
17148a1a 183code point that Unicode has assigned to the desired character. It is
b6538e4f
TC
184customary but not required to use leading zeros to pad the number to 4
185digits. Thus C<\N{U+0041}> means C<LATIN CAPITAL LETTER A>, and you will
186rarely see it written without the two leading zeros. C<\N{U+0041}> means
187"A" even on EBCDIC machines (where the ordinal value of "A" is not 0x41).
e526e8bb 188
fb121860
KW
189It is even possible to give your own names to characters and character
190sequences. For details, see L<charnames>.
8a118206 191
8c37f1d0 192(There is an expanded internal form that you may see in debug output:
b6538e4f
TC
193C<\N{U+I<code point>.I<code point>...}>.
194The C<...> means any number of these I<code point>s separated by dots.
8c37f1d0
KW
195This represents the sequence formed by the characters. This is an internal
196form only, subject to change, and you should not try to use it yourself.)
197
8a118206
RGS
198Mnemonic: I<N>amed character.
199
b6538e4f
TC
200Note that a character or character sequence expressed as a named
201or numbered character is considered a character without special
fb121860 202meaning by the regex engine, and will match "as is".
df225385 203
8a118206
RGS
204=head4 Example
205
8a118206
RGS
206 $str =~ /\N{THAI CHARACTER SO SO}/; # Matches the Thai SO SO character
207
208 use charnames 'Cyrillic'; # Loads Cyrillic names.
209 $str =~ /\N{ZHE}\N{KA}/; # Match "ZHE" followed by "KA".
210
211=head3 Octal escapes
212
f0a2b745 213There are two forms of octal escapes. Each is used to specify a character by
17148a1a 214its code point specified in octal notation.
f0a2b745
KW
215
216One form, available starting in Perl 5.14 looks like C<\o{...}>, where the dots
217represent one or more octal digits. It can be used for any Unicode character.
218
219It was introduced to avoid the potential problems with the other form,
220available in all Perls. That form consists of a backslash followed by three
221octal digits. One problem with this form is that it can look exactly like an
222old-style backreference (see
223L</Disambiguation rules between old-style octal escapes and backreferences>
224below.) You can avoid this by making the first of the three digits always a
9645299c 225zero, but that makes \077 the largest code point specifiable.
f0a2b745
KW
226
227In some contexts, a backslash followed by two or even one octal digits may be
228interpreted as an octal escape, sometimes with a warning, and because of some
229bugs, sometimes with surprising results. Also, if you are creating a regex
c69ca1d4 230out of smaller snippets concatenated together, and you use fewer than three
f0a2b745
KW
231digits, the beginning of one snippet may be interpreted as adding digits to the
232ending of the snippet before it. See L</Absolute referencing> for more
233discussion and examples of the snippet problem.
8a118206 234
b6538e4f
TC
235Note that a character expressed as an octal escape is considered
236a character without special meaning by the regex engine, and will match
8a118206
RGS
237"as is".
238
f0a2b745 239To summarize, the C<\o{}> form is always safe to use, and the other form is
17148a1a 240safe to use for code points through \077 when you use exactly three digits to
f0a2b745 241specify them.
8a118206 242
f0a2b745 243Mnemonic: I<0>ctal or I<o>ctal.
8a118206 244
f0a2b745 245=head4 Examples (assuming an ASCII platform)
8a118206 246
f0a2b745
KW
247 $str = "Perl";
248 $str =~ /\o{120}/; # Match, "\120" is "P".
249 $str =~ /\120/; # Same.
f321be7e
SK
250 $str =~ /\o{120}+/; # Match, "\120" is "P",
251 # it's repeated at least once.
f0a2b745
KW
252 $str =~ /\120+/; # Same.
253 $str =~ /P\053/; # No match, "\053" is "+" and taken literally.
254 /\o{23073}/ # Black foreground, white background smiling face.
f321be7e 255 /\o{4801234567}/ # Raises a warning, and yields chr(4).
f0a2b745
KW
256
257=head4 Disambiguation rules between old-style octal escapes and backreferences
258
259Octal escapes of the C<\000> form outside of bracketed character classes
f321be7e 260potentially clash with old-style backreferences (see L</Absolute referencing>
f0a2b745
KW
261below). They both consist of a backslash followed by numbers. So Perl has to
262use heuristics to determine whether it is a backreference or an octal escape.
