1<html> 2<head> 3<title>pcrepattern specification</title> 4</head> 5<body bgcolor="#FFFFFF" text="#00005A" link="#0066FF" alink="#3399FF" vlink="#2222BB"> 6<h1>pcrepattern man page</h1> 7<p> 8Return to the <a href="index.html">PCRE index page</a>. 9</p> 10<p> 11This page is part of the PCRE HTML documentation. It was generated automatically 12from the original man page. If there is any nonsense in it, please consult the 13man page, in case the conversion went wrong. 14<br> 15<ul> 16<li><a name="TOC1" href="#SEC1">PCRE REGULAR EXPRESSION DETAILS</a> 17<li><a name="TOC2" href="#SEC2">NEWLINE CONVENTIONS</a> 18<li><a name="TOC3" href="#SEC3">CHARACTERS AND METACHARACTERS</a> 19<li><a name="TOC4" href="#SEC4">BACKSLASH</a> 20<li><a name="TOC5" href="#SEC5">CIRCUMFLEX AND DOLLAR</a> 21<li><a name="TOC6" href="#SEC6">FULL STOP (PERIOD, DOT) AND \N</a> 22<li><a name="TOC7" href="#SEC7">MATCHING A SINGLE DATA UNIT</a> 23<li><a name="TOC8" href="#SEC8">SQUARE BRACKETS AND CHARACTER CLASSES</a> 24<li><a name="TOC9" href="#SEC9">POSIX CHARACTER CLASSES</a> 25<li><a name="TOC10" href="#SEC10">VERTICAL BAR</a> 26<li><a name="TOC11" href="#SEC11">INTERNAL OPTION SETTING</a> 27<li><a name="TOC12" href="#SEC12">SUBPATTERNS</a> 28<li><a name="TOC13" href="#SEC13">DUPLICATE SUBPATTERN NUMBERS</a> 29<li><a name="TOC14" href="#SEC14">NAMED SUBPATTERNS</a> 30<li><a name="TOC15" href="#SEC15">REPETITION</a> 31<li><a name="TOC16" href="#SEC16">ATOMIC GROUPING AND POSSESSIVE QUANTIFIERS</a> 32<li><a name="TOC17" href="#SEC17">BACK REFERENCES</a> 33<li><a name="TOC18" href="#SEC18">ASSERTIONS</a> 34<li><a name="TOC19" href="#SEC19">CONDITIONAL SUBPATTERNS</a> 35<li><a name="TOC20" href="#SEC20">COMMENTS</a> 36<li><a name="TOC21" href="#SEC21">RECURSIVE PATTERNS</a> 37<li><a name="TOC22" href="#SEC22">SUBPATTERNS AS SUBROUTINES</a> 38<li><a name="TOC23" href="#SEC23">ONIGURUMA SUBROUTINE SYNTAX</a> 39<li><a name="TOC24" href="#SEC24">CALLOUTS</a> 40<li><a name="TOC25" href="#SEC25">BACKTRACKING CONTROL</a> 41<li><a name="TOC26" href="#SEC26">SEE ALSO</a> 42<li><a name="TOC27" href="#SEC27">AUTHOR</a> 43<li><a name="TOC28" href="#SEC28">REVISION</a> 44</ul> 45<br><a name="SEC1" href="#TOC1">PCRE REGULAR EXPRESSION DETAILS</a><br> 46<P> 47The syntax and semantics of the regular expressions that are supported by PCRE 48are described in detail below. There is a quick-reference syntax summary in the 49<a href="pcresyntax.html"><b>pcresyntax</b></a> 50page. PCRE tries to match Perl syntax and semantics as closely as it can. PCRE 51also supports some alternative regular expression syntax (which does not 52conflict with the Perl syntax) in order to provide some compatibility with 53regular expressions in Python, .NET, and Oniguruma. 54</P> 55<P> 56Perl's regular expressions are described in its own documentation, and 57regular expressions in general are covered in a number of books, some of which 58have copious examples. Jeffrey Friedl's "Mastering Regular Expressions", 59published by O'Reilly, covers regular expressions in great detail. This 60description of PCRE's regular expressions is intended as reference material. 61</P> 62<P> 63The original operation of PCRE was on strings of one-byte characters. However, 64there is now also support for UTF-8 strings in the original library, and a 65second library that supports 16-bit and UTF-16 character strings. To use these 66features, PCRE must be built to include appropriate support. When using UTF 67strings you must either call the compiling function with the PCRE_UTF8 or 68PCRE_UTF16 option, or the pattern must start with one of these special 69sequences: 70<pre> 71 (*UTF8) 72 (*UTF16) 73</pre> 74Starting a pattern with such a sequence is equivalent to setting the relevant 75option. This feature is not Perl-compatible. How setting a UTF mode affects 76pattern matching is mentioned in several places below. There is also a summary 77of features in the 78<a href="pcreunicode.html"><b>pcreunicode</b></a> 79page. 80</P> 81<P> 82Another special sequence that may appear at the start of a pattern or in 83combination with (*UTF8) or (*UTF16) is: 84<pre> 85 (*UCP) 86</pre> 87This has the same effect as setting the PCRE_UCP option: it causes sequences 88such as \d and \w to use Unicode properties to determine character types, 89instead of recognizing only characters with codes less than 128 via a lookup 90table. 91</P> 92<P> 93If a pattern starts with (*NO_START_OPT), it has the same effect as setting the 94PCRE_NO_START_OPTIMIZE option either at compile or matching time. There are 95also some more of these special sequences that are concerned with the handling 96of newlines; they are described below. 97</P> 98<P> 99The remainder of this document discusses the patterns that are supported by 100PCRE when one its main matching functions, <b>pcre_exec()</b> (8-bit) or 101<b>pcre16_exec()</b> (16-bit), is used. PCRE also has alternative matching 102functions, <b>pcre_dfa_exec()</b> and <b>pcre16_dfa_exec()</b>, which match using 103a different algorithm that is not Perl-compatible. Some of the features 104discussed below are not available when DFA matching is used. The advantages and 105disadvantages of the alternative functions, and how they differ from the normal 106functions, are discussed in the 107<a href="pcrematching.html"><b>pcrematching</b></a> 108page. 109<a name="newlines"></a></P> 110<br><a name="SEC2" href="#TOC1">NEWLINE CONVENTIONS</a><br> 111<P> 112PCRE supports five different conventions for indicating line breaks in 113strings: a single CR (carriage return) character, a single LF (linefeed) 114character, the two-character sequence CRLF, any of the three preceding, or any 115Unicode newline sequence. The 116<a href="pcreapi.html"><b>pcreapi</b></a> 117page has 118<a href="pcreapi.html#newlines">further discussion</a> 119about newlines, and shows how to set the newline convention in the 120<i>options</i> arguments for the compiling and matching functions. 121</P> 122<P> 123It is also possible to specify a newline convention by starting a pattern 124string with one of the following five sequences: 125<pre> 126 (*CR) carriage return 127 (*LF) linefeed 128 (*CRLF) carriage return, followed by linefeed 129 (*ANYCRLF) any of the three above 130 (*ANY) all Unicode newline sequences 131</pre> 132These override the default and the options given to the compiling function. For 133example, on a Unix system where LF is the default newline sequence, the pattern 134<pre> 135 (*CR)a.b 136</pre> 137changes the convention to CR. That pattern matches "a\nb" because LF is no 138longer a newline. Note that these special settings, which are not 139Perl-compatible, are recognized only at the very start of a pattern, and that 140they must be in upper case. If more than one of them is present, the last one 141is used. 142</P> 143<P> 144The newline convention affects the interpretation of the dot metacharacter when 145PCRE_DOTALL is not set, and also the behaviour of \N. However, it does not 146affect what the \R escape sequence matches. By default, this is any Unicode 147newline sequence, for Perl compatibility. However, this can be changed; see the 148description of \R in the section entitled 149<a href="#newlineseq">"Newline sequences"</a> 150below. A change of \R setting can be combined with a change of newline 151convention. 152</P> 153<br><a name="SEC3" href="#TOC1">CHARACTERS AND METACHARACTERS</a><br> 154<P> 155A regular expression is a pattern that is matched against a subject string from 156left to right. Most characters stand for themselves in a pattern, and match the 157corresponding characters in the subject. As a trivial example, the pattern 158<pre> 159 The quick brown fox 160</pre> 161matches a portion of a subject string that is identical to itself. When 162caseless matching is specified (the PCRE_CASELESS option), letters are matched 163independently of case. In a UTF mode, PCRE always understands the concept of 164case for characters whose values are less than 128, so caseless matching is 165always possible. For characters with higher values, the concept of case is 166supported if PCRE is compiled with Unicode property support, but not otherwise. 167If you want to use caseless matching for characters 128 and above, you must 168ensure that PCRE is compiled with Unicode property support as well as with 169UTF support. 170</P> 171<P> 172The power of regular expressions comes from the ability to include alternatives 173and repetitions in the pattern. These are encoded in the pattern by the use of 174<i>metacharacters</i>, which do not stand for themselves but instead are 175interpreted in some special way. 176</P> 177<P> 178There are two different sets of metacharacters: those that are recognized 179anywhere in the pattern except within square brackets, and those that are 180recognized within square brackets. Outside square brackets, the metacharacters 181are as follows: 182<pre> 183 \ general escape character with several uses 184 ^ assert start of string (or line, in multiline mode) 185 $ assert end of string (or line, in multiline mode) 186 . match any character except newline (by default) 187 [ start character class definition 188 | start of alternative branch 189 ( start subpattern 190 ) end subpattern 191 ? extends the meaning of ( 192 also 0 or 1 quantifier 193 also quantifier minimizer 194 * 0 or more quantifier 195 + 1 or more quantifier 196 also "possessive quantifier" 197 { start min/max quantifier 198</pre> 199Part of a pattern that is in square brackets is called a "character class". In 200a character class the only metacharacters are: 201<pre> 202 \ general escape character 203 ^ negate the class, but only if the first character 204 - indicates character range 205 [ POSIX character class (only if followed by POSIX syntax) 206 ] terminates the character class 207</pre> 208The following sections describe the use of each of the metacharacters. 209</P> 210<br><a name="SEC4" href="#TOC1">BACKSLASH</a><br> 211<P> 212The backslash character has several uses. Firstly, if it is followed by a 213character that is not a number or a letter, it takes away any special meaning 214that character may have. This use of backslash as an escape character applies 215both inside and outside character classes. 216</P> 217<P> 218For example, if you want to match a * character, you write \* in the pattern. 219This escaping action applies whether or not the following character would 220otherwise be interpreted as a metacharacter, so it is always safe to precede a 221non-alphanumeric with backslash to specify that it stands for itself. In 222particular, if you want to match a backslash, you write \\. 223</P> 224<P> 225In a UTF mode, only ASCII numbers and letters have any special meaning after a 226backslash. All other characters (in particular, those whose codepoints are 227greater than 127) are treated as literals. 228</P> 229<P> 230If a pattern is compiled with the PCRE_EXTENDED option, white space in the 231pattern (other than in a character class) and characters between a # outside 232a character class and the next newline are ignored. An escaping backslash can 233be used to include a white space or # character as part of the pattern. 234</P> 235<P> 236If you want to remove the special meaning from a sequence of characters, you 237can do so by putting them between \Q and \E. This is different from Perl in 238that $ and @ are handled as literals in \Q...\E sequences in PCRE, whereas in 239Perl, $ and @ cause variable interpolation. Note the following examples: 240<pre> 241 Pattern PCRE matches Perl matches 242 243 \Qabc$xyz\E abc$xyz abc followed by the contents of $xyz 244 \Qabc\$xyz\E abc\$xyz abc\$xyz 245 \Qabc\E\$\Qxyz\E abc$xyz abc$xyz 246</pre> 247The \Q...\E sequence is recognized both inside and outside character classes. 248An isolated \E that is not preceded by \Q is ignored. If \Q is not followed 249by \E later in the pattern, the literal interpretation continues to the end of 250the pattern (that is, \E is assumed at the end). If the isolated \Q is inside 251a character class, this causes an error, because the character class is not 252terminated. 253<a name="digitsafterbackslash"></a></P> 254<br><b> 255Non-printing characters 256</b><br> 257<P> 258A second use of backslash provides a way of encoding non-printing characters 259in patterns in a visible manner. There is no restriction on the appearance of 260non-printing characters, apart from the binary zero that terminates a pattern, 261but when a pattern is being prepared by text editing, it is often easier to use 262one of the following escape sequences than the binary character it represents: 263<pre> 264 \a alarm, that is, the BEL character (hex 07) 265 \cx "control-x", where x is any ASCII character 266 \e escape (hex 1B) 267 \f form feed (hex 0C) 268 \n linefeed (hex 0A) 269 \r carriage return (hex 0D) 270 \t tab (hex 09) 271 \ddd character with octal code ddd, or back reference 272 \xhh character with hex code hh 273 \x{hhh..} character with hex code hhh.. (non-JavaScript mode) 274 \uhhhh character with hex code hhhh (JavaScript mode only) 275</pre> 276The precise effect of \cx is as follows: if x is a lower case letter, it 277is converted to upper case. Then bit 6 of the character (hex 40) is inverted. 278Thus \cz becomes hex 1A (z is 7A), but \c{ becomes hex 3B ({ is 7B), while 279\c; becomes hex 7B (; is 3B). If the byte following \c has a value greater 280than 127, a compile-time error occurs. This locks out non-ASCII characters in 281all modes. (When PCRE is compiled in EBCDIC mode, all byte values are valid. A 282lower case letter is converted to upper case, and then the 0xc0 bits are 283flipped.) 284</P> 285<P> 286By default, after \x, from zero to two hexadecimal digits are read (letters 287can be in upper or lower case). Any number of hexadecimal digits may appear 288between \x{ and }, but the character code is constrained as follows: 289<pre> 290 8-bit non-UTF mode less than 0x100 291 8-bit UTF-8 mode less than 0x10ffff and a valid codepoint 292 16-bit non-UTF mode less than 0x10000 293 16-bit UTF-16 mode less than 0x10ffff and a valid codepoint 294</pre> 295Invalid Unicode codepoints are the range 0xd800 to 0xdfff (the so-called 296"surrogate" codepoints). 297</P> 298<P> 299If characters other than hexadecimal digits appear between \x{ and }, or if 300there is no terminating }, this form of escape is not recognized. Instead, the 301initial \x will be interpreted as a basic hexadecimal escape, with no 302following digits, giving a character whose value is zero. 303</P> 304<P> 305If the PCRE_JAVASCRIPT_COMPAT option is set, the interpretation of \x is 306as just described only when it is followed by two hexadecimal digits. 307Otherwise, it matches a literal "x" character. In JavaScript mode, support for 308code points greater than 256 is provided by \u, which must be followed by 309four hexadecimal digits; otherwise it matches a literal "u" character. 310Character codes specified by \u in JavaScript mode are constrained in the same 311was as those specified by \x in non-JavaScript mode. 312</P> 313<P> 314Characters whose value is less than 256 can be defined by either of the two 315syntaxes for \x (or by \u in JavaScript mode). There is no difference in the 316way they are handled. For example, \xdc is exactly the same as \x{dc} (or 317\u00dc in JavaScript mode). 318</P> 319<P> 320After \0 up to two further octal digits are read. If there are fewer than two 321digits, just those that are present are used. Thus the sequence \0\x\07 322specifies two binary zeros followed by a BEL character (code value 7). Make 323sure you supply two digits after the initial zero if the pattern character that 324follows is itself an octal digit. 325</P> 326<P> 327The handling of a backslash followed by a digit other than 0 is complicated. 328Outside a character class, PCRE reads it and any following digits as a decimal 329number. If the number is less than 10, or if there have been at least that many 330previous capturing left parentheses in the expression, the entire sequence is 331taken as a <i>back reference</i>. A description of how this works is given 332<a href="#backreferences">later,</a> 333following the discussion of 334<a href="#subpattern">parenthesized subpatterns.</a> 335</P> 336<P> 337Inside a character class, or if the decimal number is greater than 9 and there 338have not been that many capturing subpatterns, PCRE re-reads up to three octal 339digits following the backslash, and uses them to generate a data character. Any 340subsequent digits stand for themselves. The value of the character is 341constrained in the same way as characters specified in hexadecimal. 