This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlrecharclass: A few clarifications
[perl5.git] / pod / perlrecharclass.pod
CommitLineData
8a118206 1=head1 NAME
ea449505 2X<character class>
8a118206
RGS
3
4perlrecharclass - Perl Regular Expression Character Classes
5
6=head1 DESCRIPTION
7
8The top level documentation about Perl regular expressions
9is found in L<perlre>.
10
11This manual page discusses the syntax and use of character
6b83a163 12classes in Perl regular expressions.
8a118206 13
6b83a163 14A character class is a way of denoting a set of characters
8a118206 15in such a way that one character of the set is matched.
6b83a163 16It's important to remember that: matching a character class
8a118206
RGS
17consumes exactly one character in the source string. (The source
18string is the string the regular expression is matched against.)
19
20There are three types of character classes in Perl regular
6b83a163 21expressions: the dot, backslash sequences, and the form enclosed in square
ea449505 22brackets. Keep in mind, though, that often the term "character class" is used
6b83a163 23to mean just the bracketed form. Certainly, most Perl documentation does that.
8a118206
RGS
24
25=head2 The dot
26
27The dot (or period), C<.> is probably the most used, and certainly
28the most well-known character class. By default, a dot matches any
5db9882c 29character, except for the newline. That default can be changed to
4a88d526 30add matching the newline by using the I<single line> modifier:
6b83a163 31for the entire regular expression with the C</s> modifier, or
4a88d526
KW
32locally with C<(?s)> (and even globally within the scope of
33L<C<use re '/s'>|re/'E<sol>flags' mode>). (The C<L</\N>> backslash
34sequence, described
6b83a163
KW
35below, matches any character except newline without regard to the
36I<single line> modifier.)
8a118206
RGS
37
38Here are some examples:
39
40 "a" =~ /./ # Match
41 "." =~ /./ # Match
42 "" =~ /./ # No match (dot has to match a character)
43 "\n" =~ /./ # No match (dot does not match a newline)
44 "\n" =~ /./s # Match (global 'single line' modifier)
45 "\n" =~ /(?s:.)/ # Match (local 'single line' modifier)
46 "ab" =~ /^.$/ # No match (dot matches one character)
47
6b83a163 48=head2 Backslash sequences
82206b5e 49X<\w> X<\W> X<\s> X<\S> X<\d> X<\D> X<\p> X<\P>
ea449505
KW
50X<\N> X<\v> X<\V> X<\h> X<\H>
51X<word> X<whitespace>
8a118206 52
6b83a163
KW
53A backslash sequence is a sequence of characters, the first one of which is a
54backslash. Perl ascribes special meaning to many such sequences, and some of
55these are character classes. That is, they match a single character each,
56provided that the character belongs to the specific set of characters defined
57by the sequence.
8a118206 58
6b83a163
KW
59Here's a list of the backslash sequences that are character classes. They
60are discussed in more detail below. (For the backslash sequences that aren't
61character classes, see L<perlrebackslash>.)
8a118206 62
6b83a163
KW
63 \d Match a decimal digit character.
64 \D Match a non-decimal-digit character.
8a118206
RGS
65 \w Match a "word" character.
66 \W Match a non-"word" character.
ea449505
KW
67 \s Match a whitespace character.
68 \S Match a non-whitespace character.
69 \h Match a horizontal whitespace character.
70 \H Match a character that isn't horizontal whitespace.
ea449505
KW
71 \v Match a vertical whitespace character.
72 \V Match a character that isn't vertical whitespace.
4e5e0888 73 \N Match a character that isn't a newline.
6b83a163 74 \pP, \p{Prop} Match a character that has the given Unicode property.
6c5a041f 75 \PP, \P{Prop} Match a character that doesn't have the Unicode property
8a118206 76
1433f837
KW
77=head3 \N
78
2171640d 79C<\N>, available starting in v5.12, like the dot, matches any
1433f837
KW
80character that is not a newline. The difference is that C<\N> is not influenced
81by the I<single line> regular expression modifier (see L</The dot> above). Note
82that the form C<\N{...}> may mean something completely different. When the
83C<{...}> is a L<quantifier|perlre/Quantifiers>, it means to match a non-newline
84character that many times. For example, C<\N{3}> means to match 3
85non-newlines; C<\N{5,}> means to match 5 or more non-newlines. But if C<{...}>
86is not a legal quantifier, it is presumed to be a named character. See
87L<charnames> for those. For example, none of C<\N{COLON}>, C<\N{4F}>, and
88C<\N{F4}> contain legal quantifiers, so Perl will try to find characters whose
89names are respectively C<COLON>, C<4F>, and C<F4>.
90
8a118206
RGS
91=head3 Digits
92
b6538e4f 93C<\d> matches a single character considered to be a decimal I<digit>.
5db9882c 94If the C</a> regular expression modifier is in effect, it matches [0-9].
582da942 95Otherwise, it
82206b5e
KW
96matches anything that is matched by C<\p{Digit}>, which includes [0-9].
97(An unlikely possible exception is that under locale matching rules, the
d66e1f56
KW
98current locale might not have C<[0-9]> matched by C<\d>, and/or might match
99other characters whose code point is less than 256. The only such locale
100definitions that are legal would be to match C<[0-9]> plus another set of
10110 consecutive digit characters; anything else would be in violation of
102the C language standard, but Perl doesn't currently assume anything in
103regard to this.)
82206b5e
KW
104
105What this means is that unless the C</a> modifier is in effect C<\d> not
106only matches the digits '0' - '9', but also Arabic, Devanagari, and
107digits from other languages. This may cause some confusion, and some
108security issues.
109
110Some digits that C<\d> matches look like some of the [0-9] ones, but
111have different values. For example, BENGALI DIGIT FOUR (U+09EA) looks
112very much like an ASCII DIGIT EIGHT (U+0038). An application that
113is expecting only the ASCII digits might be misled, or if the match is
114C<\d+>, the matched string might contain a mixture of digits from
115different writing systems that look like they signify a number different
67592e11 116than they actually do. L<Unicode::UCD/num()> can
e397bccf 117be used to safely
82206b5e
KW
118calculate the value, returning C<undef> if the input string contains
119such a mixture.
120
121What C<\p{Digit}> means (and hence C<\d> except under the C</a>
122modifier) is C<\p{General_Category=Decimal_Number}>, or synonymously,
123C<\p{General_Category=Digit}>. Starting with Unicode version 4.1, this
124is the same set of characters matched by C<\p{Numeric_Type=Decimal}>.
6b83a163
KW
125But Unicode also has a different property with a similar name,
126C<\p{Numeric_Type=Digit}>, which matches a completely different set of
82206b5e
KW
127characters. These characters are things such as C<CIRCLED DIGIT ONE>
128or subscripts, or are from writing systems that lack all ten digits.
6b83a163 129
82206b5e
KW
130The design intent is for C<\d> to exactly match the set of characters
131that can safely be used with "normal" big-endian positional decimal
132syntax, where, for example 123 means one 'hundred', plus two 'tens',
133plus three 'ones'. This positional notation does not necessarily apply
134to characters that match the other type of "digit",
135C<\p{Numeric_Type=Digit}>, and so C<\d> doesn't match them.