263Perl uses the following rules to disambiguate:
8a118206
RGS
264
265=over 4
266
267=item 1
268
353c6505 269If the backslash is followed by a single digit, it's a backreference.
8a118206
RGS
270
271=item 2
272
273If the first digit following the backslash is a 0, it's an octal escape.
274
275=item 3
276
b6538e4f
TC
277If the number following the backslash is N (in decimal), and Perl already
278has seen N capture groups, Perl considers this a backreference. Otherwise,
279it considers it an octal escape. If N has more than three digits, Perl
280takes only the first three for the octal escape; the rest are matched as is.
8a118206
RGS
281
282 my $pat = "(" x 999;
283 $pat .= "a";
284 $pat .= ")" x 999;
285 /^($pat)\1000$/; # Matches 'aa'; there are 1000 capture groups.
286 /^$pat\1000$/; # Matches 'a@0'; there are 999 capture groups
f321be7e 287 # and \1000 is seen as \100 (a '@') and a '0'.
8a118206
RGS
288
289=back
290
17148a1a 291You can force a backreference interpretation always by using the C<\g{...}>
f0a2b745
KW
292form. You can the force an octal interpretation always by using the C<\o{...}>
293form, or for numbers up through \077 (= 63 decimal), by using three digits,
294beginning with a "0".
295
8a118206
RGS
296=head3 Hexadecimal escapes
297
f0a2b745 298Like octal escapes, there are two forms of hexadecimal escapes, but both start
febd1aee 299with the sequence C<\x>. This is followed by either exactly two hexadecimal
f0a2b745
KW
300digits forming a number, or a hexadecimal number of arbitrary length surrounded
301by curly braces. The hexadecimal number is the code point of the character you
302want to express.
8a118206 303
b6538e4f
TC
304Note that a character expressed as one of these escapes is considered a
305character without special meaning by the regex engine, and will match
8a118206
RGS
306"as is".
307
308Mnemonic: heI<x>adecimal.
309
9f5650a8 310=head4 Examples (assuming an ASCII platform)
8a118206
RGS
311
312 $str = "Perl";
313 $str =~ /\x50/; # Match, "\x50" is "P".
f822d0dd 314 $str =~ /\x50+/; # Match, "\x50" is "P", it is repeated at least once
8a118206
RGS
315 $str =~ /P\x2B/; # No match, "\x2B" is "+" and taken literally.
316
317 /\x{2603}\x{2602}/ # Snowman with an umbrella.
318 # The Unicode character 2603 is a snowman,
319 # the Unicode character 2602 is an umbrella.
320 /\x{263B}/ # Black smiling face.
321 /\x{263b}/ # Same, the hex digits A - F are case insensitive.
322
323=head2 Modifiers
324
325A number of backslash sequences have to do with changing the character,
326or characters following them. C<\l> will lowercase the character following
5f2b17ca 327it, while C<\u> will uppercase (or, more accurately, titlecase) the
b6538e4f
TC
328character following it. They provide functionality similar to the
329functions C<lcfirst> and C<ucfirst>.
8a118206
RGS
330
331To uppercase or lowercase several characters, one might want to use
332C<\L> or C<\U>, which will lowercase/uppercase all characters following
b6538e4f 333them, until either the end of the pattern or the next occurrence of
17148a1a 334C<\E>, whichever comes first. They provide functionality similar to what
b6538e4f 335the functions C<lc> and C<uc> provide.
8a118206 336
736fe711
KW
337C<\Q> is used to quote (disable) pattern metacharacters, up to the next
338C<\E> or the end of the pattern. C<\Q> adds a backslash to any character
339that could have special meaning to Perl. In the ASCII range, it quotes
340every character that isn't a letter, digit, or underscore. See
341L<perlfunc/quotemeta> for details on what gets quoted for non-ASCII
342code points. Using this ensures that any character between C<\Q> and
343C<\E> will be matched literally, not interpreted as a metacharacter by
344the regex engine.
8a118206 345
628253b8
BF
346C<\F> can be used to casefold all characters following, up to the next C<\E>
347or the end of the pattern. It provides the functionality similar to
348the C<fc> function.