342For example: 343<pre> 344 \040 is another way of writing a space 345 \40 is the same, provided there are fewer than 40 previous capturing subpatterns 346 \7 is always a back reference 347 \11 might be a back reference, or another way of writing a tab 348 \011 is always a tab 349 \0113 is a tab followed by the character "3" 350 \113 might be a back reference, otherwise the character with octal code 113 351 \377 might be a back reference, otherwise the value 255 (decimal) 352 \81 is either a back reference, or a binary zero followed by the two characters "8" and "1" 353</pre> 354Note that octal values of 100 or greater must not be introduced by a leading 355zero, because no more than three octal digits are ever read. 356</P> 357<P> 358All the sequences that define a single character value can be used both inside 359and outside character classes. In addition, inside a character class, \b is 360interpreted as the backspace character (hex 08). 361</P> 362<P> 363\N is not allowed in a character class. \B, \R, and \X are not special 364inside a character class. Like other unrecognized escape sequences, they are 365treated as the literal characters "B", "R", and "X" by default, but cause an 366error if the PCRE_EXTRA option is set. Outside a character class, these 367sequences have different meanings. 368</P> 369<br><b> 370Unsupported escape sequences 371</b><br> 372<P> 373In Perl, the sequences \l, \L, \u, and \U are recognized by its string 374handler and used to modify the case of following characters. By default, PCRE 375does not support these escape sequences. However, if the PCRE_JAVASCRIPT_COMPAT 376option is set, \U matches a "U" character, and \u can be used to define a 377character by code point, as described in the previous section. 378</P> 379<br><b> 380Absolute and relative back references 381</b><br> 382<P> 383The sequence \g followed by an unsigned or a negative number, optionally 384enclosed in braces, is an absolute or relative back reference. A named back 385reference can be coded as \g{name}. Back references are discussed 386<a href="#backreferences">later,</a> 387following the discussion of 388<a href="#subpattern">parenthesized subpatterns.</a> 389</P> 390<br><b> 391Absolute and relative subroutine calls 392</b><br> 393<P> 394For compatibility with Oniguruma, the non-Perl syntax \g followed by a name or 395a number enclosed either in angle brackets or single quotes, is an alternative 396syntax for referencing a subpattern as a "subroutine". Details are discussed 397<a href="#onigurumasubroutines">later.</a> 398Note that \g{...} (Perl syntax) and \g<...> (Oniguruma syntax) are <i>not</i> 399synonymous. The former is a back reference; the latter is a 400<a href="#subpatternsassubroutines">subroutine</a> 401call. 402<a name="genericchartypes"></a></P> 403<br><b> 404Generic character types 405</b><br> 406<P> 407Another use of backslash is for specifying generic character types: 408<pre> 409 \d any decimal digit 410 \D any character that is not a decimal digit 411 \h any horizontal white space character 412 \H any character that is not a horizontal white space character 413 \s any white space character 414 \S any character that is not a white space character 415 \v any vertical white space character 416 \V any character that is not a vertical white space character 417 \w any "word" character 418 \W any "non-word" character 419</pre> 420There is also the single sequence \N, which matches a non-newline character. 421This is the same as 422<a href="#fullstopdot">the "." metacharacter</a> 423when PCRE_DOTALL is not set. Perl also uses \N to match characters by name; 424PCRE does not support this. 425</P> 426<P> 427Each pair of lower and upper case escape sequences partitions the complete set 428of characters into two disjoint sets. Any given character matches one, and only 429one, of each pair. The sequences can appear both inside and outside character 430classes. They each match one character of the appropriate type. If the current 431matching point is at the end of the subject string, all of them fail, because 432there is no character to match. 433</P> 434<P> 435For compatibility with Perl, \s does not match the VT character (code 11). 436This makes it different from the the POSIX "space" class. The \s characters 437are HT (9), LF (10), FF (12), CR (13), and space (32). If "use locale;" is 438included in a Perl script, \s may match the VT character. In PCRE, it never 439does. 440</P> 441<P> 442A "word" character is an underscore or any character that is a letter or digit. 443By default, the definition of letters and digits is controlled by PCRE's 444low-valued character tables, and may vary if locale-specific matching is taking 445place (see 446<a href="pcreapi.html#localesupport">"Locale support"</a> 447in the 448<a href="pcreapi.html"><b>pcreapi</b></a> 449page). For example, in a French locale such as "fr_FR" in Unix-like systems, 450or "french" in Windows, some character codes greater than 128 are used for 451accented letters, and these are then matched by \w. The use of locales with 452Unicode is discouraged. 453</P> 454<P> 455By default, in a UTF mode, characters with values greater than 128 never match 456\d, \s, or \w, and always match \D, \S, and \W. These sequences retain 457their original meanings from before UTF support was available, mainly for 458efficiency reasons. However, if PCRE is compiled with Unicode property support, 459and the PCRE_UCP option is set, the behaviour is changed so that Unicode 460properties are used to determine character types, as follows: 461<pre> 462 \d any character that \p{Nd} matches (decimal digit) 463 \s any character that \p{Z} matches, plus HT, LF, FF, CR 464 \w any character that \p{L} or \p{N} matches, plus underscore 465</pre> 466The upper case escapes match the inverse sets of characters. Note that \d 467matches only decimal digits, whereas \w matches any Unicode digit, as well as 468any Unicode letter, and underscore. Note also that PCRE_UCP affects \b, and 469\B because they are defined in terms of \w and \W. Matching these sequences 470is noticeably slower when PCRE_UCP is set. 471</P> 472<P> 473The sequences \h, \H, \v, and \V are features that were added to Perl at 474release 5.10. In contrast to the other sequences, which match only ASCII 475characters by default, these always match certain high-valued codepoints, 476whether or not PCRE_UCP is set. The horizontal space characters are: 477<pre> 478 U+0009 Horizontal tab 479 U+0020 Space 480 U+00A0 Non-break space 481 U+1680 Ogham space mark 482 U+180E Mongolian vowel separator 483 U+2000 En quad 484 U+2001 Em quad 485 U+2002 En space 486 U+2003 Em space 487 U+2004 Three-per-em space 488 U+2005 Four-per-em space 489 U+2006 Six-per-em space 490 U+2007 Figure space 491 U+2008 Punctuation space 492 U+2009 Thin space 493 U+200A Hair space 494 U+202F Narrow no-break space 495 U+205F Medium mathematical space 496 U+3000 Ideographic space 497</pre> 498The vertical space characters are: 499<pre> 500 U+000A Linefeed 501 U+000B Vertical tab 502 U+000C Form feed 503 U+000D Carriage return 504 U+0085 Next line 505 U+2028 Line separator 506 U+2029 Paragraph separator 507</pre> 508In 8-bit, non-UTF-8 mode, only the characters with codepoints less than 256 are 509relevant. 510<a name="newlineseq"></a></P> 511<br><b> 512Newline sequences 513</b><br> 514<P> 515Outside a character class, by default, the escape sequence \R matches any 516Unicode newline sequence. In 8-bit non-UTF-8 mode \R is equivalent to the 517following: 518<pre> 519 (?>\r\n|\n|\x0b|\f|\r|\x85) 520</pre> 521This is an example of an "atomic group", details of which are given 522<a href="#atomicgroup">below.</a> 523This particular group matches either the two-character sequence CR followed by 524LF, or one of the single characters LF (linefeed, U+000A), VT (vertical tab, 525U+000B), FF (form feed, U+000C), CR (carriage return, U+000D), or NEL (next 526line, U+0085). The two-character sequence is treated as a single unit that 527cannot be split. 528</P> 529<P> 530In other modes, two additional characters whose codepoints are greater than 255 531are added: LS (line separator, U+2028) and PS (paragraph separator, U+2029). 532Unicode character property support is not needed for these characters to be 533recognized. 534</P> 535<P> 536It is possible to restrict \R to match only CR, LF, or CRLF (instead of the 537complete set of Unicode line endings) by setting the option PCRE_BSR_ANYCRLF 538either at compile time or when the pattern is matched. (BSR is an abbrevation 539for "backslash R".) This can be made the default when PCRE is built; if this is 540the case, the other behaviour can be requested via the PCRE_BSR_UNICODE option. 541It is also possible to specify these settings by starting a pattern string with 542one of the following sequences: 543<pre> 544 (*BSR_ANYCRLF) CR, LF, or CRLF only 545 (*BSR_UNICODE) any Unicode newline sequence 546</pre> 547These override the default and the options given to the compiling function, but 548they can themselves be overridden by options given to a matching function. Note 549that these special settings, which are not Perl-compatible, are recognized only 550at the very start of a pattern, and that they must be in upper case. If more 551than one of them is present, the last one is used. They can be combined with a 552change of newline convention; for example, a pattern can start with: 553<pre> 554 (*ANY)(*BSR_ANYCRLF) 555</pre> 556They can also be combined with the (*UTF8), (*UTF16), or (*UCP) special 557sequences. Inside a character class, \R is treated as an unrecognized escape 558sequence, and so matches the letter "R" by default, but causes an error if 559PCRE_EXTRA is set. 560<a name="uniextseq"></a></P> 561<br><b> 562Unicode character properties 563</b><br> 564<P> 565When PCRE is built with Unicode character property support, three additional 566escape sequences that match characters with specific properties are available. 567When in 8-bit non-UTF-8 mode, these sequences are of course limited to testing 568characters whose codepoints are less than 256, but they do work in this mode. 569The extra escape sequences are: 570<pre> 571 \p{<i>xx</i>} a character with the <i>xx</i> property 572 \P{<i>xx</i>} a character without the <i>xx</i> property 573 \X an extended Unicode sequence 574</pre> 575The property names represented by <i>xx</i> above are limited to the Unicode 576script names, the general category properties, "Any", which matches any 577character (including newline), and some special PCRE properties (described 578in the 579<a href="#extraprops">next section).</a> 580Other Perl properties such as "InMusicalSymbols" are not currently supported by 581PCRE. Note that \P{Any} does not match any characters, so always causes a 582match failure. 583</P> 584<P> 585Sets of Unicode characters are defined as belonging to certain scripts. A 586character from one of these sets can be matched using a script name. For 587example: 588<pre> 589 \p{Greek} 590 \P{Han} 591</pre> 592Those that are not part of an identified script are lumped together as 593"Common". The current list of scripts is: 594</P> 595<P> 596Arabic, 597Armenian, 598Avestan, 599Balinese, 600Bamum, 601Batak, 602Bengali, 603Bopomofo, 604Brahmi, 605Braille, 606Buginese, 607Buhid, 608Canadian_Aboriginal, 609Carian, 610Chakma, 611Cham, 612Cherokee, 613Common, 614Coptic, 615Cuneiform, 616Cypriot, 617Cyrillic, 618Deseret, 619Devanagari, 620Egyptian_Hieroglyphs, 621Ethiopic, 622Georgian, 623Glagolitic, 624Gothic, 625Greek, 626Gujarati, 627Gurmukhi, 628Han, 629Hangul, 630Hanunoo, 631Hebrew, 632Hiragana, 633Imperial_Aramaic, 634Inherited, 635Inscriptional_Pahlavi, 636Inscriptional_Parthian, 637Javanese, 638Kaithi, 639Kannada, 640Katakana, 641Kayah_Li, 642Kharoshthi, 643Khmer, 644Lao, 645Latin, 646Lepcha, 647Limbu, 648Linear_B, 649Lisu, 650Lycian, 651Lydian, 652Malayalam, 653Mandaic, 654Meetei_Mayek, 655Meroitic_Cursive, 656Meroitic_Hieroglyphs, 657Miao, 658Mongolian, 659Myanmar, 660New_Tai_Lue, 661Nko, 662Ogham, 663Old_Italic, 664Old_Persian, 665Old_South_Arabian, 666Old_Turkic, 667Ol_Chiki, 668Oriya, 669Osmanya, 670Phags_Pa, 671Phoenician, 672Rejang, 673Runic, 674Samaritan, 675Saurashtra, 676Sharada, 677Shavian, 678Sinhala, 679Sora_Sompeng, 680Sundanese, 681Syloti_Nagri, 682Syriac, 683Tagalog, 684Tagbanwa, 685Tai_Le, 686Tai_Tham, 687Tai_Viet, 688Takri, 689Tamil, 690Telugu, 691Thaana, 692Thai, 693Tibetan, 694Tifinagh, 695Ugaritic, 696Vai, 697Yi. 698</P> 699<P> 700Each character has exactly one Unicode general category property, specified by 701a two-letter abbreviation. For compatibility with Perl, negation can be 702specified by including a circumflex between the opening brace and the property 703name. For example, \p{^Lu} is the same as \P{Lu}. 704</P> 705<P> 706If only one letter is specified with \p or \P, it includes all the general 707category properties that start with that letter. In this case, in the absence 708of negation, the curly brackets in the escape sequence are optional; these two 709examples have the same effect: 710<pre> 711 \p{L} 712 \pL 713</pre> 714The following general category property codes are supported: 715<pre> 716 C Other 717 Cc Control 718 Cf Format 719 Cn Unassigned 720 Co Private use 721 Cs Surrogate 722 723 L Letter 724 Ll Lower case letter 725 Lm Modifier letter 726 Lo Other letter 727 Lt Title case letter 728 Lu Upper case letter 729 730 M Mark 731 Mc Spacing mark 732 Me Enclosing mark 733 Mn Non-spacing mark 734 735 N Number 736 Nd Decimal number 737 Nl Letter number 738 No Other number 739 740 P Punctuation 741 Pc Connector punctuation 742 Pd Dash punctuation 743 Pe Close punctuation 744 Pf Final punctuation 745 Pi Initial punctuation 746 Po Other punctuation 747 Ps Open punctuation 748 749 S Symbol 750 Sc Currency symbol 751 Sk Modifier symbol 752 Sm Mathematical symbol 753 So Other symbol 754 755 Z Separator 756 Zl Line separator 757 Zp Paragraph separator 758 Zs Space separator 759</pre> 760The special property L& is also supported: it matches a character that has 761the Lu, Ll, or Lt property, in other words, a letter that is not classified as 762a modifier or "other". 763</P> 764<P> 765The Cs (Surrogate) property applies only to characters in the range U+D800 to 766U+DFFF. Such characters are not valid in Unicode strings and so 767cannot be tested by PCRE, unless UTF validity checking has been turned off 768(see the discussion of PCRE_NO_UTF8_CHECK and PCRE_NO_UTF16_CHECK in the 769<a href="pcreapi.html"><b>pcreapi</b></a> 770page). Perl does not support the Cs property. 771</P> 772<P> 773The long synonyms for property names that Perl supports (such as \p{Letter}) 774are not supported by PCRE, nor is it permitted to prefix any of these 775properties with "Is". 776</P> 777<P> 778No character that is in the Unicode table has the Cn (unassigned) property. 779Instead, this property is assumed for any code point that is not in the 780Unicode table. 781</P> 782<P> 783Specifying caseless matching does not affect these escape sequences. For 784example, \p{Lu} always matches only upper case letters. 785</P> 786<P> 787The \X escape matches any number of Unicode characters that form an extended 788Unicode sequence. \X is equivalent to 789<pre> 790 (?>\PM\pM*) 791</pre> 792That is, it matches a character without the "mark" property, followed by zero 793or more characters with the "mark" property, and treats the sequence as an 794atomic group 795<a href="#atomicgroup">(see below).</a> 796Characters with the "mark" property are typically accents that affect the 797preceding character. None of them have codepoints less than 256, so in 7988-bit non-UTF-8 mode \X matches any one character. 799</P> 800<P> 801Note that recent versions of Perl have changed \X to match what Unicode calls 802an "extended grapheme cluster", which has a more complicated definition. 803</P> 804<P> 805Matching characters by Unicode property is not fast, because PCRE has to search 806a structure that contains data for over fifteen thousand characters. That is 807why the traditional escape sequences such as \d and \w do not use Unicode 808properties in PCRE by default, though you can make them do so by setting the 809PCRE_UCP option or by starting the pattern with (*UCP). 