6b83a163 136
e2cfb18c 137The Tamil digits (U+0BE6 - U+0BEF) can also legally be
82206b5e
KW
138used in old-style Tamil numbers in which they would appear no more than
139one in a row, separated by characters that mean "times 10", "times 100",
140etc. (See L<http://www.unicode.org/notes/tn21>.)
8a118206 141
b6538e4f 142Any character not matched by C<\d> is matched by C<\D>.
8a118206
RGS
143
144=head3 Word characters
145
ea449505 146A C<\w> matches a single alphanumeric character (an alphabetic character, or a
41805eb9
KW
147decimal digit); or a connecting punctuation character, such as an
148underscore ("_"); or a "mark" character (like some sort of accent) that
149attaches to one of those. It does not match a whole word. To match a
150whole word, use C<\w+>. This isn't the same thing as matching an
151English word, but in the ASCII range it is the same as a string of
152Perl-identifier characters.
82206b5e
KW
153
154=over
155
156=item If the C</a> modifier is in effect ...
157
158C<\w> matches the 63 characters [a-zA-Z0-9_].
159
160=item otherwise ...
161
162=over
163
164=item For code points above 255 ...
165
166C<\w> matches the same as C<\p{Word}> matches in this range. That is,
167it matches Thai letters, Greek letters, etc. This includes connector
d35dd6c6 168punctuation (like the underscore) which connect two words together, or
b6538e4f 169diacritics, such as a C<COMBINING TILDE> and the modifier letters, which
82206b5e
KW
170are generally used to add auxiliary markings to letters.
171
172=item For code points below 256 ...
173
174=over
175
176=item if locale rules are in effect ...
177
178C<\w> matches the platform's native underscore character plus whatever
179the locale considers to be alphanumeric.
180
4a88d526 181=item if instead, Unicode rules are in effect ...
82206b5e
KW
182
183C<\w> matches exactly what C<\p{Word}> matches.
184
185=item otherwise ...
186
187C<\w> matches [a-zA-Z0-9_].
188
189=back
190
191=back
192
193=back
194
195Which rules apply are determined as described in L<perlre/Which character set modifier is in effect?>.
8a118206 196
6b83a163
KW
197There are a number of security issues with the full Unicode list of word
198characters. See L<http://unicode.org/reports/tr36>.
199
200Also, for a somewhat finer-grained set of characters that are in programming
201language identifiers beyond the ASCII range, you may wish to instead use the
e2cfb18c
KW
202more customized L</Unicode Properties>, C<\p{ID_Start}>,
203C<\p{ID_Continue}>, C<\p{XID_Start}>, and C<\p{XID_Continue}>. See
204L<http://unicode.org/reports/tr31>.
6b83a163 205
b6538e4f 206Any character not matched by C<\w> is matched by C<\W>.
8a118206 207
ea449505
KW
208=head3 Whitespace
209
82206b5e
KW
210C<\s> matches any single character considered whitespace.
211
212=over
213
214=item If the C</a> modifier is in effect ...
215
d28d8023
KW
216In all Perl versions, C<\s> matches the 5 characters [\t\n\f\r ]; that
217is, the horizontal tab,
218the newline, the form feed, the carriage return, and the space.
779cf272 219Starting in Perl v5.18, it also matches the vertical tab, C<\cK>.
d28d8023 220See note C<[1]> below for a discussion of this.
82206b5e
KW
221
222=item otherwise ...
223
224=over
225
226=item For code points above 255 ...
227
228C<\s> matches exactly the code points above 255 shown with an "s" column
229in the table below.
230
231=item For code points below 256 ...
232
233=over
234
235=item if locale rules are in effect ...
236
d28d8023 237C<\s> matches whatever the locale considers to be whitespace.
82206b5e 238
4a88d526 239=item if instead, Unicode rules are in effect ...
82206b5e
KW
240
241C<\s> matches exactly the characters shown with an "s" column in the
242table below.
243
244=item otherwise ...
245
779cf272 246C<\s> matches [\t\n\f\r ] and, starting in Perl
d28d8023
KW
247v5.18, the vertical tab, C<\cK>.
248(See note C<[1]> below for a discussion of this.)
82206b5e
KW
249Note that this list doesn't include the non-breaking space.
250
251=back
252
253=back
254
255=back
256
257Which rules apply are determined as described in L<perlre/Which character set modifier is in effect?>.
8a118206 258
b6538e4f 259Any character not matched by C<\s> is matched by C<\S>.
8a118206 260
b6538e4f 261C<\h> matches any character considered horizontal whitespace;
8129baca 262this includes the platform's space and tab characters and several others
b6538e4f 263listed in the table below. C<\H> matches any character
8129baca
KW
264not considered horizontal whitespace. They use the platform's native
265character set, and do not consider any locale that may otherwise be in
266use.
ea449505 267
b6538e4f 268C<\v> matches any character considered vertical whitespace;
8129baca 269this includes the platform's carriage return and line feed characters (newline)
b6538e4f
TC
270plus several other characters, all listed in the table below.
271C<\V> matches any character not considered vertical whitespace.
8129baca
KW
272They use the platform's native character set, and do not consider any
273locale that may otherwise be in use.
8a118206
RGS
274
275C<\R> matches anything that can be considered a newline under Unicode
412a49a2
KW
276rules. It can match a multi-character sequence. It cannot be used inside
277a bracketed character class; use C<\v> instead (vertical whitespace).
278It uses the platform's
8129baca
KW
279native character set, and does not consider any locale that may
280otherwise be in use.
ea449505 281Details are discussed in L<perlrebackslash>.
8a118206 282
82206b5e 283Note that unlike C<\s> (and C<\d> and C<\w>), C<\h> and C<\v> always match
8129baca
KW
284the same characters, without regard to other factors, such as the active
285locale or whether the source string is in UTF-8 format.
8a118206 286
d28d8023
KW
287One might think that C<\s> is equivalent to C<[\h\v]>. This is indeed true
288starting in Perl v5.18, but prior to that, the sole difference was that the
289vertical tab (C<"\cK">) was not matched by C<\s>.
8a118206
RGS
290
291The following table is a complete listing of characters matched by
a9c9e371 292C<\s>, C<\h> and C<\v> as of Unicode 6.3.
8a118206 293
582da942 294The first column gives the Unicode code point of the character (in hex format),
8a118206 295the second column gives the (Unicode) name. The third column indicates
4b9734bf
KW
296by which class(es) the character is matched (assuming no locale is in
297effect that changes the C<\s> matching).