349
350Mnemonic: I<L>owercase, I<U>ppercase, I<F>old-case, I<Q>uotemeta, I<E>nd.
8a118206
RGS
351
352=head4 Examples
353
354 $sid = "sid";
355 $greg = "GrEg";
356 $miranda = "(Miranda)";
357 $str =~ /\u$sid/; # Matches 'Sid'
358 $str =~ /\L$greg/; # Matches 'greg'
359 $str =~ /\Q$miranda\E/; # Matches '(Miranda)', as if the pattern
360 # had been written as /\(Miranda\)/
361
362=head2 Character classes
363
364Perl regular expressions have a large range of character classes. Some of
365the character classes are written as a backslash sequence. We will briefly
366discuss those here; full details of character classes can be found in
367L<perlrecharclass>.
368
d35dd6c6
KW
369C<\w> is a character class that matches any single I<word> character
370(letters, digits, Unicode marks, and connector punctuation (like the
371underscore)). C<\d> is a character class that matches any decimal
372digit, while the character class C<\s> matches any whitespace character.
99d59c4d 373New in perl 5.10.0 are the classes C<\h> and C<\v> which match horizontal
418e7b04 374and vertical whitespace characters.
cfaf538b
KW
375
376The exact set of characters matched by C<\d>, C<\s>, and C<\w> varies
9645299c
KW
377depending on various pragma and regular expression modifiers. It is
378possible to restrict the match to the ASCII range by using the C</a>
379regular expression modifier. See L<perlrecharclass>.
8a118206
RGS
380
381The uppercase variants (C<\W>, C<\D>, C<\S>, C<\H>, and C<\V>) are
e486b3cc
KW
382character classes that match, respectively, any character that isn't a
383word character, digit, whitespace, horizontal whitespace, or vertical
384whitespace.
8a118206
RGS
385
386Mnemonics: I<w>ord, I<d>igit, I<s>pace, I<h>orizontal, I<v>ertical.
387
388=head3 Unicode classes
389
390C<\pP> (where C<P> is a single letter) and C<\p{Property}> are used to
391match a character that matches the given Unicode property; properties
392include things like "letter", or "thai character". Capitalizing the
393sequence to C<\PP> and C<\P{Property}> make the sequence match a character
394that doesn't match the given Unicode property. For more details, see
4948b50f 395L<perlrecharclass/Backslash sequences> and
8a118206
RGS
396L<perlunicode/Unicode Character Properties>.
397
398Mnemonic: I<p>roperty.
399
8a118206
RGS
400=head2 Referencing
401
402If capturing parenthesis are used in a regular expression, we can refer
403to the part of the source string that was matched, and match exactly the
1843fd28
RGS
404same thing. There are three ways of referring to such I<backreference>:
405absolutely, relatively, and by name.
406
407=for later add link to perlrecapture
8a118206
RGS
408
409=head3 Absolute referencing
410
c27a5cfe 411Either C<\gI<N>> (starting in Perl 5.10.0), or C<\I<N>> (old-style) where I<N>
d8b950dc 412is a positive (unsigned) decimal number of any length is an absolute reference
c27a5cfe
KW
413to a capturing group.
414
8e4698ef
KW
415I<N> refers to the Nth set of parentheses, so C<\gI<N>> refers to whatever has
416been matched by that set of parentheses. Thus C<\g1> refers to the first
c27a5cfe
KW
417capture group in the regex.
418
419The C<\gI<N>> form can be equivalently written as C<\g{I<N>}>
420which avoids ambiguity when building a regex by concatenating shorter
d8b950dc
KW
421strings. Otherwise if you had a regex C<qr/$a$b/>, and C<$a> contained
422C<"\g1">, and C<$b> contained C<"37">, you would get C</\g137/> which is
423probably not what you intended.
c27a5cfe
KW
424
425In the C<\I<N>> form, I<N> must not begin with a "0", and there must be at
b6538e4f
TC
426least I<N> capturing groups, or else I<N> is considered an octal escape
427(but something like C<\18> is the same as C<\0018>; that is, the octal escape
c27a5cfe
KW
428C<"\001"> followed by a literal digit C<"8">).
429
430Mnemonic: I<g>roup.
