5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #define RF_tainted 1 /* tainted information used? */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
97 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
103 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
104 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
106 #define HOPc(pos,off) \
107 (char *)(PL_reg_match_utf8 \
108 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
110 #define HOPBACKc(pos, off) \
111 (char*)(PL_reg_match_utf8\
112 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
113 : (pos - off >= PL_bostr) \
117 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
118 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
120 /* these are unrolled below in the CCC_TRY_XXX defined */
121 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
122 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
124 /* Doesn't do an assert to verify that is correct */
125 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
126 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)" "); LEAVE; } } STMT_END
128 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
129 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
130 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
132 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
133 LOAD_UTF8_CHARCLASS(X_begin, " "); \
134 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
135 /* These are utf8 constants, and not utf-ebcdic constants, so the \
136 * assert should likely and hopefully fail on an EBCDIC machine */ \
137 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
139 /* No asserts are done for these, in case called on an early \
140 * Unicode version in which they map to nothing */ \
141 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
142 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
143 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
144 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
145 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
146 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
147 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
150 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
151 so that it is possible to override the option here without having to
152 rebuild the entire core. as we are required to do if we change regcomp.h
153 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
155 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
156 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
159 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
160 #define LOAD_UTF8_CHARCLASS_PERL_WORD() LOAD_UTF8_CHARCLASS_ALNUM()
161 #define LOAD_UTF8_CHARCLASS_PERL_SPACE() LOAD_UTF8_CHARCLASS_SPACE()
162 #define LOAD_UTF8_CHARCLASS_POSIX_DIGIT() LOAD_UTF8_CHARCLASS_DIGIT()
163 #define RE_utf8_perl_word PL_utf8_alnum
164 #define RE_utf8_perl_space PL_utf8_space
165 #define RE_utf8_posix_digit PL_utf8_digit
166 #define perl_word alnum
167 #define perl_space space
168 #define posix_digit digit
170 #define LOAD_UTF8_CHARCLASS_PERL_WORD() LOAD_UTF8_CHARCLASS(perl_word,"a")
171 #define LOAD_UTF8_CHARCLASS_PERL_SPACE() LOAD_UTF8_CHARCLASS(perl_space," ")
172 #define LOAD_UTF8_CHARCLASS_POSIX_DIGIT() LOAD_UTF8_CHARCLASS(posix_digit,"0")
173 #define RE_utf8_perl_word PL_utf8_perl_word
174 #define RE_utf8_perl_space PL_utf8_perl_space
175 #define RE_utf8_posix_digit PL_utf8_posix_digit
179 #define _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
181 PL_reg_flags |= RF_tainted; \
186 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
187 if (!CAT2(PL_utf8_,CLASS)) { \
191 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
195 if (!(OP(scan) == NAME \
196 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, utf8_target)) \
197 : LCFUNC_utf8((U8*)locinput))) \
201 locinput += PL_utf8skip[nextchr]; \
202 nextchr = UCHARAT(locinput); \
205 /* Drops through to the macro that calls this one */
207 #define CCC_TRY_AFF(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
208 _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
209 if (!(OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
211 nextchr = UCHARAT(++locinput); \
214 /* Almost identical to the above, but has a case for a node that matches chars
215 * between 128 and 255 using Unicode (latin1) semantics. */
216 #define CCC_TRY_AFF_U(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU,LCFUNC) \
217 _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
218 if (!(OP(scan) == NAMEL ? LCFUNC(nextchr) : (FUNCU(nextchr) && (isASCII(nextchr) || (FLAGS(scan) & USE_UNI))))) \
220 nextchr = UCHARAT(++locinput); \
223 #define _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
225 PL_reg_flags |= RF_tainted; \
228 if (!nextchr && locinput >= PL_regeol) \
230 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
231 if (!CAT2(PL_utf8_,CLASS)) { \
235 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
239 if ((OP(scan) == NAME \
240 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, utf8_target)) \
241 : LCFUNC_utf8((U8*)locinput))) \
245 locinput += PL_utf8skip[nextchr]; \
246 nextchr = UCHARAT(locinput); \
250 #define CCC_TRY_NEG(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
251 _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
252 if ((OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
254 nextchr = UCHARAT(++locinput); \
258 #define CCC_TRY_NEG_U(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU,LCFUNC) \
259 _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU) \
260 if ((OP(scan) == NAMEL ? LCFUNC(nextchr) : (FUNCU(nextchr) && (isASCII(nextchr) || (FLAGS(scan) & USE_UNI))))) \
262 nextchr = UCHARAT(++locinput); \
267 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
269 /* for use after a quantifier and before an EXACT-like node -- japhy */
270 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
272 * NOTE that *nothing* that affects backtracking should be in here, specifically
273 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
274 * node that is in between two EXACT like nodes when ascertaining what the required
275 * "follow" character is. This should probably be moved to regex compile time
276 * although it may be done at run time beause of the REF possibility - more
277 * investigation required. -- demerphq
279 #define JUMPABLE(rn) ( \
281 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
283 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
284 OP(rn) == PLUS || OP(rn) == MINMOD || \
286 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
288 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
290 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
293 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
294 we don't need this definition. */
295 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
296 #define IS_TEXTF(rn) ( OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
297 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
300 /* ... so we use this as its faster. */
301 #define IS_TEXT(rn) ( OP(rn)==EXACT )
302 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
303 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
308 Search for mandatory following text node; for lookahead, the text must
309 follow but for lookbehind (rn->flags != 0) we skip to the next step.