8a118206 298
fc28d2a3
KW
299 0x0009 CHARACTER TABULATION h s
300 0x000a LINE FEED (LF) vs
d28d8023 301 0x000b LINE TABULATION vs [1]
fc28d2a3
KW
302 0x000c FORM FEED (FF) vs
303 0x000d CARRIAGE RETURN (CR) vs
304 0x0020 SPACE h s
d28d8023
KW
305 0x0085 NEXT LINE (NEL) vs [2]
306 0x00a0 NO-BREAK SPACE h s [2]
fc28d2a3 307 0x1680 OGHAM SPACE MARK h s
fc28d2a3
KW
308 0x2000 EN QUAD h s
309 0x2001 EM QUAD h s
310 0x2002 EN SPACE h s
311 0x2003 EM SPACE h s
312 0x2004 THREE-PER-EM SPACE h s
313 0x2005 FOUR-PER-EM SPACE h s
314 0x2006 SIX-PER-EM SPACE h s
315 0x2007 FIGURE SPACE h s
316 0x2008 PUNCTUATION SPACE h s
317 0x2009 THIN SPACE h s
318 0x200a HAIR SPACE h s
319 0x2028 LINE SEPARATOR vs
320 0x2029 PARAGRAPH SEPARATOR vs
321 0x202f NARROW NO-BREAK SPACE h s
322 0x205f MEDIUM MATHEMATICAL SPACE h s
323 0x3000 IDEOGRAPHIC SPACE h s
8a118206
RGS
324
325=over 4
326
327=item [1]
328
779cf272
KW
329Prior to Perl v5.18, C<\s> did not match the vertical tab.
330C<[^\S\cK]> (obscurely) matches what C<\s> traditionally did.
d28d8023
KW
331
332=item [2]
333
82206b5e
KW
334NEXT LINE and NO-BREAK SPACE may or may not match C<\s> depending
335on the rules in effect. See
336L<the beginning of this section|/Whitespace>.
8a118206
RGS
337
338=back
339
8a118206
RGS
340=head3 Unicode Properties
341
c1c4ae3a
KW
342C<\pP> and C<\p{Prop}> are character classes to match characters that fit given
343Unicode properties. One letter property names can be used in the C<\pP> form,
344with the property name following the C<\p>, otherwise, braces are required.
345When using braces, there is a single form, which is just the property name
346enclosed in the braces, and a compound form which looks like C<\p{name=value}>,
b6538e4f 347which means to match if the property "name" for the character has that particular
c1c4ae3a 348"value".
e1b711da
KW
349For instance, a match for a number can be written as C</\pN/> or as
350C</\p{Number}/>, or as C</\p{Number=True}/>.
351Lowercase letters are matched by the property I<Lowercase_Letter> which
e2cfb18c 352has the short form I<Ll>. They need the braces, so are written as C</\p{Ll}/> or
e1b711da
KW
353C</\p{Lowercase_Letter}/>, or C</\p{General_Category=Lowercase_Letter}/>
354(the underscores are optional).
355C</\pLl/> is valid, but means something different.
8a118206
RGS
356It matches a two character string: a letter (Unicode property C<\pL>),
357followed by a lowercase C<l>.
358
bc943be5 359If locale rules are not in effect, the use of
82206b5e 360a Unicode property will force the regular expression into using Unicode
bc943be5 361rules, if it isn't already.
82206b5e 362
56ca34ca
KW
363Note that almost all properties are immune to case-insensitive matching.
364That is, adding a C</i> regular expression modifier does not change what
82206b5e 365they match. There are two sets that are affected. The first set is
56ca34ca
KW
366C<Uppercase_Letter>,
367C<Lowercase_Letter>,
368and C<Titlecase_Letter>,
369all of which match C<Cased_Letter> under C</i> matching.
b6538e4f 370The second set is
56ca34ca
KW
371C<Uppercase>,
372C<Lowercase>,
373and C<Titlecase>,
374all of which match C<Cased> under C</i> matching.
375(The difference between these sets is that some things, such as Roman
e2cfb18c 376numerals, come in both upper and lower case, so they are C<Cased>, but
b6538e4f 377aren't considered to be letters, so they aren't C<Cased_Letter>s. They're
82206b5e
KW
378actually C<Letter_Number>s.)
379This set also includes its subsets C<PosixUpper> and C<PosixLower>, both
e2cfb18c 380of which under C</i> match C<PosixAlpha>.
56ca34ca
KW
381
382For more details on Unicode properties, see L<perlunicode/Unicode
383Character Properties>; for a
e1b711da 384complete list of possible properties, see
56ca34ca
KW
385L<perluniprops/Properties accessible through \p{} and \P{}>,
386which notes all forms that have C</i> differences.
e1b711da 387It is also possible to define your own properties. This is discussed in
8a118206
RGS
388L<perlunicode/User-Defined Character Properties>.
389
94b42e47 390Unicode properties are defined (surprise!) only on Unicode code points.
2d88a86a
KW
391Starting in v5.20, when matching against C<\p> and C<\P>, Perl treats
392non-Unicode code points (those above the legal Unicode maximum of
3930x10FFFF) as if they were typical unassigned Unicode code points.
94b42e47 394
2d88a86a
KW
395Prior to v5.20, Perl raised a warning and made all matches fail on
396non-Unicode code points. This could be somewhat surprising:
94b42e47 397
2d88a86a
KW
398 chr(0x110000) =~ \p{ASCII_Hex_Digit=True} # Fails on Perls < v5.20.
399 chr(0x110000) =~ \p{ASCII_Hex_Digit=False} # Also fails on Perls
400 # < v5.20
401
402Even though these two matches might be thought of as complements, until
403v5.20 they were so only on Unicode code points.
94b42e47 404
8a118206
RGS
405=head4 Examples
406
407 "a" =~ /\w/ # Match, "a" is a 'word' character.
408 "7" =~ /\w/ # Match, "7" is a 'word' character as well.
409 "a" =~ /\d/ # No match, "a" isn't a digit.
410 "7" =~ /\d/ # Match, "7" is a digit.
ea449505 411 " " =~ /\s/ # Match, a space is whitespace.
8a118206
RGS
412 "a" =~ /\D/ # Match, "a" is a non-digit.
413 "7" =~ /\D/ # No match, "7" is not a non-digit.
ea449505 414 " " =~ /\S/ # No match, a space is not non-whitespace.
8a118206 415
ea449505
KW
416 " " =~ /\h/ # Match, space is horizontal whitespace.
417 " " =~ /\v/ # No match, space is not vertical whitespace.
418 "\r" =~ /\v/ # Match, a return is vertical whitespace.
8a118206
RGS
419
420 "a" =~ /\pL/ # Match, "a" is a letter.
421 "a" =~ /\p{Lu}/ # No match, /\p{Lu}/ matches upper case letters.
422
423 "\x{0e0b}" =~ /\p{Thai}/ # Match, \x{0e0b} is the character
424 # 'THAI CHARACTER SO SO', and that's in
425 # Thai Unicode class.
ea449505 426 "a" =~ /\P{Lao}/ # Match, as "a" is not a Laotian character.
8a118206 427
82206b5e
KW
428It is worth emphasizing that C<\d>, C<\w>, etc, match single characters, not
429complete numbers or words. To match a number (that consists of digits),
430use C<\d+>; to match a word, use C<\w+>. But be aware of the security
431considerations in doing so, as mentioned above.
8a118206
RGS
432
433=head2 Bracketed Character Classes
434
435The third form of character class you can use in Perl regular expressions
6b83a163 436is the bracketed character class. In its simplest form, it lists the characters
c1c4ae3a 437that may be matched, surrounded by square brackets, like this: C<[aeiou]>.
ea449505 438This matches one of C<a>, C<e>, C<i>, C<o> or C<u>. Like the other
1f59b283 439character classes, exactly one character is matched.* To match
ea449505 440a longer string consisting of characters mentioned in the character
6b83a163 441class, follow the character class with a L<quantifier|perlre/Quantifiers>. For
b6538e4f 442instance, C<[aeiou]+> matches one or more lowercase English vowels.