8a118206
RGS
431
432=head4 Examples
433
c27a5cfe 434 /(\w+) \g1/; # Finds a duplicated word, (e.g. "cat cat").
f321be7e 435 /(\w+) \1/; # Same thing; written old-style.
c27a5cfe 436 /(.)(.)\g2\g1/; # Match a four letter palindrome (e.g. "ABBA").
8a118206
RGS
437
438
439=head3 Relative referencing
440
c27a5cfe
KW
441C<\g-I<N>> (starting in Perl 5.10.0) is used for relative addressing. (It can
442be written as C<\g{-I<N>>.) It refers to the I<N>th group before the
443C<\g{-I<N>}>.
8a118206 444
c27a5cfe 445The big advantage of this form is that it makes it much easier to write
8a118206
RGS
446patterns with references that can be interpolated in larger patterns,
447even if the larger pattern also contains capture groups.
448
8a118206
RGS
449=head4 Examples
450
c27a5cfe
KW
451 /(A) # Group 1
452 ( # Group 2
453 (B) # Group 3
454 \g{-1} # Refers to group 3 (B)
455 \g{-3} # Refers to group 1 (A)
8a118206
RGS
456 )
457 /x; # Matches "ABBA".
458
459 my $qr = qr /(.)(.)\g{-2}\g{-1}/; # Matches 'abab', 'cdcd', etc.
460 /$qr$qr/ # Matches 'ababcdcd'.
461
462=head3 Named referencing
463
d8b950dc
KW
464C<\g{I<name>}> (starting in Perl 5.10.0) can be used to back refer to a
465named capture group, dispensing completely with having to think about capture
466buffer positions.
8a118206
RGS
467
468To be compatible with .Net regular expressions, C<\g{name}> may also be
469written as C<\k{name}>, C<< \k<name> >> or C<\k'name'>.
470
d8b950dc
KW
471To prevent any ambiguity, I<name> must not start with a digit nor contain a
472hyphen.
8a118206
RGS
473
474=head4 Examples
475
476 /(?<word>\w+) \g{word}/ # Finds duplicated word, (e.g. "cat cat")
477 /(?<word>\w+) \k{word}/ # Same.
478 /(?<word>\w+) \k<word>/ # Same.
479 /(?<letter1>.)(?<letter2>.)\g{letter2}\g{letter1}/
480 # Match a four letter palindrome (e.g. "ABBA")
481
482=head2 Assertions
483
ac036724 484Assertions are conditions that have to be true; they don't actually
8a118206
RGS
485match parts of the substring. There are six assertions that are written as
486backslash sequences.
487
488=over 4
489
490=item \A
491
492C<\A> only matches at the beginning of the string. If the C</m> modifier
1726f7e8 493isn't used, then C</\A/> is equivalent to C</^/>. However, if the C</m>
8a118206
RGS
494modifier is used, then C</^/> matches internal newlines, but the meaning
495of C</\A/> isn't changed by the C</m> modifier. C<\A> matches at the beginning
496of the string regardless whether the C</m> modifier is used.
497
498=item \z, \Z
499
500C<\z> and C<\Z> match at the end of the string. If the C</m> modifier isn't
b6538e4f
TC
501used, then C</\Z/> is equivalent to C</$/>; that is, it matches at the
502end of the string, or one before the newline at the end of the string. If the
8a118206
RGS
503C</m> modifier is used, then C</$/> matches at internal newlines, but the
504meaning of C</\Z/> isn't changed by the C</m> modifier. C<\Z> matches at
505the end of the string (or just before a trailing newline) regardless whether
506the C</m> modifier is used.
507
b6538e4f
TC
508C<\z> is just like C<\Z>, except that it does not match before a trailing
509newline. C<\z> matches at the end of the string only, regardless of the
510modifiers used, and not just before a newline. It is how to anchor the
511match to the true end of the string under all conditions.
8a118206
RGS
512
513=item \G
514
b6538e4f
TC
515C<\G> is usually used only in combination with the C</g> modifier. If the
516C</g> modifier is used and the match is done in scalar context, Perl
517remembers where in the source string the last match ended, and the next time,
8a118206
RGS
518it will start the match from where it ended the previous time.
519
b6538e4f
TC
520C<\G> matches the point where the previous match on that string ended,
521or the beginning of that string if there was no previous match.