311 #define FIND_NEXT_IMPT(rn) STMT_START { \
312 while (JUMPABLE(rn)) { \
313 const OPCODE type = OP(rn); \
314 if (type == SUSPEND || PL_regkind[type] == CURLY) \
315 rn = NEXTOPER(NEXTOPER(rn)); \
316 else if (type == PLUS) \
318 else if (type == IFMATCH) \
319 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
320 else rn += NEXT_OFF(rn); \
325 static void restore_pos(pTHX_ void *arg);
327 #define REGCP_PAREN_ELEMS 4
328 #define REGCP_OTHER_ELEMS 5
329 #define REGCP_FRAME_ELEMS 1
330 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
331 * are needed for the regexp context stack bookkeeping. */
334 S_regcppush(pTHX_ I32 parenfloor)
337 const int retval = PL_savestack_ix;
338 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
339 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
340 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
342 GET_RE_DEBUG_FLAGS_DECL;
344 if (paren_elems_to_push < 0)
345 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
347 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
348 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
349 " out of range (%lu-%ld)",
350 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
352 SSGROW(total_elems + REGCP_FRAME_ELEMS);
354 for (p = PL_regsize; p > parenfloor; p--) {
355 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
356 SSPUSHINT(PL_regoffs[p].end);
357 SSPUSHINT(PL_regoffs[p].start);
358 SSPUSHPTR(PL_reg_start_tmp[p]);
360 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
361 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
362 (UV)p, (IV)PL_regoffs[p].start,
363 (IV)(PL_reg_start_tmp[p] - PL_bostr),
364 (IV)PL_regoffs[p].end
367 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
368 SSPUSHPTR(PL_regoffs);
369 SSPUSHINT(PL_regsize);
370 SSPUSHINT(*PL_reglastparen);
371 SSPUSHINT(*PL_reglastcloseparen);
372 SSPUSHPTR(PL_reginput);
373 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
378 /* These are needed since we do not localize EVAL nodes: */
379 #define REGCP_SET(cp) \
381 PerlIO_printf(Perl_debug_log, \
382 " Setting an EVAL scope, savestack=%"IVdf"\n", \
383 (IV)PL_savestack_ix)); \
386 #define REGCP_UNWIND(cp) \
388 if (cp != PL_savestack_ix) \
389 PerlIO_printf(Perl_debug_log, \
390 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
391 (IV)(cp), (IV)PL_savestack_ix)); \
395 S_regcppop(pTHX_ const regexp *rex)
400 GET_RE_DEBUG_FLAGS_DECL;
402 PERL_ARGS_ASSERT_REGCPPOP;
404 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
406 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
407 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
408 input = (char *) SSPOPPTR;
409 *PL_reglastcloseparen = SSPOPINT;
410 *PL_reglastparen = SSPOPINT;
411 PL_regsize = SSPOPINT;
412 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
414 i -= REGCP_OTHER_ELEMS;
415 /* Now restore the parentheses context. */
416 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
418 U32 paren = (U32)SSPOPINT;
419 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
420 PL_regoffs[paren].start = SSPOPINT;
422 if (paren <= *PL_reglastparen)
423 PL_regoffs[paren].end = tmps;
425 PerlIO_printf(Perl_debug_log,
426 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
427 (UV)paren, (IV)PL_regoffs[paren].start,
428 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
429 (IV)PL_regoffs[paren].end,
430 (paren > *PL_reglastparen ? "(no)" : ""));
434 if (*PL_reglastparen + 1 <= rex->nparens) {
435 PerlIO_printf(Perl_debug_log,
436 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
437 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
441 /* It would seem that the similar code in regtry()
442 * already takes care of this, and in fact it is in
443 * a better location to since this code can #if 0-ed out
444 * but the code in regtry() is needed or otherwise tests
445 * requiring null fields (pat.t#187 and split.t#{13,14}
446 * (as of patchlevel 7877) will fail. Then again,
447 * this code seems to be necessary or otherwise
448 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
449 * --jhi updated by dapm */
450 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
452 PL_regoffs[i].start = -1;
453 PL_regoffs[i].end = -1;
459 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
462 * pregexec and friends
465 #ifndef PERL_IN_XSUB_RE
467 - pregexec - match a regexp against a string
470 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
471 char *strbeg, I32 minend, SV *screamer, U32 nosave)