8a118206
RGS
443
444Repeating a character in a character class has no
445effect; it's considered to be in the set only once.
446
447Examples:
448
449 "e" =~ /[aeiou]/ # Match, as "e" is listed in the class.
450 "p" =~ /[aeiou]/ # No match, "p" is not listed in the class.
451 "ae" =~ /^[aeiou]$/ # No match, a character class only matches
452 # a single character.
453 "ae" =~ /^[aeiou]+$/ # Match, due to the quantifier.
454
1f59b283
KW
455 -------
456
8f0cd35a
KW
457* There are two exceptions to a bracketed character class matching a
458single character only. Each requires special handling by Perl to make
459things work:
460
461=over
462
463=item *
464
465When the class is to match caselessly under C</i> matching rules, and a
466character that is explicitly mentioned inside the class matches a
1f59b283 467multiple-character sequence caselessly under Unicode rules, the class
8f0cd35a
KW
468will also match that sequence. For example, Unicode says that the
469letter C<LATIN SMALL LETTER SHARP S> should match the sequence C<ss>
470under C</i> rules. Thus,
1f59b283
KW
471
472 'ss' =~ /\A\N{LATIN SMALL LETTER SHARP S}\z/i # Matches
473 'ss' =~ /\A[aeioust\N{LATIN SMALL LETTER SHARP S}]\z/i # Matches
474
8f0cd35a
KW
475For this to happen, the class must not be inverted (see L</Negation>)
476and the character must be explicitly specified, and not be part of a
477multi-character range (not even as one of its endpoints). (L</Character
478Ranges> will be explained shortly.) Therefore,
9d53c457 479
eb9e3b14
KW
480 'ss' =~ /\A[\0-\x{ff}]\z/ui # Doesn't match
481 'ss' =~ /\A[\0-\N{LATIN SMALL LETTER SHARP S}]\z/ui # No match
482 'ss' =~ /\A[\xDF-\xDF]\z/ui # Matches on ASCII platforms, since
a845303d 483 # \xDF is LATIN SMALL LETTER SHARP S,
8f0cd35a
KW
484 # and the range is just a single
485 # element
9d53c457
KW
486
487Note that it isn't a good idea to specify these types of ranges anyway.
488
8f0cd35a
KW
489=item *
490
491Some names known to C<\N{...}> refer to a sequence of multiple characters,
492instead of the usual single character. When one of these is included in
493the class, the entire sequence is matched. For example,
494
495 "\N{TAMIL LETTER KA}\N{TAMIL VOWEL SIGN AU}"
496 =~ / ^ [\N{TAMIL SYLLABLE KAU}] $ /x;
497
498matches, because C<\N{TAMIL SYLLABLE KAU}> is a named sequence
499consisting of the two characters matched against. Like the other
eb9e3b14 500instance where a bracketed class can match multiple characters, and for
8f0cd35a
KW
501similar reasons, the class must not be inverted, and the named sequence
502may not appear in a range, even one where it is both endpoints. If
4a88d526
KW
503these happen, it is a fatal error if the character class is within the
504scope of L<C<use re 'strict>|re/'strict' mode>, or within an extended
505L<C<(?[...])>|/Extended Bracketed Character Classes> class; otherwise
506only the first code point is used (with a C<regexp>-type warning
507raised).
8f0cd35a
KW
508
509=back
510
8a118206
RGS
511=head3 Special Characters Inside a Bracketed Character Class
512
513Most characters that are meta characters in regular expressions (that
df225385 514is, characters that carry a special meaning like C<.>, C<*>, or C<(>) lose
8a118206
RGS
515their special meaning and can be used inside a character class without
516the need to escape them. For instance, C<[()]> matches either an opening
517parenthesis, or a closing parenthesis, and the parens inside the character
6e16fd37
KW
518class don't group or capture. Be aware that, unless the pattern is
519evaluated in single-quotish context, variable interpolation will take
520place before the bracketed class is parsed:
521
522 $, = "\t| ";
523 $a =~ m'[$,]'; # single-quotish: matches '$' or ','
524 $a =~ q{[$,]}' # same
525 $a =~ m/[$,]/; # double-quotish: matches "\t", "|", or " "
8a118206
RGS
526
527Characters that may carry a special meaning inside a character class are:
528C<\>, C<^>, C<->, C<[> and C<]>, and are discussed below. They can be
529escaped with a backslash, although this is sometimes not needed, in which
530case the backslash may be omitted.
531
532The sequence C<\b> is special inside a bracketed character class. While
6b83a163 533outside the character class, C<\b> is an assertion indicating a point
8a118206
RGS
534that does not have either two word characters or two non-word characters
535on either side, inside a bracketed character class, C<\b> matches a
536backspace character.
537
df225385
KW
538The sequences
539C<\a>,
540C<\c>,
541C<\e>,
542C<\f>,
543C<\n>,
e526e8bb 544C<\N{I<NAME>}>,
765fa144 545C<\N{U+I<hex char>}>,
df225385
KW
546C<\r>,
547C<\t>,
548and
549C<\x>
06ee63cd 550are also special and have the same meanings as they do outside a
eb9e3b14 551bracketed character class.
df225385 552
ea449505
KW
553Also, a backslash followed by two or three octal digits is considered an octal
554number.
df225385 555
6b83a163
KW
556A C<[> is not special inside a character class, unless it's the start of a
557POSIX character class (see L</POSIX Character Classes> below). It normally does
558not need escaping.
8a118206 559
6b83a163
KW
560A C<]> is normally either the end of a POSIX character class (see
561L</POSIX Character Classes> below), or it signals the end of the bracketed
562character class. If you want to include a C<]> in the set of characters, you
563must generally escape it.
b6538e4f 564
8a118206
RGS
565However, if the C<]> is the I<first> (or the second if the first
566character is a caret) character of a bracketed character class, it
567does not denote the end of the class (as you cannot have an empty class)
568and is considered part of the set of characters that can be matched without
569escaping.
570
571Examples:
572
573 "+" =~ /[+?*]/ # Match, "+" in a character class is not special.
090752cc 574 "\cH" =~ /[\b]/ # Match, \b inside in a character class
c1c4ae3a 575 # is equivalent to a backspace.
090752cc 576 "]" =~ /[][]/ # Match, as the character class contains
8a118206
RGS
577 # both [ and ].
578 "[]" =~ /[[]]/ # Match, the pattern contains a character class
52f4d632 579 # containing just [, and the character class is
8a118206
RGS
580 # followed by a ].
581
77c8f263
KW
582=head3 Bracketed Character Classes and the C</xx> pattern modifier
583
584Normally SPACE and TAB characters have no special meaning inside a
585bracketed character class; they are just added to the list of characters
586matched by the class. But if the L<C</xx>|perlre/E<sol>x and E<sol>xx>
587pattern modifier is in effect, they are generally ignored and can be
588added to improve readability. They can't be added in the middle of a
589single construct:
590
591 / [ \x{10 FFFF} ] /xx # WRONG!
592
593The SPACE in the middle of the hex constant is illegal.