1843fd28
RGS
522
523=for later add link to perlremodifiers
8a118206
RGS
524
525Mnemonic: I<G>lobal.
526
64935bc6 527=item \b{}, \b, \B{}, \B
8a118206 528
64935bc6
KW
529C<\b{...}>, available starting in v5.22, matches a boundary (between two
530characters, or before the first character of the string, or after the
531final character of the string) based on the Unicode rules for the
6b659339 532boundary type specified inside the braces. The boundary
64935bc6
KW
533types are given a few paragraphs below. C<\B{...}> matches at any place
534between characters where C<\b{...}> of the same type doesn't match.
535
536C<\b> when not immediately followed by a C<"{"> matches at any place
537between a word (something matched by C<\w>) and a non-word character
538(C<\W>); C<\B> when not immediately followed by a C<"{"> matches at any
ae3bb8ea 539place between characters where C<\b> doesn't match. To get better
a95b3d6a 540word matching of natural language text, see L</\b{wb}> below.
64935bc6
KW
541
542C<\b>
8a118206
RGS
543and C<\B> assume there's a non-word character before the beginning and after
544the end of the source string; so C<\b> will match at the beginning (or end)
545of the source string if the source string begins (or ends) with a word
b6538e4f
TC
546character. Otherwise, C<\B> will match.
547
548Do not use something like C<\b=head\d\b> and expect it to match the
549beginning of a line. It can't, because for there to be a boundary before
550the non-word "=", there must be a word character immediately previous.
64935bc6
KW
551All plain C<\b> and C<\B> boundary determinations look for word
552characters alone, not for
553non-word characters nor for string ends. It may help to understand how
6b659339 554C<\b> and C<\B> work by equating them as follows:
b6538e4f
TC
555
556 \b really means (?:(?<=\w)(?!\w)|(?<!\w)(?=\w))
557 \B really means (?:(?<=\w)(?=\w)|(?<!\w)(?!\w))
8a118206 558
54bdcd8e
KW
559In contrast, C<\b{...}> and C<\B{...}> may or may not match at the
560beginning and end of the line, depending on the boundary type. These
561implement the Unicode default boundaries, specified in
6b659339 562L<http://www.unicode.org/reports/tr14/> and
54bdcd8e 563L<http://www.unicode.org/reports/tr29/>.
6b659339 564The boundary types are:
ae3bb8ea
KW
565
566=over
567
568=item C<\b{gcb}> or C<\b{g}>
569
570This matches a Unicode "Grapheme Cluster Boundary". (Actually Perl
571always uses the improved "extended" grapheme cluster"). These are
572explained below under L</C<\X>>. In fact, C<\X> is another way to get
573the same functionality. It is equivalent to C</.+?\b{gcb}/>. Use
574whichever is most convenient for your situation.
575
6b659339
KW
576=item C<\b{lb}>
577
578This matches according to the default Unicode Line Breaking Algorithm
579(L<http://www.unicode.org/reports/tr14/>), as customized in that
580document
581(L<Example 7 of revision 35|http://www.unicode.org/reports/tr14/tr14-35.html#Example7>)
582for better handling of numeric expressions.
583
584This is suitable for many purposes, but the L<Unicode::LineBreak> module
585is available on CPAN that provides many more features, including
586customization.
587
06ae2722
KW
588=item C<\b{sb}>
589
590This matches a Unicode "Sentence Boundary". This is an aid to parsing
591natural language sentences. It gives good, but imperfect results. For
592example, it thinks that "Mr. Smith" is two sentences. More details are
d90f6819
KW
593at L<http://www.unicode.org/reports/tr29/>. Note also that it thinks
594that anything matching L</\R> (except form feed and vertical tab) is a
54bdcd8e
KW
595sentence boundary. C<\b{sb}> works with text designed for
596word-processors which wrap lines
d90f6819
KW
597automatically for display, but hard-coded line boundaries are considered
598to be essentially the ends of text blocks (paragraphs really), and hence
54bdcd8e
KW
599the ends of sententces. C<\b{sb}> doesn't do well with text containing
600embedded newlines, like the source text of the document you are reading.
601Such text needs to be preprocessed to get rid of the line separators
602before looking for sentence boundaries. Some people view this as a bug
268e6905
KW
603in the Unicode standard, and this behavior is quite subject to change in
604future Perl versions.