810<a name="extraprops"></a></P> 811<br><b> 812PCRE's additional properties 813</b><br> 814<P> 815As well as the standard Unicode properties described in the previous 816section, PCRE supports four more that make it possible to convert traditional 817escape sequences such as \w and \s and POSIX character classes to use Unicode 818properties. PCRE uses these non-standard, non-Perl properties internally when 819PCRE_UCP is set. They are: 820<pre> 821 Xan Any alphanumeric character 822 Xps Any POSIX space character 823 Xsp Any Perl space character 824 Xwd Any Perl "word" character 825</pre> 826Xan matches characters that have either the L (letter) or the N (number) 827property. Xps matches the characters tab, linefeed, vertical tab, form feed, or 828carriage return, and any other character that has the Z (separator) property. 829Xsp is the same as Xps, except that vertical tab is excluded. Xwd matches the 830same characters as Xan, plus underscore. 831<a name="resetmatchstart"></a></P> 832<br><b> 833Resetting the match start 834</b><br> 835<P> 836The escape sequence \K causes any previously matched characters not to be 837included in the final matched sequence. For example, the pattern: 838<pre> 839 foo\Kbar 840</pre> 841matches "foobar", but reports that it has matched "bar". This feature is 842similar to a lookbehind assertion 843<a href="#lookbehind">(described below).</a> 844However, in this case, the part of the subject before the real match does not 845have to be of fixed length, as lookbehind assertions do. The use of \K does 846not interfere with the setting of 847<a href="#subpattern">captured substrings.</a> 848For example, when the pattern 849<pre> 850 (foo)\Kbar 851</pre> 852matches "foobar", the first substring is still set to "foo". 853</P> 854<P> 855Perl documents that the use of \K within assertions is "not well defined". In 856PCRE, \K is acted upon when it occurs inside positive assertions, but is 857ignored in negative assertions. 858<a name="smallassertions"></a></P> 859<br><b> 860Simple assertions 861</b><br> 862<P> 863The final use of backslash is for certain simple assertions. An assertion 864specifies a condition that has to be met at a particular point in a match, 865without consuming any characters from the subject string. The use of 866subpatterns for more complicated assertions is described 867<a href="#bigassertions">below.</a> 868The backslashed assertions are: 869<pre> 870 \b matches at a word boundary 871 \B matches when not at a word boundary 872 \A matches at the start of the subject 873 \Z matches at the end of the subject 874 also matches before a newline at the end of the subject 875 \z matches only at the end of the subject 876 \G matches at the first matching position in the subject 877</pre> 878Inside a character class, \b has a different meaning; it matches the backspace 879character. If any other of these assertions appears in a character class, by 880default it matches the corresponding literal character (for example, \B 881matches the letter B). However, if the PCRE_EXTRA option is set, an "invalid 882escape sequence" error is generated instead. 883</P> 884<P> 885A word boundary is a position in the subject string where the current character 886and the previous character do not both match \w or \W (i.e. one matches 887\w and the other matches \W), or the start or end of the string if the 888first or last character matches \w, respectively. In a UTF mode, the meanings 889of \w and \W can be changed by setting the PCRE_UCP option. When this is 890done, it also affects \b and \B. Neither PCRE nor Perl has a separate "start 891of word" or "end of word" metasequence. However, whatever follows \b normally 892determines which it is. For example, the fragment \ba matches "a" at the start 893of a word. 894</P> 895<P> 896The \A, \Z, and \z assertions differ from the traditional circumflex and 897dollar (described in the next section) in that they only ever match at the very 898start and end of the subject string, whatever options are set. Thus, they are 899independent of multiline mode. These three assertions are not affected by the 900PCRE_NOTBOL or PCRE_NOTEOL options, which affect only the behaviour of the 901circumflex and dollar metacharacters. However, if the <i>startoffset</i> 902argument of <b>pcre_exec()</b> is non-zero, indicating that matching is to start 903at a point other than the beginning of the subject, \A can never match. The 904difference between \Z and \z is that \Z matches before a newline at the end 905of the string as well as at the very end, whereas \z matches only at the end. 906</P> 907<P> 908The \G assertion is true only when the current matching position is at the 909start point of the match, as specified by the <i>startoffset</i> argument of 910<b>pcre_exec()</b>. It differs from \A when the value of <i>startoffset</i> is 911non-zero. By calling <b>pcre_exec()</b> multiple times with appropriate 912arguments, you can mimic Perl's /g option, and it is in this kind of 913implementation where \G can be useful. 914</P> 915<P> 916Note, however, that PCRE's interpretation of \G, as the start of the current 917match, is subtly different from Perl's, which defines it as the end of the 918previous match. In Perl, these can be different when the previously matched 919string was empty. Because PCRE does just one match at a time, it cannot 920reproduce this behaviour. 921</P> 922<P> 923If all the alternatives of a pattern begin with \G, the expression is anchored 924to the starting match position, and the "anchored" flag is set in the compiled 925regular expression. 926</P> 927<br><a name="SEC5" href="#TOC1">CIRCUMFLEX AND DOLLAR</a><br> 928<P> 929Outside a character class, in the default matching mode, the circumflex 930character is an assertion that is true only if the current matching point is 931at the start of the subject string. If the <i>startoffset</i> argument of 932<b>pcre_exec()</b> is non-zero, circumflex can never match if the PCRE_MULTILINE 933option is unset. Inside a character class, circumflex has an entirely different 934meaning 935<a href="#characterclass">(see below).</a> 936</P> 937<P> 938Circumflex need not be the first character of the pattern if a number of 939alternatives are involved, but it should be the first thing in each alternative 940in which it appears if the pattern is ever to match that branch. If all 941possible alternatives start with a circumflex, that is, if the pattern is 942constrained to match only at the start of the subject, it is said to be an 943"anchored" pattern. (There are also other constructs that can cause a pattern 944to be anchored.) 945</P> 946<P> 947A dollar character is an assertion that is true only if the current matching 948point is at the end of the subject string, or immediately before a newline 949at the end of the string (by default). Dollar need not be the last character of 950the pattern if a number of alternatives are involved, but it should be the last 951item in any branch in which it appears. Dollar has no special meaning in a 952character class. 953</P> 954<P> 955The meaning of dollar can be changed so that it matches only at the very end of 956the string, by setting the PCRE_DOLLAR_ENDONLY option at compile time. This 957does not affect the \Z assertion. 958</P> 959<P> 960The meanings of the circumflex and dollar characters are changed if the 961PCRE_MULTILINE option is set. When this is the case, a circumflex matches 962immediately after internal newlines as well as at the start of the subject 963string. It does not match after a newline that ends the string. A dollar 964matches before any newlines in the string, as well as at the very end, when 965PCRE_MULTILINE is set. When newline is specified as the two-character 966sequence CRLF, isolated CR and LF characters do not indicate newlines. 967</P> 968<P> 969For example, the pattern /^abc$/ matches the subject string "def\nabc" (where 970\n represents a newline) in multiline mode, but not otherwise. Consequently, 971patterns that are anchored in single line mode because all branches start with 972^ are not anchored in multiline mode, and a match for circumflex is possible 973when the <i>startoffset</i> argument of <b>pcre_exec()</b> is non-zero. The 974PCRE_DOLLAR_ENDONLY option is ignored if PCRE_MULTILINE is set. 975</P> 976<P> 977Note that the sequences \A, \Z, and \z can be used to match the start and 978end of the subject in both modes, and if all branches of a pattern start with 979\A it is always anchored, whether or not PCRE_MULTILINE is set. 980<a name="fullstopdot"></a></P> 981<br><a name="SEC6" href="#TOC1">FULL STOP (PERIOD, DOT) AND \N</a><br> 982<P> 983Outside a character class, a dot in the pattern matches any one character in 984the subject string except (by default) a character that signifies the end of a 985line. 986</P> 987<P> 988When a line ending is defined as a single character, dot never matches that 989character; when the two-character sequence CRLF is used, dot does not match CR 990if it is immediately followed by LF, but otherwise it matches all characters 991(including isolated CRs and LFs). When any Unicode line endings are being 992recognized, dot does not match CR or LF or any of the other line ending 993characters. 994</P> 995<P> 996The behaviour of dot with regard to newlines can be changed. If the PCRE_DOTALL 997option is set, a dot matches any one character, without exception. If the 998two-character sequence CRLF is present in the subject string, it takes two dots 999to match it. 1000</P> 1001<P> 1002The handling of dot is entirely independent of the handling of circumflex and 1003dollar, the only relationship being that they both involve newlines. Dot has no 1004special meaning in a character class. 1005</P> 1006<P> 1007The escape sequence \N behaves like a dot, except that it is not affected by 1008the PCRE_DOTALL option. In other words, it matches any character except one 1009that signifies the end of a line. Perl also uses \N to match characters by 1010name; PCRE does not support this. 1011</P> 1012<br><a name="SEC7" href="#TOC1">MATCHING A SINGLE DATA UNIT</a><br> 1013<P> 1014Outside a character class, the escape sequence \C matches any one data unit, 1015whether or not a UTF mode is set. In the 8-bit library, one data unit is one 1016byte; in the 16-bit library it is a 16-bit unit. Unlike a dot, \C always 1017matches line-ending characters. The feature is provided in Perl in order to 1018match individual bytes in UTF-8 mode, but it is unclear how it can usefully be 1019used. Because \C breaks up characters into individual data units, matching one 1020unit with \C in a UTF mode means that the rest of the string may start with a 1021malformed UTF character. This has undefined results, because PCRE assumes that 1022it is dealing with valid UTF strings (and by default it checks this at the 1023start of processing unless the PCRE_NO_UTF8_CHECK or PCRE_NO_UTF16_CHECK option 1024is used). 1025</P> 1026<P> 1027PCRE does not allow \C to appear in lookbehind assertions 1028<a href="#lookbehind">(described below)</a> 1029in a UTF mode, because this would make it impossible to calculate the length of 1030the lookbehind. 1031</P> 1032<P> 1033In general, the \C escape sequence is best avoided. However, one 1034way of using it that avoids the problem of malformed UTF characters is to use a 1035lookahead to check the length of the next character, as in this pattern, which 1036could be used with a UTF-8 string (ignore white space and line breaks): 1037<pre> 1038 (?| (?=[\x00-\x7f])(\C) | 1039 (?=[\x80-\x{7ff}])(\C)(\C) | 1040 (?=[\x{800}-\x{ffff}])(\C)(\C)(\C) | 1041 (?=[\x{10000}-\x{1fffff}])(\C)(\C)(\C)(\C)) 1042</pre> 1043A group that starts with (?| resets the capturing parentheses numbers in each 1044alternative (see 1045<a href="#dupsubpatternnumber">"Duplicate Subpattern Numbers"</a> 1046below). The assertions at the start of each branch check the next UTF-8 1047character for values whose encoding uses 1, 2, 3, or 4 bytes, respectively. The 1048character's individual bytes are then captured by the appropriate number of 1049groups. 1050<a name="characterclass"></a></P> 1051<br><a name="SEC8" href="#TOC1">SQUARE BRACKETS AND CHARACTER CLASSES</a><br> 1052<P> 1053An opening square bracket introduces a character class, terminated by a closing 1054square bracket. A closing square bracket on its own is not special by default. 1055However, if the PCRE_JAVASCRIPT_COMPAT option is set, a lone closing square 1056bracket causes a compile-time error. If a closing square bracket is required as 1057a member of the class, it should be the first data character in the class 1058(after an initial circumflex, if present) or escaped with a backslash. 1059</P> 1060<P> 1061A character class matches a single character in the subject. In a UTF mode, the 1062character may be more than one data unit long. A matched character must be in 1063the set of characters defined by the class, unless the first character in the 1064class definition is a circumflex, in which case the subject character must not 1065be in the set defined by the class. If a circumflex is actually required as a 1066member of the class, ensure it is not the first character, or escape it with a 1067backslash. 1068</P> 1069<P> 1070For example, the character class [aeiou] matches any lower case vowel, while 1071[^aeiou] matches any character that is not a lower case vowel. Note that a 1072circumflex is just a convenient notation for specifying the characters that 1073are in the class by enumerating those that are not. A class that starts with a 1074circumflex is not an assertion; it still consumes a character from the subject 1075string, and therefore it fails if the current pointer is at the end of the 1076string. 1077</P> 1078<P> 1079In UTF-8 (UTF-16) mode, characters with values greater than 255 (0xffff) can be 1080included in a class as a literal string of data units, or by using the \x{ 1081escaping mechanism. 1082</P> 1083<P> 1084When caseless matching is set, any letters in a class represent both their 1085upper case and lower case versions, so for example, a caseless [aeiou] matches 1086"A" as well as "a", and a caseless [^aeiou] does not match "A", whereas a 1087caseful version would. In a UTF mode, PCRE always understands the concept of 1088case for characters whose values are less than 128, so caseless matching is 1089always possible. For characters with higher values, the concept of case is 1090supported if PCRE is compiled with Unicode property support, but not otherwise. 1091If you want to use caseless matching in a UTF mode for characters 128 and 1092above, you must ensure that PCRE is compiled with Unicode property support as 1093well as with UTF support. 1094</P> 1095<P> 1096Characters that might indicate line breaks are never treated in any special way 1097when matching character classes, whatever line-ending sequence is in use, and 1098whatever setting of the PCRE_DOTALL and PCRE_MULTILINE options is used. A class 1099such as [^a] always matches one of these characters. 1100</P> 1101<P> 1102The minus (hyphen) character can be used to specify a range of characters in a 1103character class. For example, [d-m] matches any letter between d and m, 1104inclusive. If a minus character is required in a class, it must be escaped with 1105a backslash or appear in a position where it cannot be interpreted as 1106indicating a range, typically as the first or last character in the class. 1107</P> 1108<P> 1109It is not possible to have the literal character "]" as the end character of a 1110range. A pattern such as [W-]46] is interpreted as a class of two characters 1111("W" and "-") followed by a literal string "46]", so it would match "W46]" or 1112"-46]". However, if the "]" is escaped with a backslash it is interpreted as 1113the end of range, so [W-\]46] is interpreted as a class containing a range 1114followed by two other characters. The octal or hexadecimal representation of 1115"]" can also be used to end a range. 1116</P> 1117<P> 1118Ranges operate in the collating sequence of character values. They can also be 1119used for characters specified numerically, for example [\000-\037]. Ranges 1120can include any characters that are valid for the current mode. 1121</P> 1122<P> 1123If a range that includes letters is used when caseless matching is set, it 1124matches the letters in either case. For example, [W-c] is equivalent to 1125[][\\^_`wxyzabc], matched caselessly, and in a non-UTF mode, if character 1126tables for a French locale are in use, [\xc8-\xcb] matches accented E 1127characters in both cases. In UTF modes, PCRE supports the concept of case for 1128characters with values greater than 128 only when it is compiled with Unicode 1129property support. 