472 /* strend: pointer to null at end of string */
473 /* strbeg: real beginning of string */
474 /* minend: end of match must be >=minend after stringarg. */
475 /* nosave: For optimizations. */
477 PERL_ARGS_ASSERT_PREGEXEC;
480 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
481 nosave ? 0 : REXEC_COPY_STR);
486 * Need to implement the following flags for reg_anch:
488 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
490 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
491 * INTUIT_AUTORITATIVE_ML
492 * INTUIT_ONCE_NOML - Intuit can match in one location only.
495 * Another flag for this function: SECOND_TIME (so that float substrs
496 * with giant delta may be not rechecked).
499 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
501 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
502 Otherwise, only SvCUR(sv) is used to get strbeg. */
504 /* XXXX We assume that strpos is strbeg unless sv. */
506 /* XXXX Some places assume that there is a fixed substring.
507 An update may be needed if optimizer marks as "INTUITable"
508 RExen without fixed substrings. Similarly, it is assumed that
509 lengths of all the strings are no more than minlen, thus they
510 cannot come from lookahead.
511 (Or minlen should take into account lookahead.)
512 NOTE: Some of this comment is not correct. minlen does now take account
513 of lookahead/behind. Further research is required. -- demerphq
517 /* A failure to find a constant substring means that there is no need to make
518 an expensive call to REx engine, thus we celebrate a failure. Similarly,
519 finding a substring too deep into the string means that less calls to
520 regtry() should be needed.
522 REx compiler's optimizer found 4 possible hints:
523 a) Anchored substring;
525 c) Whether we are anchored (beginning-of-line or \G);
526 d) First node (of those at offset 0) which may distingush positions;
527 We use a)b)d) and multiline-part of c), and try to find a position in the
528 string which does not contradict any of them.
531 /* Most of decisions we do here should have been done at compile time.
532 The nodes of the REx which we used for the search should have been
533 deleted from the finite automaton. */
536 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
537 char *strend, const U32 flags, re_scream_pos_data *data)
540 struct regexp *const prog = (struct regexp *)SvANY(rx);
541 register I32 start_shift = 0;
542 /* Should be nonnegative! */
543 register I32 end_shift = 0;
548 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
550 register char *other_last = NULL; /* other substr checked before this */
551 char *check_at = NULL; /* check substr found at this pos */
552 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
553 RXi_GET_DECL(prog,progi);
555 const char * const i_strpos = strpos;
557 GET_RE_DEBUG_FLAGS_DECL;
559 PERL_ARGS_ASSERT_RE_INTUIT_START;
561 RX_MATCH_UTF8_set(rx,utf8_target);
564 PL_reg_flags |= RF_utf8;
567 debug_start_match(rx, utf8_target, strpos, strend,
568 sv ? "Guessing start of match in sv for"
569 : "Guessing start of match in string for");
572 /* CHR_DIST() would be more correct here but it makes things slow. */
573 if (prog->minlen > strend - strpos) {
574 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
575 "String too short... [re_intuit_start]\n"));
579 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
582 if (!prog->check_utf8 && prog->check_substr)
583 to_utf8_substr(prog);
584 check = prog->check_utf8;
586 if (!prog->check_substr && prog->check_utf8)
587 to_byte_substr(prog);
588 check = prog->check_substr;
590 if (check == &PL_sv_undef) {
591 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
592 "Non-utf8 string cannot match utf8 check string\n"));
595 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
596 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
597 || ( (prog->extflags & RXf_ANCH_BOL)
598 && !multiline ) ); /* Check after \n? */
601 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
602 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
603 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
605 && (strpos != strbeg)) {
606 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
609 if (prog->check_offset_min == prog->check_offset_max &&
610 !(prog->extflags & RXf_CANY_SEEN)) {
611 /* Substring at constant offset from beg-of-str... */
614 s = HOP3c(strpos, prog->check_offset_min, strend);
617 slen = SvCUR(check); /* >= 1 */