06ae2722 605
ae3bb8ea
KW
606=item C<\b{wb}>
607
f1f6961f
KW
608This matches a Unicode "Word Boundary", but tailored to Perl
609expectations. This gives better (though not
ae3bb8ea
KW
610perfect) results for natural language processing than plain C<\b>
611(without braces) does. For example, it understands that apostrophes can
54bdcd8e 612be in the middle of words and that parentheses aren't (see the examples
39a8449e 613below). More details are at L<http://www.unicode.org/reports/tr29/>.
ae3bb8ea 614
f1f6961f
KW
615The current Unicode definition of a Word Boundary matches between every
616white space character. Perl tailors this, starting in version 5.24, to
617generally not break up spans of white space, just as plain C<\b> has
618always functioned. This allows C<\b{wb}> to be a drop-in replacement for
619C<\b>, but with generally better results for natural language
620processing. (The exception to this tailoring is when a span of white
621space is immediately followed by something like U+0303, COMBINING TILDE.
622If the final space character in the span is a horizontal white space, it
623is broken out so that it attaches instead to the combining character.
624To be precise, if a span of white space that ends in a horizontal space
be8bb1ac 625has the character immediately following it have any of the Word
b0e24409 626Boundary property values "Extend", "Format" or "ZWJ", the boundary between the
f1f6961f
KW
627final horizontal space character and the rest of the span matches
628C<\b{wb}>. In all other cases the boundary between two white space
629characters matches C<\B{wb}>.)
630
ae3bb8ea 631=back
64935bc6 632
1ce7b77d
KW
633It is important to realize when you use these Unicode boundaries,
634that you are taking a risk that a future version of Perl which contains
635a later version of the Unicode Standard will not work precisely the same
636way as it did when your code was written. These rules are not
637considered stable and have been somewhat more subject to change than the
638rest of the Standard. Unicode reserves the right to change them at
639will, and Perl reserves the right to update its implementation to
640Unicode's new rules. In the past, some changes have been because new
641characters have been added to the Standard which have different
642characteristics than all previous characters, so new rules are
643formulated for handling them. These should not cause any backward
644compatibility issues. But some changes have changed the treatment of
645existing characters because the Unicode Technical Committee has decided
646that the change is warranted for whatever reason. This could be to fix
647a bug, or because they think better results are obtained with the new
648rule.
649
650It is also important to realize that these are default boundary
651definitions, and that implementations may wish to tailor the results for
39a8449e 652particular purposes and locales. For example, some languages, such as
be8bb1ac
KW
653Japanese and Thai, require dictionary lookup to accurately determine
654word boundaries.
54bdcd8e 655
8a118206
RGS
656Mnemonic: I<b>oundary.
657
658=back
659
660=head4 Examples
661
662 "cat" =~ /\Acat/; # Match.
663 "cat" =~ /cat\Z/; # Match.
664 "cat\n" =~ /cat\Z/; # Match.
665 "cat\n" =~ /cat\z/; # No match.
666
667 "cat" =~ /\bcat\b/; # Matches.
668 "cats" =~ /\bcat\b/; # No match.
669 "cat" =~ /\bcat\B/; # No match.
670 "cats" =~ /\bcat\B/; # Match.
671
672 while ("cat dog" =~ /(\w+)/g) {
673 print $1; # Prints 'catdog'
674 }
675 while ("cat dog" =~ /\G(\w+)/g) {
676 print $1; # Prints 'cat'
677 }
678
54bdcd8e
KW
679 my $s = "He said, \"Is pi 3.14? (I'm not sure).\"";
680 print join("|", $s =~ m/ ( .+? \b ) /xg), "\n";
681 print join("|", $s =~ m/ ( .+? \b{wb} ) /xg), "\n";
ae3bb8ea 682 prints
54bdcd8e
KW
683 He| |said|, "|Is| |pi| |3|.|14|? (|I|'|m| |not| |sure
684 He| |said|,| |"|Is| |pi| |3.14|?| |(|I'm| |not| |sure|)|.|"
ae3bb8ea 685
8a118206
RGS
686=head2 Misc
687
688Here we document the backslash sequences that don't fall in one of the
b6538e4f 689categories above. These are:
8a118206
RGS
690
691=over 4
692
8a118206
RGS
693=item \K
694
b6538e4f
TC
695This appeared in perl 5.10.0. Anything matched left of C<\K> is
696not included in C<$&>, and will not be replaced if the pattern is
697used in a substitution. This lets you write C<s/PAT1 \K PAT2/REPL/x>
8a118206
RGS
698instead of C<s/(PAT1) PAT2/${1}REPL/x> or C<s/(?<=PAT1) PAT2/REPL/x>.