1130</P> 1131<P> 1132The character escape sequences \d, \D, \h, \H, \p, \P, \s, \S, \v, 1133\V, \w, and \W may appear in a character class, and add the characters that 1134they match to the class. For example, [\dABCDEF] matches any hexadecimal 1135digit. In UTF modes, the PCRE_UCP option affects the meanings of \d, \s, \w 1136and their upper case partners, just as it does when they appear outside a 1137character class, as described in the section entitled 1138<a href="#genericchartypes">"Generic character types"</a> 1139above. The escape sequence \b has a different meaning inside a character 1140class; it matches the backspace character. The sequences \B, \N, \R, and \X 1141are not special inside a character class. Like any other unrecognized escape 1142sequences, they are treated as the literal characters "B", "N", "R", and "X" by 1143default, but cause an error if the PCRE_EXTRA option is set. 1144</P> 1145<P> 1146A circumflex can conveniently be used with the upper case character types to 1147specify a more restricted set of characters than the matching lower case type. 1148For example, the class [^\W_] matches any letter or digit, but not underscore, 1149whereas [\w] includes underscore. A positive character class should be read as 1150"something OR something OR ..." and a negative class as "NOT something AND NOT 1151something AND NOT ...". 1152</P> 1153<P> 1154The only metacharacters that are recognized in character classes are backslash, 1155hyphen (only where it can be interpreted as specifying a range), circumflex 1156(only at the start), opening square bracket (only when it can be interpreted as 1157introducing a POSIX class name - see the next section), and the terminating 1158closing square bracket. However, escaping other non-alphanumeric characters 1159does no harm. 1160</P> 1161<br><a name="SEC9" href="#TOC1">POSIX CHARACTER CLASSES</a><br> 1162<P> 1163Perl supports the POSIX notation for character classes. This uses names 1164enclosed by [: and :] within the enclosing square brackets. PCRE also supports 1165this notation. For example, 1166<pre> 1167 [01[:alpha:]%] 1168</pre> 1169matches "0", "1", any alphabetic character, or "%". The supported class names 1170are: 1171<pre> 1172 alnum letters and digits 1173 alpha letters 1174 ascii character codes 0 - 127 1175 blank space or tab only 1176 cntrl control characters 1177 digit decimal digits (same as \d) 1178 graph printing characters, excluding space 1179 lower lower case letters 1180 print printing characters, including space 1181 punct printing characters, excluding letters and digits and space 1182 space white space (not quite the same as \s) 1183 upper upper case letters 1184 word "word" characters (same as \w) 1185 xdigit hexadecimal digits 1186</pre> 1187The "space" characters are HT (9), LF (10), VT (11), FF (12), CR (13), and 1188space (32). Notice that this list includes the VT character (code 11). This 1189makes "space" different to \s, which does not include VT (for Perl 1190compatibility). 1191</P> 1192<P> 1193The name "word" is a Perl extension, and "blank" is a GNU extension from Perl 11945.8. Another Perl extension is negation, which is indicated by a ^ character 1195after the colon. For example, 1196<pre> 1197 [12[:^digit:]] 1198</pre> 1199matches "1", "2", or any non-digit. PCRE (and Perl) also recognize the POSIX 1200syntax [.ch.] and [=ch=] where "ch" is a "collating element", but these are not 1201supported, and an error is given if they are encountered. 1202</P> 1203<P> 1204By default, in UTF modes, characters with values greater than 128 do not match 1205any of the POSIX character classes. However, if the PCRE_UCP option is passed 1206to <b>pcre_compile()</b>, some of the classes are changed so that Unicode 1207character properties are used. This is achieved by replacing the POSIX classes 1208by other sequences, as follows: 1209<pre> 1210 [:alnum:] becomes \p{Xan} 1211 [:alpha:] becomes \p{L} 1212 [:blank:] becomes \h 1213 [:digit:] becomes \p{Nd} 1214 [:lower:] becomes \p{Ll} 1215 [:space:] becomes \p{Xps} 1216 [:upper:] becomes \p{Lu} 1217 [:word:] becomes \p{Xwd} 1218</pre> 1219Negated versions, such as [:^alpha:] use \P instead of \p. The other POSIX 1220classes are unchanged, and match only characters with code points less than 1221128. 1222</P> 1223<br><a name="SEC10" href="#TOC1">VERTICAL BAR</a><br> 1224<P> 1225Vertical bar characters are used to separate alternative patterns. For example, 1226the pattern 1227<pre> 1228 gilbert|sullivan 1229</pre> 1230matches either "gilbert" or "sullivan". Any number of alternatives may appear, 1231and an empty alternative is permitted (matching the empty string). The matching 1232process tries each alternative in turn, from left to right, and the first one 1233that succeeds is used. If the alternatives are within a subpattern 1234<a href="#subpattern">(defined below),</a> 1235"succeeds" means matching the rest of the main pattern as well as the 1236alternative in the subpattern. 1237</P> 1238<br><a name="SEC11" href="#TOC1">INTERNAL OPTION SETTING</a><br> 1239<P> 1240The settings of the PCRE_CASELESS, PCRE_MULTILINE, PCRE_DOTALL, and 1241PCRE_EXTENDED options (which are Perl-compatible) can be changed from within 1242the pattern by a sequence of Perl option letters enclosed between "(?" and ")". 1243The option letters are 1244<pre> 1245 i for PCRE_CASELESS 1246 m for PCRE_MULTILINE 1247 s for PCRE_DOTALL 1248 x for PCRE_EXTENDED 1249</pre> 1250For example, (?im) sets caseless, multiline matching. It is also possible to 1251unset these options by preceding the letter with a hyphen, and a combined 1252setting and unsetting such as (?im-sx), which sets PCRE_CASELESS and 1253PCRE_MULTILINE while unsetting PCRE_DOTALL and PCRE_EXTENDED, is also 1254permitted. If a letter appears both before and after the hyphen, the option is 1255unset. 1256</P> 1257<P> 1258The PCRE-specific options PCRE_DUPNAMES, PCRE_UNGREEDY, and PCRE_EXTRA can be 1259changed in the same way as the Perl-compatible options by using the characters 1260J, U and X respectively. 1261</P> 1262<P> 1263When one of these option changes occurs at top level (that is, not inside 1264subpattern parentheses), the change applies to the remainder of the pattern 1265that follows. If the change is placed right at the start of a pattern, PCRE 1266extracts it into the global options (and it will therefore show up in data 1267extracted by the <b>pcre_fullinfo()</b> function). 1268</P> 1269<P> 1270An option change within a subpattern (see below for a description of 1271subpatterns) affects only that part of the subpattern that follows it, so 1272<pre> 1273 (a(?i)b)c 1274</pre> 1275matches abc and aBc and no other strings (assuming PCRE_CASELESS is not used). 1276By this means, options can be made to have different settings in different 1277parts of the pattern. Any changes made in one alternative do carry on 1278into subsequent branches within the same subpattern. For example, 1279<pre> 1280 (a(?i)b|c) 1281</pre> 1282matches "ab", "aB", "c", and "C", even though when matching "C" the first 1283branch is abandoned before the option setting. This is because the effects of 1284option settings happen at compile time. There would be some very weird 1285behaviour otherwise. 1286</P> 1287<P> 1288<b>Note:</b> There are other PCRE-specific options that can be set by the 1289application when the compiling or matching functions are called. In some cases 1290the pattern can contain special leading sequences such as (*CRLF) to override 1291what the application has set or what has been defaulted. Details are given in 1292the section entitled 1293<a href="#newlineseq">"Newline sequences"</a> 1294above. There are also the (*UTF8), (*UTF16), and (*UCP) leading sequences that 1295can be used to set UTF and Unicode property modes; they are equivalent to 1296setting the PCRE_UTF8, PCRE_UTF16, and the PCRE_UCP options, respectively. 1297<a name="subpattern"></a></P> 1298<br><a name="SEC12" href="#TOC1">SUBPATTERNS</a><br> 1299<P> 1300Subpatterns are delimited by parentheses (round brackets), which can be nested. 1301Turning part of a pattern into a subpattern does two things: 1302<br> 1303<br> 13041. It localizes a set of alternatives. For example, the pattern 1305<pre> 1306 cat(aract|erpillar|) 1307</pre> 1308matches "cataract", "caterpillar", or "cat". Without the parentheses, it would 1309match "cataract", "erpillar" or an empty string. 1310<br> 1311<br> 13122. It sets up the subpattern as a capturing subpattern. This means that, when 1313the whole pattern matches, that portion of the subject string that matched the 1314subpattern is passed back to the caller via the <i>ovector</i> argument of the 1315matching function. (This applies only to the traditional matching functions; 1316the DFA matching functions do not support capturing.) 1317</P> 1318<P> 1319Opening parentheses are counted from left to right (starting from 1) to obtain 1320numbers for the capturing subpatterns. For example, if the string "the red 1321king" is matched against the pattern 1322<pre> 1323 the ((red|white) (king|queen)) 1324</pre> 1325the captured substrings are "red king", "red", and "king", and are numbered 1, 13262, and 3, respectively. 1327</P> 1328<P> 1329The fact that plain parentheses fulfil two functions is not always helpful. 1330There are often times when a grouping subpattern is required without a 1331capturing requirement. If an opening parenthesis is followed by a question mark 1332and a colon, the subpattern does not do any capturing, and is not counted when 1333computing the number of any subsequent capturing subpatterns. For example, if 1334the string "the white queen" is matched against the pattern 1335<pre> 1336 the ((?:red|white) (king|queen)) 1337</pre> 1338the captured substrings are "white queen" and "queen", and are numbered 1 and 13392. The maximum number of capturing subpatterns is 65535. 1340</P> 1341<P> 1342As a convenient shorthand, if any option settings are required at the start of 1343a non-capturing subpattern, the option letters may appear between the "?" and 1344the ":". Thus the two patterns 1345<pre> 1346 (?i:saturday|sunday) 1347 (?:(?i)saturday|sunday) 1348</pre> 1349match exactly the same set of strings. Because alternative branches are tried 1350from left to right, and options are not reset until the end of the subpattern 1351is reached, an option setting in one branch does affect subsequent branches, so 1352the above patterns match "SUNDAY" as well as "Saturday". 1353<a name="dupsubpatternnumber"></a></P> 1354<br><a name="SEC13" href="#TOC1">DUPLICATE SUBPATTERN NUMBERS</a><br> 1355<P> 1356Perl 5.10 introduced a feature whereby each alternative in a subpattern uses 1357the same numbers for its capturing parentheses. Such a subpattern starts with 1358(?| and is itself a non-capturing subpattern. For example, consider this 1359pattern: 1360<pre> 1361 (?|(Sat)ur|(Sun))day 1362</pre> 1363Because the two alternatives are inside a (?| group, both sets of capturing 1364parentheses are numbered one. Thus, when the pattern matches, you can look 1365at captured substring number one, whichever alternative matched. This construct 1366is useful when you want to capture part, but not all, of one of a number of 1367alternatives. Inside a (?| group, parentheses are numbered as usual, but the 1368number is reset at the start of each branch. The numbers of any capturing 1369parentheses that follow the subpattern start after the highest number used in 1370any branch. The following example is taken from the Perl documentation. The 1371numbers underneath show in which buffer the captured content will be stored. 1372<pre> 1373 # before ---------------branch-reset----------- after 1374 / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x 1375 # 1 2 2 3 2 3 4 1376</pre> 1377A back reference to a numbered subpattern uses the most recent value that is 1378set for that number by any subpattern. The following pattern matches "abcabc" 1379or "defdef": 1380<pre> 1381 /(?|(abc)|(def))\1/ 1382</pre> 1383In contrast, a subroutine call to a numbered subpattern always refers to the 1384first one in the pattern with the given number. The following pattern matches 1385"abcabc" or "defabc": 1386<pre> 1387 /(?|(abc)|(def))(?1)/ 1388</pre> 1389If a 1390<a href="#conditions">condition test</a> 1391for a subpattern's having matched refers to a non-unique number, the test is 1392true if any of the subpatterns of that number have matched. 1393</P> 1394<P> 1395An alternative approach to using this "branch reset" feature is to use 1396duplicate named subpatterns, as described in the next section. 1397</P> 1398<br><a name="SEC14" href="#TOC1">NAMED SUBPATTERNS</a><br> 1399<P> 1400Identifying capturing parentheses by number is simple, but it can be very hard 1401to keep track of the numbers in complicated regular expressions. Furthermore, 1402if an expression is modified, the numbers may change. To help with this 1403difficulty, PCRE supports the naming of subpatterns. This feature was not 1404added to Perl until release 5.10. Python had the feature earlier, and PCRE 1405introduced it at release 4.0, using the Python syntax. PCRE now supports both 1406the Perl and the Python syntax. Perl allows identically numbered subpatterns to 1407have different names, but PCRE does not. 1408</P> 1409<P> 1410In PCRE, a subpattern can be named in one of three ways: (?<name>...) or 1411(?'name'...) as in Perl, or (?P<name>...) as in Python. References to capturing 1412parentheses from other parts of the pattern, such as 1413<a href="#backreferences">back references,</a> 1414<a href="#recursion">recursion,</a> 1415and 1416<a href="#conditions">conditions,</a> 1417can be made by name as well as by number. 1418</P> 1419<P> 1420Names consist of up to 32 alphanumeric characters and underscores. Named 1421capturing parentheses are still allocated numbers as well as names, exactly as 1422if the names were not present. The PCRE API provides function calls for 1423extracting the name-to-number translation table from a compiled pattern. There 1424is also a convenience function for extracting a captured substring by name. 1425</P> 1426<P> 1427By default, a name must be unique within a pattern, but it is possible to relax 1428this constraint by setting the PCRE_DUPNAMES option at compile time. (Duplicate 1429names are also always permitted for subpatterns with the same number, set up as 1430described in the previous section.) Duplicate names can be useful for patterns 1431where only one instance of the named parentheses can match. Suppose you want to 1432match the name of a weekday, either as a 3-letter abbreviation or as the full 1433name, and in both cases you want to extract the abbreviation. This pattern 1434(ignoring the line breaks) does the job: 1435<pre> 1436 (?<DN>Mon|Fri|Sun)(?:day)?| 1437 (?<DN>Tue)(?:sday)?| 1438 (?<DN>Wed)(?:nesday)?| 1439 (?<DN>Thu)(?:rsday)?| 1440 (?<DN>Sat)(?:urday)? 1441</pre> 1442There are five capturing substrings, but only one is ever set after a match. 1443(An alternative way of solving this problem is to use a "branch reset" 1444subpattern, as described in the previous section.) 1445</P> 1446<P> 1447The convenience function for extracting the data by name returns the substring 1448for the first (and in this example, the only) subpattern of that name that 1449matched. This saves searching to find which numbered subpattern it was. 1450</P> 1451<P> 1452If you make a back reference to a non-unique named subpattern from elsewhere in 1453the pattern, the one that corresponds to the first occurrence of the name is 1454used. In the absence of duplicate numbers (see the previous section) this is 1455the one with the lowest number. If you use a named reference in a condition 1456test (see the 1457<a href="#conditions">section about conditions</a> 1458below), either to check whether a subpattern has matched, or to check for 1459recursion, all subpatterns with the same name are tested. If the condition is 1460true for any one of them, the overall condition is true. This is the same 1461behaviour as testing by number. For further details of the interfaces for 1462handling named subpatterns, see the 1463<a href="pcreapi.html"><b>pcreapi</b></a> 1464documentation. 