594
595To specify a literal SPACE character, you can escape it with a
596backslash, like:
597
598 /[ a e i o u \ ]/xx
599
600This matches the English vowels plus the SPACE character.
601
602For clarity, you should already have been using C<\t> to specify a
603literal tab, and C<\t> is unaffected by C</xx>.
604
8a118206
RGS
605=head3 Character Ranges
606
607It is not uncommon to want to match a range of characters. Luckily, instead
b6538e4f 608of listing all characters in the range, one may use the hyphen (C<->).
8a118206 609If inside a bracketed character class you have two characters separated
b6538e4f 610by a hyphen, it's treated as if all characters between the two were in
8a118206 611the class. For instance, C<[0-9]> matches any ASCII digit, and C<[a-m]>
e2cfb18c 612matches any lowercase letter from the first half of the ASCII alphabet.
8a118206
RGS
613
614Note that the two characters on either side of the hyphen are not
765fa144 615necessarily both letters or both digits. Any character is possible,
8a118206 616although not advisable. C<['-?]> contains a range of characters, but
b6538e4f 617most people will not know which characters that means. Furthermore,
8a118206
RGS
618such ranges may lead to portability problems if the code has to run on
619a platform that uses a different character set, such as EBCDIC.
620
ea449505
KW
621If a hyphen in a character class cannot syntactically be part of a range, for
622instance because it is the first or the last character of the character class,
b6538e4f
TC
623or if it immediately follows a range, the hyphen isn't special, and so is
624considered a character to be matched literally. If you want a hyphen in
625your set of characters to be matched and its position in the class is such
626that it could be considered part of a range, you must escape that hyphen
627with a backslash.
8a118206
RGS
628
629Examples:
630
631 [a-z] # Matches a character that is a lower case ASCII letter.
c1c4ae3a
KW
632 [a-fz] # Matches any letter between 'a' and 'f' (inclusive) or
633 # the letter 'z'.
8a118206
RGS
634 [-z] # Matches either a hyphen ('-') or the letter 'z'.
635 [a-f-m] # Matches any letter between 'a' and 'f' (inclusive), the
636 # hyphen ('-'), or the letter 'm'.
637 ['-?] # Matches any of the characters '()*+,-./0123456789:;<=>?
638 # (But not on an EBCDIC platform).
c7d25594
KW
639 [\N{APOSTROPHE}-\N{QUESTION MARK}]
640 # Matches any of the characters '()*+,-./0123456789:;<=>?
641 # even on an EBCDIC platform.
ad63362f 642 [\N{U+27}-\N{U+3F}] # Same. (U+27 is "'", and U+3F is "?")
c7d25594
KW
643
644As the final two examples above show, you can achieve portablity to
645non-ASCII platforms by using the C<\N{...}> form for the range
646endpoints. These indicate that the specified range is to be interpreted
647using Unicode values, so C<[\N{U+27}-\N{U+3F}]> means to match
648C<\N{U+27}>, C<\N{U+28}>, C<\N{U+29}>, ..., C<\N{U+3D}>, C<\N{U+3E}>,
649and C<\N{U+3F}>, whatever the native code point versions for those are.
b927b7e9
KW
650These are called "Unicode" ranges. If either end is of the C<\N{...}>
651form, the range is considered Unicode. A C<regexp> warning is raised
652under C<S<"use re 'strict'">> if the other endpoint is specified
653non-portably:
654
655 [\N{U+00}-\x09] # Warning under re 'strict'; \x09 is non-portable
656 [\N{U+00}-\t] # No warning;
657
658Both of the above match the characters C<\N{U+00}> C<\N{U+01}>, ...
659C<\N{U+08}>, C<\N{U+09}>, but the C<\x09> looks like it could be a
660mistake so the warning is raised (under C<re 'strict'>) for it.
c7d25594
KW
661
662Perl also guarantees that the ranges C<A-Z>, C<a-z>, C<0-9>, and any
09e43397 663subranges of these match what an English-only speaker would expect them
c7d25594
KW
664to match on any platform. That is, C<[A-Z]> matches the 26 ASCII
665uppercase letters;
09e43397
KW
666C<[a-z]> matches the 26 lowercase letters; and C<[0-9]> matches the 10
667digits. Subranges, like C<[h-k]>, match correspondingly, in this case
668just the four letters C<"h">, C<"i">, C<"j">, and C<"k">. This is the
669natural behavior on ASCII platforms where the code points (ordinal
670values) for C<"h"> through C<"k"> are consecutive integers (0x68 through
6710x6B). But special handling to achieve this may be needed on platforms
672with a non-ASCII native character set. For example, on EBCDIC
673platforms, the code point for C<"h"> is 0x88, C<"i"> is 0x89, C<"j"> is
6740x91, and C<"k"> is 0x92. Perl specially treats C<[h-k]> to exclude the
675seven code points in the gap: 0x8A through 0x90. This special handling is
676only invoked when the range is a subrange of one of the ASCII uppercase,
677lowercase, and digit ranges, AND each end of the range is expressed
678either as a literal, like C<"A">, or as a named character (C<\N{...}>,
679including the C<\N{U+...> form).
680
681EBCDIC Examples:
682
683 [i-j] # Matches either "i" or "j"
684 [i-\N{LATIN SMALL LETTER J}] # Same
685 [i-\N{U+6A}] # Same
686 [\N{U+69}-\N{U+6A}] # Same
687 [\x{89}-\x{91}] # Matches 0x89 ("i"), 0x8A .. 0x90, 0x91 ("j")
688 [i-\x{91}] # Same
689 [\x{89}-j] # Same
690 [i-J] # Matches, 0x89 ("i") .. 0xC1 ("J"); special
691 # handling doesn't apply because range is mixed
692 # case
8a118206
RGS
693
694=head3 Negation
695
696It is also possible to instead list the characters you do not want to
697match. You can do so by using a caret (C<^>) as the first character in the
b6538e4f 698character class. For instance, C<[^a-z]> matches any character that is not a
e2cfb18c
KW
699lowercase ASCII letter, which therefore includes more than a million
700Unicode code points. The class is said to be "negated" or "inverted".
8a118206
RGS
701
702This syntax make the caret a special character inside a bracketed character
703class, but only if it is the first character of the class. So if you want
82206b5e 704the caret as one of the characters to match, either escape the caret or
e2cfb18c 705else don't list it first.
8a118206 706
1f59b283 707In inverted bracketed character classes, Perl ignores the Unicode rules
8f0cd35a
KW
708that normally say that named sequence, and certain characters should
709match a sequence of multiple characters use under caseless C</i>
710matching. Following those rules could lead to highly confusing
711situations:
1f59b283 712
582da942 713 "ss" =~ /^[^\xDF]+$/ui; # Matches!
1f59b283
KW
714
715This should match any sequences of characters that aren't C<\xDF> nor
716what C<\xDF> matches under C</i>. C<"s"> isn't C<\xDF>, but Unicode
717says that C<"ss"> is what C<\xDF> matches under C</i>. So which one
718"wins"? Do you fail the match because the string has C<ss> or accept it
582da942 719because it has an C<s> followed by another C<s>? Perl has chosen the
8f0cd35a 720latter. (See note in L</Bracketed Character Classes> above.)