619 if ( strend - s > slen || strend - s < slen - 1
620 || (strend - s == slen && strend[-1] != '\n')) {
621 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
624 /* Now should match s[0..slen-2] */
626 if (slen && (*SvPVX_const(check) != *s
628 && memNE(SvPVX_const(check), s, slen)))) {
630 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
634 else if (*SvPVX_const(check) != *s
635 || ((slen = SvCUR(check)) > 1
636 && memNE(SvPVX_const(check), s, slen)))
639 goto success_at_start;
642 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
644 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
645 end_shift = prog->check_end_shift;
648 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
649 - (SvTAIL(check) != 0);
650 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
652 if (end_shift < eshift)
656 else { /* Can match at random position */
659 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
660 end_shift = prog->check_end_shift;
662 /* end shift should be non negative here */
665 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
667 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
668 (IV)end_shift, RX_PRECOMP(prog));
672 /* Find a possible match in the region s..strend by looking for
673 the "check" substring in the region corrected by start/end_shift. */
676 I32 srch_start_shift = start_shift;
677 I32 srch_end_shift = end_shift;
678 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
679 srch_end_shift -= ((strbeg - s) - srch_start_shift);
680 srch_start_shift = strbeg - s;
682 DEBUG_OPTIMISE_MORE_r({
683 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
684 (IV)prog->check_offset_min,
685 (IV)srch_start_shift,
687 (IV)prog->check_end_shift);
690 if (flags & REXEC_SCREAM) {
691 I32 p = -1; /* Internal iterator of scream. */
692 I32 * const pp = data ? data->scream_pos : &p;
694 if (PL_screamfirst[BmRARE(check)] >= 0
695 || ( BmRARE(check) == '\n'
696 && (BmPREVIOUS(check) == SvCUR(check) - 1)
698 s = screaminstr(sv, check,
699 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
702 /* we may be pointing at the wrong string */
703 if (s && RXp_MATCH_COPIED(prog))
704 s = strbeg + (s - SvPVX_const(sv));
706 *data->scream_olds = s;
711 if (prog->extflags & RXf_CANY_SEEN) {
712 start_point= (U8*)(s + srch_start_shift);
713 end_point= (U8*)(strend - srch_end_shift);
715 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
716 end_point= HOP3(strend, -srch_end_shift, strbeg);
718 DEBUG_OPTIMISE_MORE_r({
719 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
720 (int)(end_point - start_point),
721 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
725 s = fbm_instr( start_point, end_point,
726 check, multiline ? FBMrf_MULTILINE : 0);
729 /* Update the count-of-usability, remove useless subpatterns,
733 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
734 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
735 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
736 (s ? "Found" : "Did not find"),
737 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
738 ? "anchored" : "floating"),
741 (s ? " at offset " : "...\n") );
746 /* Finish the diagnostic message */
747 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
749 /* XXX dmq: first branch is for positive lookbehind...
750 Our check string is offset from the beginning of the pattern.
751 So we need to do any stclass tests offset forward from that
760 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
761 Start with the other substr.
762 XXXX no SCREAM optimization yet - and a very coarse implementation
763 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
764 *always* match. Probably should be marked during compile...
765 Probably it is right to do no SCREAM here...
768 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
769 : (prog->float_substr && prog->anchored_substr))
771 /* Take into account the "other" substring. */
772 /* XXXX May be hopelessly wrong for UTF... */
775 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
778 char * const last = HOP3c(s, -start_shift, strbeg);
780 char * const saved_s = s;
783 t = s - prog->check_offset_max;
784 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
786 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
791 t = HOP3c(t, prog->anchored_offset, strend);
792 if (t < other_last) /* These positions already checked */
794 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
797 /* XXXX It is not documented what units *_offsets are in.