1465</P> 1466<P> 1467<b>Warning:</b> You cannot use different names to distinguish between two 1468subpatterns with the same number because PCRE uses only the numbers when 1469matching. For this reason, an error is given at compile time if different names 1470are given to subpatterns with the same number. However, you can give the same 1471name to subpatterns with the same number, even when PCRE_DUPNAMES is not set. 1472</P> 1473<br><a name="SEC15" href="#TOC1">REPETITION</a><br> 1474<P> 1475Repetition is specified by quantifiers, which can follow any of the following 1476items: 1477<pre> 1478 a literal data character 1479 the dot metacharacter 1480 the \C escape sequence 1481 the \X escape sequence 1482 the \R escape sequence 1483 an escape such as \d or \pL that matches a single character 1484 a character class 1485 a back reference (see next section) 1486 a parenthesized subpattern (including assertions) 1487 a subroutine call to a subpattern (recursive or otherwise) 1488</pre> 1489The general repetition quantifier specifies a minimum and maximum number of 1490permitted matches, by giving the two numbers in curly brackets (braces), 1491separated by a comma. The numbers must be less than 65536, and the first must 1492be less than or equal to the second. For example: 1493<pre> 1494 z{2,4} 1495</pre> 1496matches "zz", "zzz", or "zzzz". A closing brace on its own is not a special 1497character. If the second number is omitted, but the comma is present, there is 1498no upper limit; if the second number and the comma are both omitted, the 1499quantifier specifies an exact number of required matches. Thus 1500<pre> 1501 [aeiou]{3,} 1502</pre> 1503matches at least 3 successive vowels, but may match many more, while 1504<pre> 1505 \d{8} 1506</pre> 1507matches exactly 8 digits. An opening curly bracket that appears in a position 1508where a quantifier is not allowed, or one that does not match the syntax of a 1509quantifier, is taken as a literal character. For example, {,6} is not a 1510quantifier, but a literal string of four characters. 1511</P> 1512<P> 1513In UTF modes, quantifiers apply to characters rather than to individual data 1514units. Thus, for example, \x{100}{2} matches two characters, each of 1515which is represented by a two-byte sequence in a UTF-8 string. Similarly, 1516\X{3} matches three Unicode extended sequences, each of which may be several 1517data units long (and they may be of different lengths). 1518</P> 1519<P> 1520The quantifier {0} is permitted, causing the expression to behave as if the 1521previous item and the quantifier were not present. This may be useful for 1522subpatterns that are referenced as 1523<a href="#subpatternsassubroutines">subroutines</a> 1524from elsewhere in the pattern (but see also the section entitled 1525<a href="#subdefine">"Defining subpatterns for use by reference only"</a> 1526below). Items other than subpatterns that have a {0} quantifier are omitted 1527from the compiled pattern. 1528</P> 1529<P> 1530For convenience, the three most common quantifiers have single-character 1531abbreviations: 1532<pre> 1533 * is equivalent to {0,} 1534 + is equivalent to {1,} 1535 ? is equivalent to {0,1} 1536</pre> 1537It is possible to construct infinite loops by following a subpattern that can 1538match no characters with a quantifier that has no upper limit, for example: 1539<pre> 1540 (a?)* 1541</pre> 1542Earlier versions of Perl and PCRE used to give an error at compile time for 1543such patterns. However, because there are cases where this can be useful, such 1544patterns are now accepted, but if any repetition of the subpattern does in fact 1545match no characters, the loop is forcibly broken. 1546</P> 1547<P> 1548By default, the quantifiers are "greedy", that is, they match as much as 1549possible (up to the maximum number of permitted times), without causing the 1550rest of the pattern to fail. The classic example of where this gives problems 1551is in trying to match comments in C programs. These appear between /* and */ 1552and within the comment, individual * and / characters may appear. An attempt to 1553match C comments by applying the pattern 1554<pre> 1555 /\*.*\*/ 1556</pre> 1557to the string 1558<pre> 1559 /* first comment */ not comment /* second comment */ 1560</pre> 1561fails, because it matches the entire string owing to the greediness of the .* 1562item. 1563</P> 1564<P> 1565However, if a quantifier is followed by a question mark, it ceases to be 1566greedy, and instead matches the minimum number of times possible, so the 1567pattern 1568<pre> 1569 /\*.*?\*/ 1570</pre> 1571does the right thing with the C comments. The meaning of the various 1572quantifiers is not otherwise changed, just the preferred number of matches. 1573Do not confuse this use of question mark with its use as a quantifier in its 1574own right. Because it has two uses, it can sometimes appear doubled, as in 1575<pre> 1576 \d??\d 1577</pre> 1578which matches one digit by preference, but can match two if that is the only 1579way the rest of the pattern matches. 1580</P> 1581<P> 1582If the PCRE_UNGREEDY option is set (an option that is not available in Perl), 1583the quantifiers are not greedy by default, but individual ones can be made 1584greedy by following them with a question mark. In other words, it inverts the 1585default behaviour. 1586</P> 1587<P> 1588When a parenthesized subpattern is quantified with a minimum repeat count that 1589is greater than 1 or with a limited maximum, more memory is required for the 1590compiled pattern, in proportion to the size of the minimum or maximum. 1591</P> 1592<P> 1593If a pattern starts with .* or .{0,} and the PCRE_DOTALL option (equivalent 1594to Perl's /s) is set, thus allowing the dot to match newlines, the pattern is 1595implicitly anchored, because whatever follows will be tried against every 1596character position in the subject string, so there is no point in retrying the 1597overall match at any position after the first. PCRE normally treats such a 1598pattern as though it were preceded by \A. 1599</P> 1600<P> 1601In cases where it is known that the subject string contains no newlines, it is 1602worth setting PCRE_DOTALL in order to obtain this optimization, or 1603alternatively using ^ to indicate anchoring explicitly. 1604</P> 1605<P> 1606However, there is one situation where the optimization cannot be used. When .* 1607is inside capturing parentheses that are the subject of a back reference 1608elsewhere in the pattern, a match at the start may fail where a later one 1609succeeds. Consider, for example: 1610<pre> 1611 (.*)abc\1 1612</pre> 1613If the subject is "xyz123abc123" the match point is the fourth character. For 1614this reason, such a pattern is not implicitly anchored. 1615</P> 1616<P> 1617When a capturing subpattern is repeated, the value captured is the substring 1618that matched the final iteration. For example, after 1619<pre> 1620 (tweedle[dume]{3}\s*)+ 1621</pre> 1622has matched "tweedledum tweedledee" the value of the captured substring is 1623"tweedledee". However, if there are nested capturing subpatterns, the 1624corresponding captured values may have been set in previous iterations. For 1625example, after 1626<pre> 1627 /(a|(b))+/ 1628</pre> 1629matches "aba" the value of the second captured substring is "b". 1630<a name="atomicgroup"></a></P> 1631<br><a name="SEC16" href="#TOC1">ATOMIC GROUPING AND POSSESSIVE QUANTIFIERS</a><br> 1632<P> 1633With both maximizing ("greedy") and minimizing ("ungreedy" or "lazy") 1634repetition, failure of what follows normally causes the repeated item to be 1635re-evaluated to see if a different number of repeats allows the rest of the 1636pattern to match. Sometimes it is useful to prevent this, either to change the 1637nature of the match, or to cause it fail earlier than it otherwise might, when 1638the author of the pattern knows there is no point in carrying on. 1639</P> 1640<P> 1641Consider, for example, the pattern \d+foo when applied to the subject line 1642<pre> 1643 123456bar 1644</pre> 1645After matching all 6 digits and then failing to match "foo", the normal 1646action of the matcher is to try again with only 5 digits matching the \d+ 1647item, and then with 4, and so on, before ultimately failing. "Atomic grouping" 1648(a term taken from Jeffrey Friedl's book) provides the means for specifying 1649that once a subpattern has matched, it is not to be re-evaluated in this way. 1650</P> 1651<P> 1652If we use atomic grouping for the previous example, the matcher gives up 1653immediately on failing to match "foo" the first time. The notation is a kind of 1654special parenthesis, starting with (?> as in this example: 1655<pre> 1656 (?>\d+)foo 1657</pre> 1658This kind of parenthesis "locks up" the part of the pattern it contains once 1659it has matched, and a failure further into the pattern is prevented from 1660backtracking into it. Backtracking past it to previous items, however, works as 1661normal. 1662</P> 1663<P> 1664An alternative description is that a subpattern of this type matches the string 1665of characters that an identical standalone pattern would match, if anchored at 1666the current point in the subject string. 1667</P> 1668<P> 1669Atomic grouping subpatterns are not capturing subpatterns. Simple cases such as 1670the above example can be thought of as a maximizing repeat that must swallow 1671everything it can. So, while both \d+ and \d+? are prepared to adjust the 1672number of digits they match in order to make the rest of the pattern match, 1673(?>\d+) can only match an entire sequence of digits. 1674</P> 1675<P> 1676Atomic groups in general can of course contain arbitrarily complicated 1677subpatterns, and can be nested. However, when the subpattern for an atomic 1678group is just a single repeated item, as in the example above, a simpler 1679notation, called a "possessive quantifier" can be used. This consists of an 1680additional + character following a quantifier. Using this notation, the 1681previous example can be rewritten as 1682<pre> 1683 \d++foo 1684</pre> 1685Note that a possessive quantifier can be used with an entire group, for 1686example: 1687<pre> 1688 (abc|xyz){2,3}+ 1689</pre> 1690Possessive quantifiers are always greedy; the setting of the PCRE_UNGREEDY 1691option is ignored. They are a convenient notation for the simpler forms of 1692atomic group. However, there is no difference in the meaning of a possessive 1693quantifier and the equivalent atomic group, though there may be a performance 1694difference; possessive quantifiers should be slightly faster. 1695</P> 1696<P> 1697The possessive quantifier syntax is an extension to the Perl 5.8 syntax. 1698Jeffrey Friedl originated the idea (and the name) in the first edition of his 1699book. Mike McCloskey liked it, so implemented it when he built Sun's Java 1700package, and PCRE copied it from there. It ultimately found its way into Perl 1701at release 5.10. 1702</P> 1703<P> 1704PCRE has an optimization that automatically "possessifies" certain simple 1705pattern constructs. For example, the sequence A+B is treated as A++B because 1706there is no point in backtracking into a sequence of A's when B must follow. 1707</P> 1708<P> 1709When a pattern contains an unlimited repeat inside a subpattern that can itself 1710be repeated an unlimited number of times, the use of an atomic group is the 1711only way to avoid some failing matches taking a very long time indeed. The 1712pattern 1713<pre> 1714 (\D+|<\d+>)*[!?] 1715</pre> 1716matches an unlimited number of substrings that either consist of non-digits, or 1717digits enclosed in <>, followed by either ! or ?. When it matches, it runs 1718quickly. However, if it is applied to 1719<pre> 1720 aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa 1721</pre> 1722it takes a long time before reporting failure. This is because the string can 1723be divided between the internal \D+ repeat and the external * repeat in a 1724large number of ways, and all have to be tried. (The example uses [!?] rather 1725than a single character at the end, because both PCRE and Perl have an 1726optimization that allows for fast failure when a single character is used. They 1727remember the last single character that is required for a match, and fail early 1728if it is not present in the string.) If the pattern is changed so that it uses 1729an atomic group, like this: 1730<pre> 1731 ((?>\D+)|<\d+>)*[!?] 1732</pre> 1733sequences of non-digits cannot be broken, and failure happens quickly. 1734<a name="backreferences"></a></P> 1735<br><a name="SEC17" href="#TOC1">BACK REFERENCES</a><br> 1736<P> 1737Outside a character class, a backslash followed by a digit greater than 0 (and 1738possibly further digits) is a back reference to a capturing subpattern earlier 1739(that is, to its left) in the pattern, provided there have been that many 1740previous capturing left parentheses. 1741</P> 1742<P> 1743However, if the decimal number following the backslash is less than 10, it is 1744always taken as a back reference, and causes an error only if there are not 1745that many capturing left parentheses in the entire pattern. In other words, the 1746parentheses that are referenced need not be to the left of the reference for 1747numbers less than 10. A "forward back reference" of this type can make sense 1748when a repetition is involved and the subpattern to the right has participated 1749in an earlier iteration. 1750</P> 1751<P> 1752It is not possible to have a numerical "forward back reference" to a subpattern 1753whose number is 10 or more using this syntax because a sequence such as \50 is 1754interpreted as a character defined in octal. See the subsection entitled 1755"Non-printing characters" 1756<a href="#digitsafterbackslash">above</a> 1757for further details of the handling of digits following a backslash. There is 1758no such problem when named parentheses are used. A back reference to any 1759subpattern is possible using named parentheses (see below). 1760</P> 1761<P> 1762Another way of avoiding the ambiguity inherent in the use of digits following a 1763backslash is to use the \g escape sequence. This escape must be followed by an 1764unsigned number or a negative number, optionally enclosed in braces. These 1765examples are all identical: 1766<pre> 1767 (ring), \1 1768 (ring), \g1 1769 (ring), \g{1} 1770</pre> 1771An unsigned number specifies an absolute reference without the ambiguity that 1772is present in the older syntax. It is also useful when literal digits follow 1773the reference. A negative number is a relative reference. Consider this 1774example: 1775<pre> 1776 (abc(def)ghi)\g{-1} 1777</pre> 1778The sequence \g{-1} is a reference to the most recently started capturing 1779subpattern before \g, that is, is it equivalent to \2 in this example. 1780Similarly, \g{-2} would be equivalent to \1. The use of relative references 1781can be helpful in long patterns, and also in patterns that are created by 1782joining together fragments that contain references within themselves. 1783</P> 1784<P> 1785A back reference matches whatever actually matched the capturing subpattern in 1786the current subject string, rather than anything matching the subpattern 1787itself (see 1788<a href="#subpatternsassubroutines">"Subpatterns as subroutines"</a> 1789below for a way of doing that). So the pattern 1790<pre> 1791 (sens|respons)e and \1ibility 1792</pre> 1793matches "sense and sensibility" and "response and responsibility", but not 1794"sense and responsibility". If caseful matching is in force at the time of the 1795back reference, the case of letters is relevant. For example, 1796<pre> 1797 ((?i)rah)\s+\1 1798</pre> 1799matches "rah rah" and "RAH RAH", but not "RAH rah", even though the original 1800capturing subpattern is matched caselessly. 1801</P> 1802<P> 1803There are several different ways of writing back references to named 1804subpatterns. The .NET syntax \k{name} and the Perl syntax \k<name> or 1805\k'name' are supported, as is the Python syntax (?P=name). Perl 5.10's unified 1806back reference syntax, in which \g can be used for both numeric and named 1807references, is also supported. We could rewrite the above example in any of 1808the following ways: 1809<pre> 1810 (?<p1>(?i)rah)\s+\k<p1> 1811 (?'p1'(?i)rah)\s+\k{p1} 1812 (?P<p1>(?i)rah)\s+(?P=p1) 1813 (?<p1>(?i)rah)\s+\g{p1} 1814</pre> 1815A subpattern that is referenced by name may appear in the pattern before or 1816after the reference. 1817</P> 1818<P> 1819There may be more than one back reference to the same subpattern. If a 1820subpattern has not actually been used in a particular match, any back 1821references to it always fail by default. For example, the pattern 1822<pre> 1823 (a|(bc))\2 1824</pre> 1825always fails if it starts to match "a" rather than "bc". However, if the 1826PCRE_JAVASCRIPT_COMPAT option is set at compile time, a back reference to an 1827unset value matches an empty string. 1828</P> 1829<P> 1830Because there may be many capturing parentheses in a pattern, all digits 1831following a backslash are taken as part of a potential back reference number. 