699
700Mnemonic: I<K>eep.
701
df225385
KW
702=item \N
703
2171640d 704This feature, available starting in v5.12, matches any character
b6538e4f 705that is B<not> a newline. It is a short-hand for writing C<[^\n]>, and is
b3b85878
KW
706identical to the C<.> metasymbol, except under the C</s> flag, which changes
707the meaning of C<.>, but not C<\N>.
df225385 708
e526e8bb 709Note that C<\N{...}> can mean a
fb121860
KW
710L<named or numbered character
711|/Named or numbered characters and character sequences>.
df225385
KW
712
713Mnemonic: Complement of I<\n>.
714
8a118206 715=item \R
6b46370c 716X<\R>
8a118206 717
b6538e4f
TC
718C<\R> matches a I<generic newline>; that is, anything considered a
719linebreak sequence by Unicode. This includes all characters matched by
720C<\v> (vertical whitespace), and the multi character sequence C<"\x0D\x0A">
721(carriage return followed by a line feed, sometimes called the network
722newline; it's the end of line sequence used in Microsoft text files opened
1978b668 723in binary mode). C<\R> is equivalent to C<< (?>\x0D\x0A|\v) >>. (The
040ac264
FC
724reason it doesn't backtrack is that the sequence is considered
725inseparable. That means that
1978b668
KW
726
727 "\x0D\x0A" =~ /^\R\x0A$/ # No match
728
729fails, because the C<\R> matches the entire string, and won't backtrack
730to match just the C<"\x0D">.) Since
b6538e4f
TC
731C<\R> can match a sequence of more than one character, it cannot be put
732inside a bracketed character class; C</[\R]/> is an error; use C<\v>
733instead. C<\R> was introduced in perl 5.10.0.
8a118206 734
8129baca
KW
735Note that this does not respect any locale that might be in effect; it
736matches according to the platform's native character set.
737
10fdd326
JH
738Mnemonic: none really. C<\R> was picked because PCRE already uses C<\R>,
739and more importantly because Unicode recommends such a regular expression
b6538e4f 740metacharacter, and suggests C<\R> as its notation.
8a118206
RGS
741
742=item \X
6b46370c 743X<\X>
8a118206 744
0111a78f 745This matches a Unicode I<extended grapheme cluster>.
8a118206 746
10fdd326 747C<\X> matches quite well what normal (non-Unicode-programmer) usage
0111a78f 748would consider a single character. As an example, consider a G with some sort
c670e63a 749of diacritic mark, such as an arrow. There is no such single character in
df225385 750Unicode, but one can be composed by using a G followed by a Unicode "COMBINING
c670e63a
KW
751UPWARDS ARROW BELOW", and would be displayed by Unicode-aware software as if it
752were a single character.
10fdd326 753
aa9e685b
KW
754The match is greedy and non-backtracking, so that the cluster is never
755broken up into smaller components.
756
64935bc6
KW
757See also L<C<\b{gcb}>|/\b{}, \b, \B{}, \B>.
758
8a118206
RGS
759Mnemonic: eI<X>tended Unicode character.
760
761=back
762
763=head4 Examples
764
f822d0dd 765 $str =~ s/foo\Kbar/baz/g; # Change any 'bar' following a 'foo' to 'baz'
d8b950dc 766 $str =~ s/(.)\K\g1//g; # Delete duplicated characters.
8a118206
RGS
767
768 "\n" =~ /^\R$/; # Match, \n is a generic newline.
769 "\r" =~ /^\R$/; # Match, \r is a generic newline.
770 "\r\n" =~ /^\R$/; # Match, \r\n is a generic newline.
771
b6538e4f 772 "P\x{307}" =~ /^\X$/ # \X matches a P with a dot above.
8a118206
RGS
773
774=cut