798 We assume bytes, but this is clearly wrong.
799 Meaning this code needs to be carefully reviewed for errors.
803 /* On end-of-str: see comment below. */
804 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
805 if (must == &PL_sv_undef) {
807 DEBUG_r(must = prog->anchored_utf8); /* for debug */
812 HOP3(HOP3(last1, prog->anchored_offset, strend)
813 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
815 multiline ? FBMrf_MULTILINE : 0
818 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
819 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
820 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
821 (s ? "Found" : "Contradicts"),
822 quoted, RE_SV_TAIL(must));
827 if (last1 >= last2) {
828 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
829 ", giving up...\n"));
832 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
833 ", trying floating at offset %ld...\n",
834 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
835 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
836 s = HOP3c(last, 1, strend);
840 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
841 (long)(s - i_strpos)));
842 t = HOP3c(s, -prog->anchored_offset, strbeg);
843 other_last = HOP3c(s, 1, strend);
851 else { /* Take into account the floating substring. */
853 char * const saved_s = s;
856 t = HOP3c(s, -start_shift, strbeg);
858 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
859 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
860 last = HOP3c(t, prog->float_max_offset, strend);
861 s = HOP3c(t, prog->float_min_offset, strend);
864 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
865 must = utf8_target ? prog->float_utf8 : prog->float_substr;
866 /* fbm_instr() takes into account exact value of end-of-str
867 if the check is SvTAIL(ed). Since false positives are OK,
868 and end-of-str is not later than strend we are OK. */
869 if (must == &PL_sv_undef) {
871 DEBUG_r(must = prog->float_utf8); /* for debug message */
874 s = fbm_instr((unsigned char*)s,
875 (unsigned char*)last + SvCUR(must)
877 must, multiline ? FBMrf_MULTILINE : 0);
879 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
880 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
881 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
882 (s ? "Found" : "Contradicts"),
883 quoted, RE_SV_TAIL(must));
887 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
888 ", giving up...\n"));
891 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
892 ", trying anchored starting at offset %ld...\n",
893 (long)(saved_s + 1 - i_strpos)));
895 s = HOP3c(t, 1, strend);
899 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
900 (long)(s - i_strpos)));
901 other_last = s; /* Fix this later. --Hugo */
911 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
913 DEBUG_OPTIMISE_MORE_r(
914 PerlIO_printf(Perl_debug_log,
915 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
916 (IV)prog->check_offset_min,
917 (IV)prog->check_offset_max,
925 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
927 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
930 /* Fixed substring is found far enough so that the match
931 cannot start at strpos. */
933 if (ml_anch && t[-1] != '\n') {
934 /* Eventually fbm_*() should handle this, but often
935 anchored_offset is not 0, so this check will not be wasted. */
936 /* XXXX In the code below we prefer to look for "^" even in
937 presence of anchored substrings. And we search even
938 beyond the found float position. These pessimizations
939 are historical artefacts only. */
941 while (t < strend - prog->minlen) {
943 if (t < check_at - prog->check_offset_min) {
944 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
945 /* Since we moved from the found position,
946 we definitely contradict the found anchored
947 substr. Due to the above check we do not
948 contradict "check" substr.
949 Thus we can arrive here only if check substr
950 is float. Redo checking for "other"=="fixed".
953 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
954 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
955 goto do_other_anchored;
957 /* We don't contradict the found floating substring. */
958 /* XXXX Why not check for STCLASS? */
960 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
961 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
964 /* Position contradicts check-string */
965 /* XXXX probably better to look for check-string
966 than for "\n", so one should lower the limit for t? */
967 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
968 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
969 other_last = strpos = s = t + 1;
974 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
975 PL_colors[0], PL_colors[1]));
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
980 PL_colors[0], PL_colors[1]));
984 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
987 /* The found string does not prohibit matching at strpos,
988 - no optimization of calling REx engine can be performed,
989 unless it was an MBOL and we are not after MBOL,
990 or a future STCLASS check will fail this. */
992 /* Even in this situation we may use MBOL flag if strpos is offset
993 wrt the start of the string. */
994 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
995 && (strpos != strbeg) && strpos[-1] != '\n'
996 /* May be due to an implicit anchor of m{.*foo} */
997 && !(prog->intflags & PREGf_IMPLICIT))
1002 DEBUG_EXECUTE_r( if (ml_anch)
1003 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1004 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1007 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1009 prog->check_utf8 /* Could be deleted already */
1010 && --BmUSEFUL(prog->check_utf8) < 0
1011 && (prog->check_utf8 == prog->float_utf8)
1013 prog->check_substr /* Could be deleted already */
1014 && --BmUSEFUL(prog->check_substr) < 0
1015 && (prog->check_substr == prog->float_substr)
1018 /* If flags & SOMETHING - do not do it many times on the same match */
1019 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1020 /* XXX Does the destruction order has to change with utf8_target? */
1021 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1022 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1023 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1024 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1025 check = NULL; /* abort */
1027 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevent flag
1028 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1029 if (prog->intflags & PREGf_IMPLICIT)
1030 prog->extflags &= ~RXf_ANCH_MBOL;
1031 /* XXXX This is a remnant of the old implementation. It
1032 looks wasteful, since now INTUIT can use many
1033 other heuristics. */
1034 prog->extflags &= ~RXf_USE_INTUIT;
1035 /* XXXX What other flags might need to be cleared in this branch? */
1041 /* Last resort... */
1042 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1043 /* trie stclasses are too expensive to use here, we are better off to
1044 leave it to regmatch itself */
1045 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1046 /* minlen == 0 is possible if regstclass is \b or \B,
1047 and the fixed substr is ''$.