1832If the pattern continues with a digit character, some delimiter must be used to 1833terminate the back reference. If the PCRE_EXTENDED option is set, this can be 1834white space. Otherwise, the \g{ syntax or an empty comment (see 1835<a href="#comments">"Comments"</a> 1836below) can be used. 1837</P> 1838<br><b> 1839Recursive back references 1840</b><br> 1841<P> 1842A back reference that occurs inside the parentheses to which it refers fails 1843when the subpattern is first used, so, for example, (a\1) never matches. 1844However, such references can be useful inside repeated subpatterns. For 1845example, the pattern 1846<pre> 1847 (a|b\1)+ 1848</pre> 1849matches any number of "a"s and also "aba", "ababbaa" etc. At each iteration of 1850the subpattern, the back reference matches the character string corresponding 1851to the previous iteration. In order for this to work, the pattern must be such 1852that the first iteration does not need to match the back reference. This can be 1853done using alternation, as in the example above, or by a quantifier with a 1854minimum of zero. 1855</P> 1856<P> 1857Back references of this type cause the group that they reference to be treated 1858as an 1859<a href="#atomicgroup">atomic group.</a> 1860Once the whole group has been matched, a subsequent matching failure cannot 1861cause backtracking into the middle of the group. 1862<a name="bigassertions"></a></P> 1863<br><a name="SEC18" href="#TOC1">ASSERTIONS</a><br> 1864<P> 1865An assertion is a test on the characters following or preceding the current 1866matching point that does not actually consume any characters. The simple 1867assertions coded as \b, \B, \A, \G, \Z, \z, ^ and $ are described 1868<a href="#smallassertions">above.</a> 1869</P> 1870<P> 1871More complicated assertions are coded as subpatterns. There are two kinds: 1872those that look ahead of the current position in the subject string, and those 1873that look behind it. An assertion subpattern is matched in the normal way, 1874except that it does not cause the current matching position to be changed. 1875</P> 1876<P> 1877Assertion subpatterns are not capturing subpatterns. If such an assertion 1878contains capturing subpatterns within it, these are counted for the purposes of 1879numbering the capturing subpatterns in the whole pattern. However, substring 1880capturing is carried out only for positive assertions, because it does not make 1881sense for negative assertions. 1882</P> 1883<P> 1884For compatibility with Perl, assertion subpatterns may be repeated; though 1885it makes no sense to assert the same thing several times, the side effect of 1886capturing parentheses may occasionally be useful. In practice, there only three 1887cases: 1888<br> 1889<br> 1890(1) If the quantifier is {0}, the assertion is never obeyed during matching. 1891However, it may contain internal capturing parenthesized groups that are called 1892from elsewhere via the 1893<a href="#subpatternsassubroutines">subroutine mechanism.</a> 1894<br> 1895<br> 1896(2) If quantifier is {0,n} where n is greater than zero, it is treated as if it 1897were {0,1}. At run time, the rest of the pattern match is tried with and 1898without the assertion, the order depending on the greediness of the quantifier. 1899<br> 1900<br> 1901(3) If the minimum repetition is greater than zero, the quantifier is ignored. 1902The assertion is obeyed just once when encountered during matching. 1903</P> 1904<br><b> 1905Lookahead assertions 1906</b><br> 1907<P> 1908Lookahead assertions start with (?= for positive assertions and (?! for 1909negative assertions. For example, 1910<pre> 1911 \w+(?=;) 1912</pre> 1913matches a word followed by a semicolon, but does not include the semicolon in 1914the match, and 1915<pre> 1916 foo(?!bar) 1917</pre> 1918matches any occurrence of "foo" that is not followed by "bar". Note that the 1919apparently similar pattern 1920<pre> 1921 (?!foo)bar 1922</pre> 1923does not find an occurrence of "bar" that is preceded by something other than 1924"foo"; it finds any occurrence of "bar" whatsoever, because the assertion 1925(?!foo) is always true when the next three characters are "bar". A 1926lookbehind assertion is needed to achieve the other effect. 1927</P> 1928<P> 1929If you want to force a matching failure at some point in a pattern, the most 1930convenient way to do it is with (?!) because an empty string always matches, so 1931an assertion that requires there not to be an empty string must always fail. 1932The backtracking control verb (*FAIL) or (*F) is a synonym for (?!). 1933<a name="lookbehind"></a></P> 1934<br><b> 1935Lookbehind assertions 1936</b><br> 1937<P> 1938Lookbehind assertions start with (?<= for positive assertions and (?<! for 1939negative assertions. For example, 1940<pre> 1941 (?<!foo)bar 1942</pre> 1943does find an occurrence of "bar" that is not preceded by "foo". The contents of 1944a lookbehind assertion are restricted such that all the strings it matches must 1945have a fixed length. However, if there are several top-level alternatives, they 1946do not all have to have the same fixed length. Thus 1947<pre> 1948 (?<=bullock|donkey) 1949</pre> 1950is permitted, but 1951<pre> 1952 (?<!dogs?|cats?) 1953</pre> 1954causes an error at compile time. Branches that match different length strings 1955are permitted only at the top level of a lookbehind assertion. This is an 1956extension compared with Perl, which requires all branches to match the same 1957length of string. An assertion such as 1958<pre> 1959 (?<=ab(c|de)) 1960</pre> 1961is not permitted, because its single top-level branch can match two different 1962lengths, but it is acceptable to PCRE if rewritten to use two top-level 1963branches: 1964<pre> 1965 (?<=abc|abde) 1966</pre> 1967In some cases, the escape sequence \K 1968<a href="#resetmatchstart">(see above)</a> 1969can be used instead of a lookbehind assertion to get round the fixed-length 1970restriction. 1971</P> 1972<P> 1973The implementation of lookbehind assertions is, for each alternative, to 1974temporarily move the current position back by the fixed length and then try to 1975match. If there are insufficient characters before the current position, the 1976assertion fails. 1977</P> 1978<P> 1979In a UTF mode, PCRE does not allow the \C escape (which matches a single data 1980unit even in a UTF mode) to appear in lookbehind assertions, because it makes 1981it impossible to calculate the length of the lookbehind. The \X and \R 1982escapes, which can match different numbers of data units, are also not 1983permitted. 1984</P> 1985<P> 1986<a href="#subpatternsassubroutines">"Subroutine"</a> 1987calls (see below) such as (?2) or (?&X) are permitted in lookbehinds, as long 1988as the subpattern matches a fixed-length string. 1989<a href="#recursion">Recursion,</a> 1990however, is not supported. 1991</P> 1992<P> 1993Possessive quantifiers can be used in conjunction with lookbehind assertions to 1994specify efficient matching of fixed-length strings at the end of subject 1995strings. Consider a simple pattern such as 1996<pre> 1997 abcd$ 1998</pre> 1999when applied to a long string that does not match. Because matching proceeds 2000from left to right, PCRE will look for each "a" in the subject and then see if 2001what follows matches the rest of the pattern. If the pattern is specified as 2002<pre> 2003 ^.*abcd$ 2004</pre> 2005the initial .* matches the entire string at first, but when this fails (because 2006there is no following "a"), it backtracks to match all but the last character, 2007then all but the last two characters, and so on. Once again the search for "a" 2008covers the entire string, from right to left, so we are no better off. However, 2009if the pattern is written as 2010<pre> 2011 ^.*+(?<=abcd) 2012</pre> 2013there can be no backtracking for the .*+ item; it can match only the entire 2014string. The subsequent lookbehind assertion does a single test on the last four 2015characters. If it fails, the match fails immediately. For long strings, this 2016approach makes a significant difference to the processing time. 2017</P> 2018<br><b> 2019Using multiple assertions 2020</b><br> 2021<P> 2022Several assertions (of any sort) may occur in succession. For example, 2023<pre> 2024 (?<=\d{3})(?<!999)foo 2025</pre> 2026matches "foo" preceded by three digits that are not "999". Notice that each of 2027the assertions is applied independently at the same point in the subject 2028string. First there is a check that the previous three characters are all 2029digits, and then there is a check that the same three characters are not "999". 2030This pattern does <i>not</i> match "foo" preceded by six characters, the first 2031of which are digits and the last three of which are not "999". For example, it 2032doesn't match "123abcfoo". A pattern to do that is 2033<pre> 2034 (?<=\d{3}...)(?<!999)foo 2035</pre> 2036This time the first assertion looks at the preceding six characters, checking 2037that the first three are digits, and then the second assertion checks that the 2038preceding three characters are not "999". 2039</P> 2040<P> 2041Assertions can be nested in any combination. For example, 2042<pre> 2043 (?<=(?<!foo)bar)baz 2044</pre> 2045matches an occurrence of "baz" that is preceded by "bar" which in turn is not 2046preceded by "foo", while 2047<pre> 2048 (?<=\d{3}(?!999)...)foo 2049</pre> 2050is another pattern that matches "foo" preceded by three digits and any three 2051characters that are not "999". 2052<a name="conditions"></a></P> 2053<br><a name="SEC19" href="#TOC1">CONDITIONAL SUBPATTERNS</a><br> 2054<P> 2055It is possible to cause the matching process to obey a subpattern 2056conditionally or to choose between two alternative subpatterns, depending on 2057the result of an assertion, or whether a specific capturing subpattern has 2058already been matched. The two possible forms of conditional subpattern are: 2059<pre> 2060 (?(condition)yes-pattern) 2061 (?(condition)yes-pattern|no-pattern) 2062</pre> 2063If the condition is satisfied, the yes-pattern is used; otherwise the 2064no-pattern (if present) is used. If there are more than two alternatives in the 2065subpattern, a compile-time error occurs. Each of the two alternatives may 2066itself contain nested subpatterns of any form, including conditional 2067subpatterns; the restriction to two alternatives applies only at the level of 2068the condition. This pattern fragment is an example where the alternatives are 2069complex: 2070<pre> 2071 (?(1) (A|B|C) | (D | (?(2)E|F) | E) ) 2072 2073</PRE> 2074</P> 2075<P> 2076There are four kinds of condition: references to subpatterns, references to 2077recursion, a pseudo-condition called DEFINE, and assertions. 2078</P> 2079<br><b> 2080Checking for a used subpattern by number 2081</b><br> 2082<P> 2083If the text between the parentheses consists of a sequence of digits, the 2084condition is true if a capturing subpattern of that number has previously 2085matched. If there is more than one capturing subpattern with the same number 2086(see the earlier 2087<a href="#recursion">section about duplicate subpattern numbers),</a> 2088the condition is true if any of them have matched. An alternative notation is 2089to precede the digits with a plus or minus sign. In this case, the subpattern 2090number is relative rather than absolute. The most recently opened parentheses 2091can be referenced by (?(-1), the next most recent by (?(-2), and so on. Inside 2092loops it can also make sense to refer to subsequent groups. The next 2093parentheses to be opened can be referenced as (?(+1), and so on. (The value 2094zero in any of these forms is not used; it provokes a compile-time error.) 2095</P> 2096<P> 2097Consider the following pattern, which contains non-significant white space to 2098make it more readable (assume the PCRE_EXTENDED option) and to divide it into 2099three parts for ease of discussion: 2100<pre> 2101 ( \( )? [^()]+ (?(1) \) ) 2102</pre> 2103The first part matches an optional opening parenthesis, and if that 2104character is present, sets it as the first captured substring. The second part 2105matches one or more characters that are not parentheses. The third part is a 2106conditional subpattern that tests whether or not the first set of parentheses 2107matched. If they did, that is, if subject started with an opening parenthesis, 2108the condition is true, and so the yes-pattern is executed and a closing 2109parenthesis is required. Otherwise, since no-pattern is not present, the 2110subpattern matches nothing. In other words, this pattern matches a sequence of 2111non-parentheses, optionally enclosed in parentheses. 2112</P> 2113<P> 2114If you were embedding this pattern in a larger one, you could use a relative 2115reference: 2116<pre> 2117 ...other stuff... ( \( )? [^()]+ (?(-1) \) ) ... 2118</pre> 2119This makes the fragment independent of the parentheses in the larger pattern. 2120</P> 2121<br><b> 2122Checking for a used subpattern by name 2123</b><br> 2124<P> 2125Perl uses the syntax (?(<name>)...) or (?('name')...) to test for a used 2126subpattern by name. For compatibility with earlier versions of PCRE, which had 2127this facility before Perl, the syntax (?(name)...) is also recognized. However, 2128there is a possible ambiguity with this syntax, because subpattern names may 2129consist entirely of digits. PCRE looks first for a named subpattern; if it 2130cannot find one and the name consists entirely of digits, PCRE looks for a 2131subpattern of that number, which must be greater than zero. Using subpattern 2132names that consist entirely of digits is not recommended. 2133</P> 2134<P> 2135Rewriting the above example to use a named subpattern gives this: 2136<pre> 2137 (?<OPEN> \( )? [^()]+ (?(<OPEN>) \) ) 2138</pre> 2139If the name used in a condition of this kind is a duplicate, the test is 2140applied to all subpatterns of the same name, and is true if any one of them has 2141matched. 2142</P> 2143<br><b> 2144Checking for pattern recursion 2145</b><br> 2146<P> 2147If the condition is the string (R), and there is no subpattern with the name R, 2148the condition is true if a recursive call to the whole pattern or any 2149subpattern has been made. If digits or a name preceded by ampersand follow the 2150letter R, for example: 2151<pre> 2152 (?(R3)...) or (?(R&name)...) 2153</pre> 2154the condition is true if the most recent recursion is into a subpattern whose 2155number or name is given. This condition does not check the entire recursion 2156stack. If the name used in a condition of this kind is a duplicate, the test is 2157applied to all subpatterns of the same name, and is true if any one of them is 2158the most recent recursion. 2159</P> 2160<P> 2161At "top level", all these recursion test conditions are false. 2162<a href="#recursion">The syntax for recursive patterns</a> 2163is described below. 2164<a name="subdefine"></a></P> 2165<br><b> 2166Defining subpatterns for use by reference only 2167</b><br> 2168<P> 2169If the condition is the string (DEFINE), and there is no subpattern with the 2170name DEFINE, the condition is always false. In this case, there may be only one 2171alternative in the subpattern. It is always skipped if control reaches this 2172point in the pattern; the idea of DEFINE is that it can be used to define 2173subroutines that can be referenced from elsewhere. (The use of 2174<a href="#subpatternsassubroutines">subroutines</a> 2175is described below.) For example, a pattern to match an IPv4 address such as 2176"192.168.23.245" could be written like this (ignore white space and line 2177breaks): 2178<pre> 2179 (?(DEFINE) (?<byte> 2[0-4]\d | 25[0-5] | 1\d\d | [1-9]?\d) ) 2180 \b (?&byte) (\.(?&byte)){3} \b 2181</pre> 2182The first part of the pattern is a DEFINE group inside which a another group 2183named "byte" is defined. This matches an individual component of an IPv4 2184address (a number less than 256). When matching takes place, this part of the 2185pattern is skipped because DEFINE acts like a false condition. The rest of the 2186pattern uses references to the named group to match the four dot-separated 2187components of an IPv4 address, insisting on a word boundary at each end. 2188</P> 2189<br><b> 2190Assertion conditions 2191</b><br> 2192<P> 2193If the condition is not in any of the above formats, it must be an assertion. 2194This may be a positive or negative lookahead or lookbehind assertion. Consider 2195this pattern, again containing non-significant white space, and with the two 2196alternatives on the second line: 2197<pre> 2198 (?(?