1f59b283 721
8a118206
RGS
722Examples:
723
724 "e" =~ /[^aeiou]/ # No match, the 'e' is listed.
725 "x" =~ /[^aeiou]/ # Match, as 'x' isn't a lowercase vowel.
726 "^" =~ /[^^]/ # No match, matches anything that isn't a caret.
727 "^" =~ /[x^]/ # Match, caret is not special here.
728
729=head3 Backslash Sequences
730
ea449505 731You can put any backslash sequence character class (with the exception of
765fa144 732C<\N> and C<\R>) inside a bracketed character class, and it will act just
b6538e4f
TC
733as if you had put all characters matched by the backslash sequence inside the
734character class. For instance, C<[a-f\d]> matches any decimal digit, or any
6b83a163
KW
735of the lowercase letters between 'a' and 'f' inclusive.
736
737C<\N> within a bracketed character class must be of the forms C<\N{I<name>}>
765fa144 738or C<\N{U+I<hex char>}>, and NOT be the form that matches non-newlines,
6b83a163
KW
739for the same reason that a dot C<.> inside a bracketed character class loses
740its special meaning: it matches nearly anything, which generally isn't what you
741want to happen.
df225385 742
8a118206
RGS
743
744Examples:
745
746 /[\p{Thai}\d]/ # Matches a character that is either a Thai
747 # character, or a digit.
748 /[^\p{Arabic}()]/ # Matches a character that is neither an Arabic
749 # character, nor a parenthesis.
750
751Backslash sequence character classes cannot form one of the endpoints
6b83a163
KW
752of a range. Thus, you can't say:
753
754 /[\p{Thai}-\d]/ # Wrong!
8a118206 755
6b83a163 756=head3 POSIX Character Classes
ea449505 757X<character class> X<\p> X<\p{}>
ea449505
KW
758X<alpha> X<alnum> X<ascii> X<blank> X<cntrl> X<digit> X<graph>
759X<lower> X<print> X<punct> X<space> X<upper> X<word> X<xdigit>
8a118206 760
d66e1f56 761POSIX character classes have the form C<[:class:]>, where I<class> is the
6b83a163 762name, and the C<[:> and C<:]> delimiters. POSIX character classes only appear
8a118206 763I<inside> bracketed character classes, and are a convenient and descriptive
82206b5e 764way of listing a group of characters.
6b83a163
KW
765
766Be careful about the syntax,
8a118206
RGS
767
768 # Correct:
769 $string =~ /[[:alpha:]]/
770
771 # Incorrect (will warn):
772 $string =~ /[:alpha:]/
773
774The latter pattern would be a character class consisting of a colon,
775and the letters C<a>, C<l>, C<p> and C<h>.
d66e1f56 776
82206b5e 777POSIX character classes can be part of a larger bracketed character class.
b6538e4f 778For example,
ea449505
KW
779
780 [01[:alpha:]%]
781
782is valid and matches '0', '1', any alphabetic character, and the percent sign.
8a118206
RGS
783
784Perl recognizes the following POSIX character classes:
785
ea449505 786 alpha Any alphabetical character ("[A-Za-z]").
48cbae4f 787 alnum Any alphanumeric character ("[A-Za-z0-9]").
ea449505 788 ascii Any character in the ASCII character set.
ea8b8ad2 789 blank A GNU extension, equal to a space or a horizontal tab ("\t").
ea449505
KW
790 cntrl Any control character. See Note [2] below.
791 digit Any decimal digit ("[0-9]"), equivalent to "\d".
792 graph Any printable character, excluding a space. See Note [3] below.
793 lower Any lowercase character ("[a-z]").
794 print Any printable character, including a space. See Note [4] below.
c1c4ae3a 795 punct Any graphical character excluding "word" characters. Note [5].
d28d8023
KW
796 space Any whitespace character. "\s" including the vertical tab
797 ("\cK").
ea449505
KW
798 upper Any uppercase character ("[A-Z]").
799 word A Perl extension ("[A-Za-z0-9_]"), equivalent to "\w".
800 xdigit Any hexadecimal digit ("[0-9a-fA-F]").
801
93106464
KW
802Like the L<Unicode properties|/Unicode Properties>, most of the POSIX
803properties match the same regardless of whether case-insensitive (C</i>)
804matching is in effect or not. The two exceptions are C<[:upper:]> and
805C<[:lower:]>. Under C</i>, they each match the union of C<[:upper:]> and
806C<[:lower:]>.
807
ea449505
KW
808Most POSIX character classes have two Unicode-style C<\p> property
809counterparts. (They are not official Unicode properties, but Perl extensions
810derived from official Unicode properties.) The table below shows the relation
811between POSIX character classes and these counterparts.
812
813One counterpart, in the column labelled "ASCII-range Unicode" in
b6538e4f 814the table, matches only characters in the ASCII character set.
ea449505
KW
815
816The other counterpart, in the column labelled "Full-range Unicode", matches any
817appropriate characters in the full Unicode character set. For example,
b6538e4f 818C<\p{Alpha}> matches not just the ASCII alphabetic characters, but any
82206b5e 819character in the entire Unicode character set considered alphabetic.
582da942 820An entry in the column labelled "backslash sequence" is a (short)
5db9882c 821equivalent.
ea449505 822
cbc24f92
KW
823 [[:...:]] ASCII-range Full-range backslash Note
824 Unicode Unicode sequence
ea449505 825 -----------------------------------------------------
cbc24f92
KW
826 alpha \p{PosixAlpha} \p{XPosixAlpha}
827 alnum \p{PosixAlnum} \p{XPosixAlnum}
82206b5e 828 ascii \p{ASCII}
cbc24f92
KW
829 blank \p{PosixBlank} \p{XPosixBlank} \h [1]
830 or \p{HorizSpace} [1]
831 cntrl \p{PosixCntrl} \p{XPosixCntrl} [2]
832 digit \p{PosixDigit} \p{XPosixDigit} \d
833 graph \p{PosixGraph} \p{XPosixGraph} [3]
834 lower \p{PosixLower} \p{XPosixLower}
835 print \p{PosixPrint} \p{XPosixPrint} [4]
836 punct \p{PosixPunct} \p{XPosixPunct} [5]
837 \p{PerlSpace} \p{XPerlSpace} \s [6]
838 space \p{PosixSpace} \p{XPosixSpace} [6]
839 upper \p{PosixUpper} \p{XPosixUpper}
840 word \p{PosixWord} \p{XPosixWord} \w
82206b5e 841 xdigit \p{PosixXDigit} \p{XPosixXDigit}
8a118206
RGS
842
843=over 4
844
ea449505
KW
845=item [1]
846
847C<\p{Blank}> and C<\p{HorizSpace}> are synonyms.
848
849=item [2]
8a118206 850
ea449505 851Control characters don't produce output as such, but instead usually control
b6538e4f 852the terminal somehow: for example, newline and backspace are control characters.
93106464
KW
853On ASCII platforms, in the ASCII range, characters whose code points are
854between 0 and 31 inclusive, plus 127 (C<DEL>) are control characters; on
855EBCDIC platforms, their counterparts are control characters.
8a118206 856
ea449505 857=item [3]
8a118206
RGS
858
859Any character that is I<graphical>, that is, visible. This class consists
b6538e4f 860of all alphanumeric characters and all punctuation characters.