1048 Since minlen is already taken into account, s+1 is before strend;
1049 accidentally, minlen >= 1 guaranties no false positives at s + 1
1050 even for \b or \B. But (minlen? 1 : 0) below assumes that
1051 regstclass does not come from lookahead... */
1052 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1053 This leaves EXACTF only, which is dealt with in find_byclass(). */
1054 const U8* const str = (U8*)STRING(progi->regstclass);
1055 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1056 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1059 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1060 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1061 else if (prog->float_substr || prog->float_utf8)
1062 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1066 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1067 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1070 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1073 const char *what = NULL;
1075 if (endpos == strend) {
1076 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1077 "Could not match STCLASS...\n") );
1080 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1081 "This position contradicts STCLASS...\n") );
1082 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1084 /* Contradict one of substrings */
1085 if (prog->anchored_substr || prog->anchored_utf8) {
1086 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1087 DEBUG_EXECUTE_r( what = "anchored" );
1089 s = HOP3c(t, 1, strend);
1090 if (s + start_shift + end_shift > strend) {
1091 /* XXXX Should be taken into account earlier? */
1092 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1093 "Could not match STCLASS...\n") );
1098 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1099 "Looking for %s substr starting at offset %ld...\n",
1100 what, (long)(s + start_shift - i_strpos)) );
1103 /* Have both, check_string is floating */
1104 if (t + start_shift >= check_at) /* Contradicts floating=check */
1105 goto retry_floating_check;
1106 /* Recheck anchored substring, but not floating... */
1110 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1111 "Looking for anchored substr starting at offset %ld...\n",
1112 (long)(other_last - i_strpos)) );
1113 goto do_other_anchored;
1115 /* Another way we could have checked stclass at the
1116 current position only: */
1121 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1122 "Looking for /%s^%s/m starting at offset %ld...\n",
1123 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1126 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1128 /* Check is floating subtring. */
1129 retry_floating_check:
1130 t = check_at - start_shift;
1131 DEBUG_EXECUTE_r( what = "floating" );
1132 goto hop_and_restart;
1135 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1136 "By STCLASS: moving %ld --> %ld\n",
1137 (long)(t - i_strpos), (long)(s - i_strpos))
1141 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1142 "Does not contradict STCLASS...\n");
1147 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1148 PL_colors[4], (check ? "Guessed" : "Giving up"),
1149 PL_colors[5], (long)(s - i_strpos)) );
1152 fail_finish: /* Substring not found */
1153 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1154 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1156 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1157 PL_colors[4], PL_colors[5]));
1161 #define DECL_TRIE_TYPE(scan) \
1162 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1163 trie_type = (scan->flags != EXACT) \
1164 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1165 : (utf8_target ? trie_utf8 : trie_plain)
1167 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1168 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1169 switch (trie_type) { \
1170 case trie_utf8_fold: \
1171 if ( foldlen>0 ) { \
1172 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1177 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1178 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1179 foldlen -= UNISKIP( uvc ); \
1180 uscan = foldbuf + UNISKIP( uvc ); \
1183 case trie_latin_utf8_fold: \
1184 if ( foldlen>0 ) { \
1185 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1191 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1192 foldlen -= UNISKIP( uvc ); \
1193 uscan = foldbuf + UNISKIP( uvc ); \
1197 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1204 charid = trie->charmap[ uvc ]; \
1208 if (widecharmap) { \
1209 SV** const svpp = hv_fetch(widecharmap, \
1210 (char*)&uvc, sizeof(UV), 0); \
1212 charid = (U16)SvIV(*svpp); \
1217 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1219 char *my_strend= (char *)strend; \