=[^a-z]*[a-z]) 2199 \d{2}-[a-z]{3}-\d{2} | \d{2}-\d{2}-\d{2} ) 2200</pre> 2201The condition is a positive lookahead assertion that matches an optional 2202sequence of non-letters followed by a letter. In other words, it tests for the 2203presence of at least one letter in the subject. If a letter is found, the 2204subject is matched against the first alternative; otherwise it is matched 2205against the second. This pattern matches strings in one of the two forms 2206dd-aaa-dd or dd-dd-dd, where aaa are letters and dd are digits. 2207<a name="comments"></a></P> 2208<br><a name="SEC20" href="#TOC1">COMMENTS</a><br> 2209<P> 2210There are two ways of including comments in patterns that are processed by 2211PCRE. In both cases, the start of the comment must not be in a character class, 2212nor in the middle of any other sequence of related characters such as (?: or a 2213subpattern name or number. The characters that make up a comment play no part 2214in the pattern matching. 2215</P> 2216<P> 2217The sequence (?# marks the start of a comment that continues up to the next 2218closing parenthesis. Nested parentheses are not permitted. If the PCRE_EXTENDED 2219option is set, an unescaped # character also introduces a comment, which in 2220this case continues to immediately after the next newline character or 2221character sequence in the pattern. Which characters are interpreted as newlines 2222is controlled by the options passed to a compiling function or by a special 2223sequence at the start of the pattern, as described in the section entitled 2224<a href="#newlines">"Newline conventions"</a> 2225above. Note that the end of this type of comment is a literal newline sequence 2226in the pattern; escape sequences that happen to represent a newline do not 2227count. For example, consider this pattern when PCRE_EXTENDED is set, and the 2228default newline convention is in force: 2229<pre> 2230 abc #comment \n still comment 2231</pre> 2232On encountering the # character, <b>pcre_compile()</b> skips along, looking for 2233a newline in the pattern. The sequence \n is still literal at this stage, so 2234it does not terminate the comment. Only an actual character with the code value 22350x0a (the default newline) does so. 2236<a name="recursion"></a></P> 2237<br><a name="SEC21" href="#TOC1">RECURSIVE PATTERNS</a><br> 2238<P> 2239Consider the problem of matching a string in parentheses, allowing for 2240unlimited nested parentheses. Without the use of recursion, the best that can 2241be done is to use a pattern that matches up to some fixed depth of nesting. It 2242is not possible to handle an arbitrary nesting depth. 2243</P> 2244<P> 2245For some time, Perl has provided a facility that allows regular expressions to 2246recurse (amongst other things). It does this by interpolating Perl code in the 2247expression at run time, and the code can refer to the expression itself. A Perl 2248pattern using code interpolation to solve the parentheses problem can be 2249created like this: 2250<pre> 2251 $re = qr{\( (?: (?>[^()]+) | (?p{$re}) )* \)}x; 2252</pre> 2253The (?p{...}) item interpolates Perl code at run time, and in this case refers 2254recursively to the pattern in which it appears. 2255</P> 2256<P> 2257Obviously, PCRE cannot support the interpolation of Perl code. Instead, it 2258supports special syntax for recursion of the entire pattern, and also for 2259individual subpattern recursion. After its introduction in PCRE and Python, 2260this kind of recursion was subsequently introduced into Perl at release 5.10. 2261</P> 2262<P> 2263A special item that consists of (? followed by a number greater than zero and a 2264closing parenthesis is a recursive subroutine call of the subpattern of the 2265given number, provided that it occurs inside that subpattern. (If not, it is a 2266<a href="#subpatternsassubroutines">non-recursive subroutine</a> 2267call, which is described in the next section.) The special item (?R) or (?0) is 2268a recursive call of the entire regular expression. 2269</P> 2270<P> 2271This PCRE pattern solves the nested parentheses problem (assume the 2272PCRE_EXTENDED option is set so that white space is ignored): 2273<pre> 2274 \( ( [^()]++ | (?R) )* \) 2275</pre> 2276First it matches an opening parenthesis. Then it matches any number of 2277substrings which can either be a sequence of non-parentheses, or a recursive 2278match of the pattern itself (that is, a correctly parenthesized substring). 2279Finally there is a closing parenthesis. Note the use of a possessive quantifier 2280to avoid backtracking into sequences of non-parentheses. 2281</P> 2282<P> 2283If this were part of a larger pattern, you would not want to recurse the entire 2284pattern, so instead you could use this: 2285<pre> 2286 ( \( ( [^()]++ | (?1) )* \) ) 2287</pre> 2288We have put the pattern into parentheses, and caused the recursion to refer to 2289them instead of the whole pattern. 2290</P> 2291<P> 2292In a larger pattern, keeping track of parenthesis numbers can be tricky. This 2293is made easier by the use of relative references. Instead of (?1) in the 2294pattern above you can write (?-2) to refer to the second most recently opened 2295parentheses preceding the recursion. In other words, a negative number counts 2296capturing parentheses leftwards from the point at which it is encountered. 2297</P> 2298<P> 2299It is also possible to refer to subsequently opened parentheses, by writing 2300references such as (?+2). However, these cannot be recursive because the 2301reference is not inside the parentheses that are referenced. They are always 2302<a href="#subpatternsassubroutines">non-recursive subroutine</a> 2303calls, as described in the next section. 2304</P> 2305<P> 2306An alternative approach is to use named parentheses instead. The Perl syntax 2307for this is (?&name); PCRE's earlier syntax (?P>name) is also supported. We 2308could rewrite the above example as follows: 2309<pre> 2310 (?<pn> \( ( [^()]++ | (?&pn) )* \) ) 2311</pre> 2312If there is more than one subpattern with the same name, the earliest one is 2313used. 2314</P> 2315<P> 2316This particular example pattern that we have been looking at contains nested 2317unlimited repeats, and so the use of a possessive quantifier for matching 2318strings of non-parentheses is important when applying the pattern to strings 2319that do not match. For example, when this pattern is applied to 2320<pre> 2321 (aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa() 2322</pre> 2323it yields "no match" quickly. However, if a possessive quantifier is not used, 2324the match runs for a very long time indeed because there are so many different 2325ways the + and * repeats can carve up the subject, and all have to be tested 2326before failure can be reported. 2327</P> 2328<P> 2329At the end of a match, the values of capturing parentheses are those from 2330the outermost level. If you want to obtain intermediate values, a callout 2331function can be used (see below and the 2332<a href="pcrecallout.html"><b>pcrecallout</b></a> 2333documentation). If the pattern above is matched against 2334<pre> 2335 (ab(cd)ef) 2336</pre> 2337the value for the inner capturing parentheses (numbered 2) is "ef", which is 2338the last value taken on at the top level. If a capturing subpattern is not 2339matched at the top level, its final captured value is unset, even if it was 2340(temporarily) set at a deeper level during the matching process. 2341</P> 2342<P> 2343If there are more than 15 capturing parentheses in a pattern, PCRE has to 2344obtain extra memory to store data during a recursion, which it does by using 2345<b>pcre_malloc</b>, freeing it via <b>pcre_free</b> afterwards. If no memory can 2346be obtained, the match fails with the PCRE_ERROR_NOMEMORY error. 2347</P> 2348<P> 2349Do not confuse the (?R) item with the condition (R), which tests for recursion. 2350Consider this pattern, which matches text in angle brackets, allowing for 2351arbitrary nesting. Only digits are allowed in nested brackets (that is, when 2352recursing), whereas any characters are permitted at the outer level. 2353<pre> 2354 < (?: (?(R) \d++ | [^<>]*+) | (?R)) * > 2355</pre> 2356In this pattern, (?(R) is the start of a conditional subpattern, with two 2357different alternatives for the recursive and non-recursive cases. The (?R) item 2358is the actual recursive call. 2359<a name="recursiondifference"></a></P> 2360<br><b> 2361Differences in recursion processing between PCRE and Perl 2362</b><br> 2363<P> 2364Recursion processing in PCRE differs from Perl in two important ways. In PCRE 2365(like Python, but unlike Perl), a recursive subpattern call is always treated 2366as an atomic group. That is, once it has matched some of the subject string, it 2367is never re-entered, even if it contains untried alternatives and there is a 2368subsequent matching failure. This can be illustrated by the following pattern, 2369which purports to match a palindromic string that contains an odd number of 2370characters (for example, "a", "aba", "abcba", "abcdcba"): 2371<pre> 2372 ^(.|(.)(?1)\2)$ 2373</pre> 2374The idea is that it either matches a single character, or two identical 2375characters surrounding a sub-palindrome. In Perl, this pattern works; in PCRE 2376it does not if the pattern is longer than three characters. Consider the 2377subject string "abcba": 2378</P> 2379<P> 2380At the top level, the first character is matched, but as it is not at the end 2381of the string, the first alternative fails; the second alternative is taken 2382and the recursion kicks in. The recursive call to subpattern 1 successfully 2383matches the next character ("b"). (Note that the beginning and end of line 2384tests are not part of the recursion). 2385</P> 2386<P> 2387Back at the top level, the next character ("c") is compared with what 2388subpattern 2 matched, which was "a". This fails. Because the recursion is 2389treated as an atomic group, there are now no backtracking points, and so the 2390entire match fails. (Perl is able, at this point, to re-enter the recursion and 2391try the second alternative.) However, if the pattern is written with the 2392alternatives in the other order, things are different: 2393<pre> 2394 ^((.)(?1)\2|.)$ 2395</pre> 2396This time, the recursing alternative is tried first, and continues to recurse 2397until it runs out of characters, at which point the recursion fails. But this 2398time we do have another alternative to try at the higher level. That is the big 2399difference: in the previous case the remaining alternative is at a deeper 2400recursion level, which PCRE cannot use. 2401</P> 2402<P> 2403To change the pattern so that it matches all palindromic strings, not just 2404those with an odd number of characters, it is tempting to change the pattern to 2405this: 2406<pre> 2407 ^((.)(?1)\2|.?)$ 2408</pre> 2409Again, this works in Perl, but not in PCRE, and for the same reason. When a 2410deeper recursion has matched a single character, it cannot be entered again in 2411order to match an empty string. The solution is to separate the two cases, and 2412write out the odd and even cases as alternatives at the higher level: 2413<pre> 2414 ^(?:((.)(?1)\2|)|((.)(?3)\4|.)) 2415</pre> 2416If you want to match typical palindromic phrases, the pattern has to ignore all 2417non-word characters, which can be done like this: 2418<pre> 2419 ^\W*+(?:((.)\W*+(?1)\W*+\2|)|((.)\W*+(?3)\W*+\4|\W*+.\W*+))\W*+$ 2420</pre> 2421If run with the PCRE_CASELESS option, this pattern matches phrases such as "A 2422man, a plan, a canal: Panama!" and it works well in both PCRE and Perl. Note 2423the use of the possessive quantifier *+ to avoid backtracking into sequences of 2424non-word characters. Without this, PCRE takes a great deal longer (ten times or 2425more) to match typical phrases, and Perl takes so long that you think it has 2426gone into a loop. 2427</P> 2428<P> 2429<b>WARNING</b>: The palindrome-matching patterns above work only if the subject 2430string does not start with a palindrome that is shorter than the entire string. 2431For example, although "abcba" is correctly matched, if the subject is "ababa", 2432PCRE finds the palindrome "aba" at the start, then fails at top level because 2433the end of the string does not follow. Once again, it cannot jump back into the 2434recursion to try other alternatives, so the entire match fails. 2435</P> 2436<P> 2437The second way in which PCRE and Perl differ in their recursion processing is 2438in the handling of captured values. In Perl, when a subpattern is called 2439recursively or as a subpattern (see the next section), it has no access to any 2440values that were captured outside the recursion, whereas in PCRE these values 2441can be referenced. Consider this pattern: 2442<pre> 2443 ^(.)(\1|a(?2)) 2444</pre> 2445In PCRE, this pattern matches "bab". The first capturing parentheses match "b", 2446then in the second group, when the back reference \1 fails to match "b", the 2447second alternative matches "a" and then recurses. In the recursion, \1 does 2448now match "b" and so the whole match succeeds. In Perl, the pattern fails to 2449match because inside the recursive call \1 cannot access the externally set 2450value. 2451<a name="subpatternsassubroutines"></a></P> 2452<br><a name="SEC22" href="#TOC1">SUBPATTERNS AS SUBROUTINES</a><br> 2453<P> 2454If the syntax for a recursive subpattern call (either by number or by 2455name) is used outside the parentheses to which it refers, it operates like a 2456subroutine in a programming language. The called subpattern may be defined 2457before or after the reference. A numbered reference can be absolute or 2458relative, as in these examples: 2459<pre> 2460 (...(absolute)...)...(?2)... 2461 (...(relative)...)...(?-1)... 2462 (...(?+1)...(relative)... 2463</pre> 2464An earlier example pointed out that the pattern 2465<pre> 2466 (sens|respons)e and \1ibility 2467</pre> 2468matches "sense and sensibility" and "response and responsibility", but not 2469"sense and responsibility". If instead the pattern 2470<pre> 2471 (sens|respons)e and (?1)ibility 2472</pre> 2473is used, it does match "sense and responsibility" as well as the other two 2474strings. Another example is given in the discussion of DEFINE above. 2475</P> 2476<P> 2477All subroutine calls, whether recursive or not, are always treated as atomic 2478groups. That is, once a subroutine has matched some of the subject string, it 2479is never re-entered, even if it contains untried alternatives and there is a 2480subsequent matching failure. Any capturing parentheses that are set during the 2481subroutine call revert to their previous values afterwards. 2482</P> 2483<P> 2484Processing options such as case-independence are fixed when a subpattern is 2485defined, so if it is used as a subroutine, such options cannot be changed for 2486different calls. For example, consider this pattern: 2487<pre> 2488 (abc)(?i:(?-1)) 2489</pre> 2490It matches "abcabc". It does not match "abcABC" because the change of 2491processing option does not affect the called subpattern. 2492<a name="onigurumasubroutines"></a></P> 2493<br><a name="SEC23" href="#TOC1">ONIGURUMA SUBROUTINE SYNTAX</a><br> 2494<P> 2495For compatibility with Oniguruma, the non-Perl syntax \g followed by a name or 2496a number enclosed either in angle brackets or single quotes, is an alternative 2497syntax for referencing a subpattern as a subroutine, possibly recursively. Here 2498are two of the examples used above, rewritten using this syntax: 2499<pre> 2500 (?<pn> \( ( (?>[^()]+) | \g<pn> )* \) ) 2501 (sens|respons)e and \g'1'ibility 2502</pre> 2503PCRE supports an extension to Oniguruma: if a number is preceded by a 2504plus or a minus sign it is taken as a relative reference. For example: 2505<pre> 2506 (abc)(?i:\g<-1>) 2507</pre> 2508Note that \g{...} (Perl syntax) and \g<...> (Oniguruma syntax) are <i>not</i> 2509synonymous. The former is a back reference; the latter is a subroutine call. 2510</P> 2511<br><a name="SEC24" href="#TOC1">CALLOUTS</a><br> 2512<P> 2513Perl has a feature whereby using the sequence (?{...}) causes arbitrary Perl 2514code to be obeyed in the middle of matching a regular expression. This makes it 2515possible, amongst other things, to extract different substrings that match the 2516same pair of parentheses when there is a repetition. 2517</P> 2518<P> 2519PCRE provides a similar feature, but of course it cannot obey arbitrary Perl 2520code. The feature is called "callout". The caller of PCRE provides an external 2521function by putting its entry point in the global variable <i>pcre_callout</i> 2522(8-bit library) or <i>pcre16_callout</i> (16-bit library). By default, this 2523variable contains NULL, which disables all calling out. 2524</P> 2525<P> 2526Within a regular expression, (?C) indicates the points at which the external 2527function is to be called. If you want to identify different callout points, you 2528can put a number less than 256 after the letter C. The default value is zero. 2529For example, this pattern has two callout points: 2530<pre> 2531 (?C1)abc(?C2)def 2532</pre> 2533If the PCRE_AUTO_CALLOUT flag is passed to a compiling function, callouts are 2534automatically installed before each item in the pattern. They are all numbered 2535255. 