8a118206 861
ea449505 862=item [4]
8a118206 863
b6538e4f
TC
864All printable characters, which is the set of all graphical characters
865plus those whitespace characters which are not also controls.
ea449505 866
b6dac59a 867=item [5]
ea449505 868
b6538e4f 869C<\p{PosixPunct}> and C<[[:punct:]]> in the ASCII range match all
ea449505
KW
870non-controls, non-alphanumeric, non-space characters:
871C<[-!"#$%&'()*+,./:;<=E<gt>?@[\\\]^_`{|}~]> (although if a locale is in effect,
872it could alter the behavior of C<[[:punct:]]>).
873
cbc24f92
KW
874The similarly named property, C<\p{Punct}>, matches a somewhat different
875set in the ASCII range, namely
0be9b861
KW
876C<[-!"#%&'()*,./:;?@[\\\]_{}]>. That is, it is missing the nine
877characters C<[$+E<lt>=E<gt>^`|~]>.
6c5a041f
KW
878This is because Unicode splits what POSIX considers to be punctuation into two
879categories, Punctuation and Symbols.
880
e2cfb18c 881C<\p{XPosixPunct}> and (under Unicode rules) C<[[:punct:]]>, match what
765fa144
KW
882C<\p{PosixPunct}> matches in the ASCII range, plus what C<\p{Punct}>
883matches. This is different than strictly matching according to
884C<\p{Punct}>. Another way to say it is that
82206b5e
KW
885if Unicode rules are in effect, C<[[:punct:]]> matches all characters
886that Unicode considers punctuation, plus all ASCII-range characters that
887Unicode considers symbols.
8a118206 888
ea449505 889=item [6]
8a118206 890
7fa2fdc0 891C<\p{XPerlSpace}> and C<\p{Space}> match identically starting with Perl
d28d8023 892v5.18. In earlier versions, these differ only in that in non-locale
779cf272 893matching, C<\p{XPerlSpace}> did not match the vertical tab, C<\cK>.
d28d8023 894Same for the two ASCII-only range forms.
8a118206
RGS
895
896=back
897
ab6199be 898There are various other synonyms that can be used besides the names
4cb26c52 899listed in the table. For example, C<\p{XPosixAlpha}> can be written as
ab6199be 900C<\p{Alpha}>. All are listed in
d66e1f56 901L<perluniprops/Properties accessible through \p{} and \P{}>.
ab6199be
KW
902
903Both the C<\p> counterparts always assume Unicode rules are in effect.
904On ASCII platforms, this means they assume that the code points from 128
905to 255 are Latin-1, and that means that using them under locale rules is
906unwise unless the locale is guaranteed to be Latin-1 or UTF-8. In contrast, the
907POSIX character classes are useful under locale rules. They are
908affected by the actual rules in effect, as follows:
909
910=over
911
912=item If the C</a> modifier, is in effect ...
913
914Each of the POSIX classes matches exactly the same as their ASCII-range
915counterparts.
916
917=item otherwise ...
918
919=over
920
921=item For code points above 255 ...
922
923The POSIX class matches the same as its Full-range counterpart.
924
925=item For code points below 256 ...
926
927=over
928
929=item if locale rules are in effect ...
930
a145a423
KW
931The POSIX class matches according to the locale, except:
932
933=over
934
935=item C<word>
936
937also includes the platform's native underscore character, no matter what
8129baca 938the locale is.
ab6199be 939
a145a423
KW
940=item C<ascii>
941
942on platforms that don't have the POSIX C<ascii> extension, this matches
943just the platform's native ASCII-range characters.
944
945=item C<blank>
946
947on platforms that don't have the POSIX C<blank> extension, this matches
948just the platform's native tab and space characters.
949
950=back
951
4a88d526 952=item if instead, Unicode rules are in effect ...
ab6199be
KW
953
954The POSIX class matches the same as the Full-range counterpart.
955
956=item otherwise ...
957
958The POSIX class matches the same as the ASCII range counterpart.
959
960=back
961
962=back
963
964=back
965
966Which rules apply are determined as described in
967L<perlre/Which character set modifier is in effect?>.
968
969It is proposed to change this behavior in a future release of Perl so that
970whether or not Unicode rules are in effect would not change the
4b9734bf 971behavior: Outside of locale, the POSIX classes
ab6199be
KW
972would behave like their ASCII-range counterparts. If you wish to
973comment on this proposal, send email to C<perl5-porters@perl.org>.
cbc24f92 974
1f59b283 975=head4 Negation of POSIX character classes
ea449505 976X<character class, negation>
8a118206
RGS
977
978A Perl extension to the POSIX character class is the ability to
979negate it. This is done by prefixing the class name with a caret (C<^>).
980Some examples:
981
ea449505
KW
982 POSIX ASCII-range Full-range backslash
983 Unicode Unicode sequence
984 -----------------------------------------------------
cbc24f92
KW
985 [[:^digit:]] \P{PosixDigit} \P{XPosixDigit} \D
986 [[:^space:]] \P{PosixSpace} \P{XPosixSpace}
987 \P{PerlSpace} \P{XPerlSpace} \S
988 [[:^word:]] \P{PerlWord} \P{XPosixWord} \W
989
765fa144 990The backslash sequence can mean either ASCII- or Full-range Unicode,
82206b5e 991depending on various factors as described in L<perlre/Which character set modifier is in effect?>.
8a118206
RGS
992
993=head4 [= =] and [. .]
994
b6538e4f 995Perl recognizes the POSIX character classes C<[=class=]> and
82206b5e 996C<[.class.]>, but does not (yet?) support them. Any attempt to use
b6538e4f 997either construct raises an exception.
8a118206
RGS
998
999=head4 Examples
1000
1001 /[[:digit:]]/ # Matches a character that is a digit.
1002 /[01[:lower:]]/ # Matches a character that is either a
1003 # lowercase letter, or '0' or '1'.
c1c4ae3a 1004 /[[:digit:][:^xdigit:]]/ # Matches a character that can be anything
bc943be5
KW
1005 # except the letters 'a' to 'f' and 'A' to
1006 # 'F'. This is because the main character
1007 # class is composed of two POSIX character
1008 # classes that are ORed together, one that
1009 # matches any digit, and the other that
1010 # matches anything that isn't a hex digit.
1011 # The OR adds the digits, leaving only the
1012 # letters 'a' to 'f' and 'A' to 'F' excluded.
572224ce
KW
1013
1014=head3 Extended Bracketed Character Classes
1015X<character class>
1016X<set operations>
1017
1018This is a fancy bracketed character class that can be used for more
1019readable and less error-prone classes, and to perform set operations,
1020such as intersection. An example is
1021
1022 /(?[ \p{Thai} & \p{Digit} ])/
1023
1024This will match all the digit characters that are in the Thai script.
1025
1026This is an experimental feature available starting in 5.18, and is
1027subject to change as we gain field experience with it. Any attempt to
1028use it will raise a warning, unless disabled via
1029
1030 no warnings "experimental::regex_sets";
1031
1032Comments on this feature are welcome; send email to
1033C<perl5-porters@perl.org>.
1034
a60b7922
KW
1035The rules used by L<C<use re 'strict>|re/'strict' mode> apply to this
1036construct.