1222 foldEQ_utf8(s, &my_strend, 0, utf8_target, \
1223 m, NULL, ln, cBOOL(UTF_PATTERN))) \
1224 && (!reginfo || regtry(reginfo, &s)) ) \
1227 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1228 uvchr_to_utf8(tmpbuf, c); \
1229 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1231 && (f == c1 || f == c2) \
1233 foldEQ_utf8(s, &my_strend, 0, utf8_target,\
1234 m, NULL, ln, cBOOL(UTF_PATTERN)))\
1235 && (!reginfo || regtry(reginfo, &s)) ) \
1241 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1245 && (ln == 1 || (OP(c) == EXACTF \
1246 ? foldEQ(s, m, ln) \
1247 : foldEQ_locale(s, m, ln))) \
1248 && (!reginfo || regtry(reginfo, &s)) ) \
1254 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1256 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1262 #define REXEC_FBC_SCAN(CoDe) \
1264 while (s < strend) { \
1270 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1271 REXEC_FBC_UTF8_SCAN( \
1273 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1282 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1285 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1294 #define REXEC_FBC_TRYIT \
1295 if ((!reginfo || regtry(reginfo, &s))) \
1298 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1299 if (utf8_target) { \
1300 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1303 REXEC_FBC_CLASS_SCAN(CoNd); \
1307 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1308 if (utf8_target) { \
1310 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1313 REXEC_FBC_CLASS_SCAN(CoNd); \
1317 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1318 PL_reg_flags |= RF_tainted; \
1319 if (utf8_target) { \
1320 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1323 REXEC_FBC_CLASS_SCAN(CoNd); \
1327 #define DUMP_EXEC_POS(li,s,doutf8) \
1328 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1330 /* We know what class REx starts with. Try to find this position... */
1331 /* if reginfo is NULL, its a dryrun */
1332 /* annoyingly all the vars in this routine have different names from their counterparts
1333 in regmatch. /grrr */
1336 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1337 const char *strend, regmatch_info *reginfo)
1340 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1344 register STRLEN uskip;
1348 register I32 tmp = 1; /* Scratch variable? */
1349 register const bool utf8_target = PL_reg_match_utf8;
1350 RXi_GET_DECL(prog,progi);
1352 PERL_ARGS_ASSERT_FIND_BYCLASS;
1354 /* We know what class it must start with. */
1358 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1359 !UTF8_IS_INVARIANT((U8)s[0]) ?
1360 reginclass(prog, c, (U8*)s, 0, utf8_target) :
1361 REGINCLASS(prog, c, (U8*)s));
1364 while (s < strend) {
1367 if (REGINCLASS(prog, c, (U8*)s) ||
1368 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1369 /* The assignment of 2 is intentional:
1370 * for the folded sharp s, the skip is 2. */
1371 (skip = SHARP_S_SKIP))) {
1372 if (tmp && (!reginfo || regtry(reginfo, &s)))
1385 if (tmp && (!reginfo || regtry(reginfo, &s)))
1393 ln = STR_LEN(c); /* length to match in octets/bytes */
1394 lnc = (I32) ln; /* length to match in characters */
1396 STRLEN ulen1, ulen2;
1398 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1399 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1400 /* used by commented-out code below */
1401 /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
1403 /* XXX: Since the node will be case folded at compile
1404 time this logic is a little odd, although im not
1405 sure that its actually wrong. --dmq */
1407 c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1408 c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1410 /* XXX: This is kinda strange. to_utf8_XYZ returns the
1411 codepoint of the first character in the converted
1412 form, yet originally we did the extra step.
1413 No tests fail by commenting this code out however
1414 so Ive left it out. -- dmq.
1416 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1418 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1423 while (sm < ((U8 *) m + ln)) {
1438 c2 = PL_fold_locale[c1];
1440 e = HOP3c(strend, -((I32)lnc), s);
1442 if (!reginfo && e < s)
1443 e = s; /* Due to minlen logic of intuit() */
1445 /* The idea in the EXACTF* cases is to first find the
1446 * first character of the EXACTF* node and then, if
1447 * necessary, case-insensitively compare the full
1448 * text of the node. The c1 and c2 are the first
1449 * characters (though in Unicode it gets a bit
1450 * more complicated because there are more cases
1451 * than just upper and lower: one needs to use
1452 * the so-called folding case for case-insensitive
1453 * matching (called "loose matching" in Unicode).