2536</P> 2537<P> 2538During matching, when PCRE reaches a callout point, the external function is 2539called. It is provided with the number of the callout, the position in the 2540pattern, and, optionally, one item of data originally supplied by the caller of 2541the matching function. The callout function may cause matching to proceed, to 2542backtrack, or to fail altogether. A complete description of the interface to 2543the callout function is given in the 2544<a href="pcrecallout.html"><b>pcrecallout</b></a> 2545documentation. 2546<a name="backtrackcontrol"></a></P> 2547<br><a name="SEC25" href="#TOC1">BACKTRACKING CONTROL</a><br> 2548<P> 2549Perl 5.10 introduced a number of "Special Backtracking Control Verbs", which 2550are described in the Perl documentation as "experimental and subject to change 2551or removal in a future version of Perl". It goes on to say: "Their usage in 2552production code should be noted to avoid problems during upgrades." The same 2553remarks apply to the PCRE features described in this section. 2554</P> 2555<P> 2556Since these verbs are specifically related to backtracking, most of them can be 2557used only when the pattern is to be matched using one of the traditional 2558matching functions, which use a backtracking algorithm. With the exception of 2559(*FAIL), which behaves like a failing negative assertion, they cause an error 2560if encountered by a DFA matching function. 2561</P> 2562<P> 2563If any of these verbs are used in an assertion or in a subpattern that is 2564called as a subroutine (whether or not recursively), their effect is confined 2565to that subpattern; it does not extend to the surrounding pattern, with one 2566exception: the name from a *(MARK), (*PRUNE), or (*THEN) that is encountered in 2567a successful positive assertion <i>is</i> passed back when a match succeeds 2568(compare capturing parentheses in assertions). Note that such subpatterns are 2569processed as anchored at the point where they are tested. Note also that Perl's 2570treatment of subroutines and assertions is different in some cases. 2571</P> 2572<P> 2573The new verbs make use of what was previously invalid syntax: an opening 2574parenthesis followed by an asterisk. They are generally of the form 2575(*VERB) or (*VERB:NAME). Some may take either form, with differing behaviour, 2576depending on whether or not an argument is present. A name is any sequence of 2577characters that does not include a closing parenthesis. The maximum length of 2578name is 255 in the 8-bit library and 65535 in the 16-bit library. If the name 2579is empty, that is, if the closing parenthesis immediately follows the colon, 2580the effect is as if the colon were not there. Any number of these verbs may 2581occur in a pattern. 2582<a name="nooptimize"></a></P> 2583<br><b> 2584Optimizations that affect backtracking verbs 2585</b><br> 2586<P> 2587PCRE contains some optimizations that are used to speed up matching by running 2588some checks at the start of each match attempt. For example, it may know the 2589minimum length of matching subject, or that a particular character must be 2590present. When one of these optimizations suppresses the running of a match, any 2591included backtracking verbs will not, of course, be processed. You can suppress 2592the start-of-match optimizations by setting the PCRE_NO_START_OPTIMIZE option 2593when calling <b>pcre_compile()</b> or <b>pcre_exec()</b>, or by starting the 2594pattern with (*NO_START_OPT). There is more discussion of this option in the 2595section entitled 2596<a href="pcreapi.html#execoptions">"Option bits for <b>pcre_exec()</b>"</a> 2597in the 2598<a href="pcreapi.html"><b>pcreapi</b></a> 2599documentation. 2600</P> 2601<P> 2602Experiments with Perl suggest that it too has similar optimizations, sometimes 2603leading to anomalous results. 2604</P> 2605<br><b> 2606Verbs that act immediately 2607</b><br> 2608<P> 2609The following verbs act as soon as they are encountered. They may not be 2610followed by a name. 2611<pre> 2612 (*ACCEPT) 2613</pre> 2614This verb causes the match to end successfully, skipping the remainder of the 2615pattern. However, when it is inside a subpattern that is called as a 2616subroutine, only that subpattern is ended successfully. Matching then continues 2617at the outer level. If (*ACCEPT) is inside capturing parentheses, the data so 2618far is captured. For example: 2619<pre> 2620 A((?:A|B(*ACCEPT)|C)D) 2621</pre> 2622This matches "AB", "AAD", or "ACD"; when it matches "AB", "B" is captured by 2623the outer parentheses. 2624<pre> 2625 (*FAIL) or (*F) 2626</pre> 2627This verb causes a matching failure, forcing backtracking to occur. It is 2628equivalent to (?!) but easier to read. The Perl documentation notes that it is 2629probably useful only when combined with (?{}) or (??{}). Those are, of course, 2630Perl features that are not present in PCRE. The nearest equivalent is the 2631callout feature, as for example in this pattern: 2632<pre> 2633 a+(?C)(*FAIL) 2634</pre> 2635A match with the string "aaaa" always fails, but the callout is taken before 2636each backtrack happens (in this example, 10 times). 2637</P> 2638<br><b> 2639Recording which path was taken 2640</b><br> 2641<P> 2642There is one verb whose main purpose is to track how a match was arrived at, 2643though it also has a secondary use in conjunction with advancing the match 2644starting point (see (*SKIP) below). 2645<pre> 2646 (*MARK:NAME) or (*:NAME) 2647</pre> 2648A name is always required with this verb. There may be as many instances of 2649(*MARK) as you like in a pattern, and their names do not have to be unique. 2650</P> 2651<P> 2652When a match succeeds, the name of the last-encountered (*MARK) on the matching 2653path is passed back to the caller as described in the section entitled 2654<a href="pcreapi.html#extradata">"Extra data for <b>pcre_exec()</b>"</a> 2655in the 2656<a href="pcreapi.html"><b>pcreapi</b></a> 2657documentation. Here is an example of <b>pcretest</b> output, where the /K 2658modifier requests the retrieval and outputting of (*MARK) data: 2659<pre> 2660 re> /X(*MARK:A)Y|X(*MARK:B)Z/K 2661 data> XY 2662 0: XY 2663 MK: A 2664 XZ 2665 0: XZ 2666 MK: B 2667</pre> 2668The (*MARK) name is tagged with "MK:" in this output, and in this example it 2669indicates which of the two alternatives matched. This is a more efficient way 2670of obtaining this information than putting each alternative in its own 2671capturing parentheses. 2672</P> 2673<P> 2674If (*MARK) is encountered in a positive assertion, its name is recorded and 2675passed back if it is the last-encountered. This does not happen for negative 2676assertions. 2677</P> 2678<P> 2679After a partial match or a failed match, the name of the last encountered 2680(*MARK) in the entire match process is returned. For example: 2681<pre> 2682 re> /X(*MARK:A)Y|X(*MARK:B)Z/K 2683 data> XP 2684 No match, mark = B 2685</pre> 2686Note that in this unanchored example the mark is retained from the match 2687attempt that started at the letter "X" in the subject. Subsequent match 2688attempts starting at "P" and then with an empty string do not get as far as the 2689(*MARK) item, but nevertheless do not reset it. 2690</P> 2691<P> 2692If you are interested in (*MARK) values after failed matches, you should 2693probably set the PCRE_NO_START_OPTIMIZE option 2694<a href="#nooptimize">(see above)</a> 2695to ensure that the match is always attempted. 2696</P> 2697<br><b> 2698Verbs that act after backtracking 2699</b><br> 2700<P> 2701The following verbs do nothing when they are encountered. Matching continues 2702with what follows, but if there is no subsequent match, causing a backtrack to 2703the verb, a failure is forced. That is, backtracking cannot pass to the left of 2704the verb. However, when one of these verbs appears inside an atomic group, its 2705effect is confined to that group, because once the group has been matched, 2706there is never any backtracking into it. In this situation, backtracking can 2707"jump back" to the left of the entire atomic group. (Remember also, as stated 2708above, that this localization also applies in subroutine calls and assertions.) 2709</P> 2710<P> 2711These verbs differ in exactly what kind of failure occurs when backtracking 2712reaches them. 2713<pre> 2714 (*COMMIT) 2715</pre> 2716This verb, which may not be followed by a name, causes the whole match to fail 2717outright if the rest of the pattern does not match. Even if the pattern is 2718unanchored, no further attempts to find a match by advancing the starting point 2719take place. Once (*COMMIT) has been passed, <b>pcre_exec()</b> is committed to 2720finding a match at the current starting point, or not at all. For example: 2721<pre> 2722 a+(*COMMIT)b 2723</pre> 2724This matches "xxaab" but not "aacaab". It can be thought of as a kind of 2725dynamic anchor, or "I've started, so I must finish." The name of the most 2726recently passed (*MARK) in the path is passed back when (*COMMIT) forces a 2727match failure. 2728</P> 2729<P> 2730Note that (*COMMIT) at the start of a pattern is not the same as an anchor, 2731unless PCRE's start-of-match optimizations are turned off, as shown in this 2732<b>pcretest</b> example: 2733<pre> 2734 re> /(*COMMIT)abc/ 2735 data> xyzabc 2736 0: abc 2737 xyzabc\Y 2738 No match 2739</pre> 2740PCRE knows that any match must start with "a", so the optimization skips along 2741the subject to "a" before running the first match attempt, which succeeds. When 2742the optimization is disabled by the \Y escape in the second subject, the match 2743starts at "x" and so the (*COMMIT) causes it to fail without trying any other 2744starting points. 2745<pre> 2746 (*PRUNE) or (*PRUNE:NAME) 2747</pre> 2748This verb causes the match to fail at the current starting position in the 2749subject if the rest of the pattern does not match. If the pattern is 2750unanchored, the normal "bumpalong" advance to the next starting character then 2751happens. Backtracking can occur as usual to the left of (*PRUNE), before it is 2752reached, or when matching to the right of (*PRUNE), but if there is no match to 2753the right, backtracking cannot cross (*PRUNE). In simple cases, the use of 2754(*PRUNE) is just an alternative to an atomic group or possessive quantifier, 2755but there are some uses of (*PRUNE) that cannot be expressed in any other way. 2756The behaviour of (*PRUNE:NAME) is the same as (*MARK:NAME)(*PRUNE). In an 2757anchored pattern (*PRUNE) has the same effect as (*COMMIT). 2758<pre> 2759 (*SKIP) 2760</pre> 2761This verb, when given without a name, is like (*PRUNE), except that if the 2762pattern is unanchored, the "bumpalong" advance is not to the next character, 2763but to the position in the subject where (*SKIP) was encountered. (*SKIP) 2764signifies that whatever text was matched leading up to it cannot be part of a 2765successful match. Consider: 2766<pre> 2767 a+(*SKIP)b 2768</pre> 2769If the subject is "aaaac...", after the first match attempt fails (starting at 2770the first character in the string), the starting point skips on to start the 2771next attempt at "c". Note that a possessive quantifer does not have the same 2772effect as this example; although it would suppress backtracking during the 2773first match attempt, the second attempt would start at the second character 2774instead of skipping on to "c". 2775<pre> 2776 (*SKIP:NAME) 2777</pre> 2778When (*SKIP) has an associated name, its behaviour is modified. If the 2779following pattern fails to match, the previous path through the pattern is 2780searched for the most recent (*MARK) that has the same name. If one is found, 2781the "bumpalong" advance is to the subject position that corresponds to that 2782(*MARK) instead of to where (*SKIP) was encountered. If no (*MARK) with a 2783matching name is found, the (*SKIP) is ignored. 2784<pre> 2785 (*THEN) or (*THEN:NAME) 2786</pre> 2787This verb causes a skip to the next innermost alternative if the rest of the 2788pattern does not match. That is, it cancels pending backtracking, but only 2789within the current alternative. Its name comes from the observation that it can 2790be used for a pattern-based if-then-else block: 2791<pre> 2792 ( COND1 (*THEN) FOO | COND2 (*THEN) BAR | COND3 (*THEN) BAZ ) ... 2793</pre> 2794If the COND1 pattern matches, FOO is tried (and possibly further items after 2795the end of the group if FOO succeeds); on failure, the matcher skips to the 2796second alternative and tries COND2, without backtracking into COND1. The 2797behaviour of (*THEN:NAME) is exactly the same as (*MARK:NAME)(*THEN). 2798If (*THEN) is not inside an alternation, it acts like (*PRUNE). 2799</P> 2800<P> 2801Note that a subpattern that does not contain a | character is just a part of 2802the enclosing alternative; it is not a nested alternation with only one 2803alternative. The effect of (*THEN) extends beyond such a subpattern to the 2804enclosing alternative. Consider this pattern, where A, B, etc. are complex 2805pattern fragments that do not contain any | characters at this level: 2806<pre> 2807 A (B(*THEN)C) | D 2808</pre> 2809If A and B are matched, but there is a failure in C, matching does not 2810backtrack into A; instead it moves to the next alternative, that is, D. 2811However, if the subpattern containing (*THEN) is given an alternative, it 2812behaves differently: 2813<pre> 2814 A (B(*THEN)C | (*FAIL)) | D 2815</pre> 2816The effect of (*THEN) is now confined to the inner subpattern. After a failure 2817in C, matching moves to (*FAIL), which causes the whole subpattern to fail 2818because there are no more alternatives to try. In this case, matching does now 2819backtrack into A. 2820</P> 2821<P> 2822Note also that a conditional subpattern is not considered as having two 2823alternatives, because only one is ever used. In other words, the | character in 2824a conditional subpattern has a different meaning. Ignoring white space, 2825consider: 2826<pre> 2827 ^.*? (?(?=a) a | b(*THEN)c ) 2828</pre> 2829If the subject is "ba", this pattern does not match. Because .*? is ungreedy, 2830it initially matches zero characters. The condition (?=a) then fails, the 2831character "b" is matched, but "c" is not. At this point, matching does not 2832backtrack to .*? as might perhaps be expected from the presence of the | 2833character. The conditional subpattern is part of the single alternative that 2834comprises the whole pattern, and so the match fails. (If there was a backtrack 2835into .*?, allowing it to match "b", the match would succeed.) 2836</P> 2837<P> 2838The verbs just described provide four different "strengths" of control when 2839subsequent matching fails. (*THEN) is the weakest, carrying on the match at the 2840next alternative. (*PRUNE) comes next, failing the match at the current 2841starting position, but allowing an advance to the next character (for an 2842unanchored pattern). (*SKIP) is similar, except that the advance may be more 2843than one character. (*COMMIT) is the strongest, causing the entire match to 2844fail. 2845</P> 2846<P> 2847If more than one such verb is present in a pattern, the "strongest" one wins. 2848For example, consider this pattern, where A, B, etc. are complex pattern 2849fragments: 2850<pre> 2851 (A(*COMMIT)B(*THEN)C|D) 2852</pre> 2853Once A has matched, PCRE is committed to this match, at the current starting 2854position. If subsequently B matches, but C does not, the normal (*THEN) action 2855of trying the next alternative (that is, D) does not happen because (*COMMIT) 2856overrides. 2857</P> 2858<br><a name="SEC26" href="#TOC1">SEE ALSO</a><br> 2859<P> 2860<b>pcreapi</b>(3), <b>pcrecallout</b>(3), <b>pcrematching</b>(3), 2861<b>pcresyntax</b>(3), <b>pcre</b>(3), <b>pcre16(3)</b>. 2862</P> 2863<br><a name="SEC27" href="#TOC1">AUTHOR</a><br> 2864<P> 2865Philip Hazel 2866<br> 2867University Computing Service 2868<br> 2869Cambridge CB2 3QH, England. 2870<br> 2871</P> 2872<br><a name="SEC28" href="#TOC1">REVISION</a><br> 2873<P> 2874Last updated: 17 June 2012 2875<br> 2876Copyright © 1997-2012 University of Cambridge. 2877<br> 2878<p> 2879Return to the <a href="index.html">PCRE index page</a>. 2880</p> 2881