1037
572224ce
KW
1038We can extend the example above:
1039
1040 /(?[ ( \p{Thai} + \p{Lao} ) & \p{Digit} ])/
1041
1042This matches digits that are in either the Thai or Laotian scripts.
1043
1044Notice the white space in these examples. This construct always has
77c8f263 1045the C<E<sol>xx> modifier turned on within it.
572224ce
KW
1046
1047The available binary operators are:
1048
1049 & intersection
1050 + union
1051 | another name for '+', hence means union
1052 - subtraction (the result matches the set consisting of those
1053 code points matched by the first operand, excluding any that
1054 are also matched by the second operand)
1055 ^ symmetric difference (the union minus the intersection). This
1056 is like an exclusive or, in that the result is the set of code
1057 points that are matched by either, but not both, of the
1058 operands.
1059
1060There is one unary operator:
1061
1062 ! complement
1063
6798c95d
KW
1064All the binary operators left associate; C<"&"> is higher precedence
1065than the others, which all have equal precedence. The unary operator
1066right associates, and has highest precedence. Thus this follows the
1067normal Perl precedence rules for logical operators. Use parentheses to
1068override the default precedence and associativity.
572224ce
KW
1069
1070The main restriction is that everything is a metacharacter. Thus,
1071you cannot refer to single characters by doing something like this:
1072
1073 /(?[ a + b ])/ # Syntax error!
1074
1075The easiest way to specify an individual typable character is to enclose
1076it in brackets:
1077
1078 /(?[ [a] + [b] ])/
1079
1080(This is the same thing as C<[ab]>.) You could also have said the
1081equivalent:
1082
1083 /(?[[ a b ]])/
1084
de36fb2e
KW
1085(You can, of course, specify single characters by using, C<\x{...}>,
1086C<\N{...}>, etc.)
572224ce
KW
1087
1088This last example shows the use of this construct to specify an ordinary
1089bracketed character class without additional set operations. Note the
77c8f263
KW
1090white space within it. This is allowed because C<E<sol>xx> is
1091automatically turned on within this construct.
572224ce 1092
572224ce 1093All the other escapes accepted by normal bracketed character classes are
7d4c055d
KW
1094accepted here as well.
1095
1096Because this construct compiles under
1097L<C<use re 'strict>|re/'strict' mode>, unrecognized escapes that
1098generate warnings in normal classes are fatal errors here, as well as
1099all other warnings from these class elements, as well as some
1100practices that don't currently warn outside C<re 'strict'>. For example
1101you cannot say
572224ce
KW
1102
1103 /(?[ [ \xF ] ])/ # Syntax error!
1104
1105You have to have two hex digits after a braceless C<\x> (use a leading
1106zero to make two). These restrictions are to lower the incidence of
1107typos causing the class to not match what you thought it would.
1108
f194034a
KW
1109If a regular bracketed character class contains a C<\p{}> or C<\P{}> and
1110is matched against a non-Unicode code point, a warning may be
1111raised, as the result is not Unicode-defined. No such warning will come
1112when using this extended form.
1113
572224ce
KW
1114The final difference between regular bracketed character classes and
1115these, is that it is not possible to get these to match a
1116multi-character fold. Thus,
1117
1118 /(?[ [\xDF] ])/iu
1119
1120does not match the string C<ss>.
1121
1122You don't have to enclose POSIX class names inside double brackets,
1123hence both of the following work:
1124
1125 /(?[ [:word:] - [:lower:] ])/
1126 /(?[ [[:word:]] - [[:lower:]] ])/
1127
1128Any contained POSIX character classes, including things like C<\w> and C<\D>
1129respect the C<E<sol>a> (and C<E<sol>aa>) modifiers.
1130
1131C<< (?[ ]) >> is a regex-compile-time construct. Any attempt to use
1132something which isn't knowable at the time the containing regular
1133expression is compiled is a fatal error. In practice, this means
11a9b3e0 1134just three limitations:
572224ce
KW
1135
1136=over 4
1137
1138=item 1
1139
a0bd1a30
KW
1140When compiled within the scope of C<use locale> (or the C<E<sol>l> regex
1141modifier), this construct assumes that the execution-time locale will be
1142a UTF-8 one, and the generated pattern always uses Unicode rules. What
1143gets matched or not thus isn't dependent on the actual runtime locale, so
1144tainting is not enabled. But a C<locale> category warning is raised
1145if the runtime locale turns out to not be UTF-8.
572224ce
KW
1146
1147=item 2
1148
1149Any
1150L<user-defined property|perlunicode/"User-Defined Character Properties">
1151used must be already defined by the time the regular expression is
1152compiled (but note that this construct can be used instead of such
1153properties).
1154
1155=item 3
1156
1157A regular expression that otherwise would compile
1158using C<E<sol>d> rules, and which uses this construct will instead
1159use C<E<sol>u>. Thus this construct tells Perl that you don't want
1160C<E<sol>d> rules for the entire regular expression containing it.
1161
1162=back
1163
572224ce
KW
1164Note that skipping white space applies only to the interior of this
1165construct. There must not be any space between any of the characters
1166that form the initial C<(?[>. Nor may there be space between the
1167closing C<])> characters.
1168
11a9b3e0 1169Just as in all regular expressions, the pattern can be built up by
572224ce
KW
1170including variables that are interpolated at regex compilation time.
1171Care must be taken to ensure that you are getting what you expect. For
1172example:
1173
1174 my $thai_or_lao = '\p{Thai} + \p{Lao}';
1175 ...
1176 qr/(?[ \p{Digit} & $thai_or_lao ])/;
1177
1178compiles to
1179
1180 qr/(?[ \p{Digit} & \p{Thai} + \p{Lao} ])/;
1181
1182But this does not have the effect that someone reading the code would
1183likely expect, as the intersection applies just to C<\p{Thai}>,
1184excluding the Laotian. Pitfalls like this can be avoided by
1185parenthesizing the component pieces:
1186
1187 my $thai_or_lao = '( \p{Thai} + \p{Lao} )';
1188
1189But any modifiers will still apply to all the components:
1190
1191 my $lower = '\p{Lower} + \p{Digit}';
1192 qr/(?[ \p{Greek} & $lower ])/i;
1193
1194matches upper case things. You can avoid surprises by making the
1195components into instances of this construct by compiling them:
1196
1197 my $thai_or_lao = qr/(?[ \p{Thai} + \p{Lao} ])/;
1198 my $lower = qr/(?[ \p{Lower} + \p{Digit} ])/;
1199
1200When these are embedded in another pattern, what they match does not
1201change, regardless of parenthesization or what modifiers are in effect
1202in that outer pattern.
1203
1204Due to the way that Perl parses things, your parentheses and brackets
1205may need to be balanced, even including comments. If you run into any
1206examples, please send them to C<perlbug@perl.org>, so that we can have a
1207concrete example for this man page.
1208
1209We may change it so that things that remain legal uses in normal bracketed
1210character classes might become illegal within this experimental
1211construct. One proposal, for example, is to forbid adjacent uses of the
1212same character, as in C<(?[ [aa] ])>. The motivation for such a change
1213is that this usage is likely a typo, as the second "a" adds nothing.