1454 * foldEQ_utf8() will do just that. */
1456 if (utf8_target || UTF_PATTERN) {
1458 U8 tmpbuf [UTF8_MAXBYTES+1];
1461 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1463 /* Upper and lower of 1st char are equal -
1464 * probably not a "letter". */
1467 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1472 REXEC_FBC_EXACTISH_CHECK(c == c1);
1478 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1484 /* Handle some of the three Greek sigmas cases.
1485 * Note that not all the possible combinations
1486 * are handled here: some of them are handled
1487 * by the standard folding rules, and some of
1488 * them (the character class or ANYOF cases)
1489 * are handled during compiletime in
1490 * regexec.c:S_regclass(). */
1491 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1492 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1493 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1495 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1500 /* Neither pattern nor string are UTF8 */
1502 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1504 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1508 PL_reg_flags |= RF_tainted;
1515 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1516 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1518 tmp = ((OP(c) == BOUND ?
1519 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1520 LOAD_UTF8_CHARCLASS_ALNUM();
1521 REXEC_FBC_UTF8_SCAN(
1522 if (tmp == !(OP(c) == BOUND ?
1523 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)) :
1524 isALNUM_LC_utf8((U8*)s)))
1531 else { /* Not utf8 */
1532 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1533 tmp = cBOOL((OP(c) == BOUNDL)
1535 : (isWORDCHAR_L1(tmp)
1536 && (isASCII(tmp) || (FLAGS(c) & USE_UNI))));
1541 : (isWORDCHAR_L1((U8) *s)
1542 && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))))
1549 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1553 PL_reg_flags |= RF_tainted;
1560 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1561 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1563 tmp = ((OP(c) == NBOUND ?
1564 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1565 LOAD_UTF8_CHARCLASS_ALNUM();
1566 REXEC_FBC_UTF8_SCAN(
1567 if (tmp == !(OP(c) == NBOUND ?
1568 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)) :
1569 isALNUM_LC_utf8((U8*)s)))
1571 else REXEC_FBC_TRYIT;
1575 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1576 tmp = cBOOL((OP(c) == NBOUNDL)
1578 : (isWORDCHAR_L1(tmp)
1579 && (isASCII(tmp) || (FLAGS(c) & USE_UNI))));
1584 : (isWORDCHAR_L1((U8) *s)
1585 && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))))
1589 else REXEC_FBC_TRYIT;
1592 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1596 REXEC_FBC_CSCAN_PRELOAD(
1597 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1598 swash_fetch(RE_utf8_perl_word, (U8*)s, utf8_target),
1599 (FLAGS(c) & USE_UNI) ? isWORDCHAR_L1((U8) *s) : isALNUM(*s)
1602 REXEC_FBC_CSCAN_TAINT(
1603 isALNUM_LC_utf8((U8*)s),
1607 REXEC_FBC_CSCAN_PRELOAD(
1608 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1609 !swash_fetch(RE_utf8_perl_word, (U8*)s, utf8_target),
1610 ! ((FLAGS(c) & USE_UNI) ? isWORDCHAR_L1((U8) *s) : isALNUM(*s))
1613 REXEC_FBC_CSCAN_TAINT(
1614 !isALNUM_LC_utf8((U8*)s),
1618 REXEC_FBC_CSCAN_PRELOAD(
1619 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1620 *s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, utf8_target),
1621 isSPACE_L1((U8) *s) && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI))
1624 REXEC_FBC_CSCAN_TAINT(
1625 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1629 REXEC_FBC_CSCAN_PRELOAD(
1630 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1631 !(*s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, utf8_target)),
1632 !(isSPACE_L1((U8) *s) && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))
1635 REXEC_FBC_CSCAN_TAINT(
1636 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1640 REXEC_FBC_CSCAN_PRELOAD(
1641 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1642 swash_fetch(RE_utf8_posix_digit,(U8*)s, utf8_target),
1646 REXEC_FBC_CSCAN_TAINT(
1647 isDIGIT_LC_utf8((U8*)s),
1651 REXEC_FBC_CSCAN_PRELOAD(
1652 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1653 !swash_fetch(RE_utf8_posix_digit,(U8*)s, utf8_target),
1657 REXEC_FBC_CSCAN_TAINT(
1658 !isDIGIT_LC_utf8((U8*)s),
1664 is_LNBREAK_latin1(s)
1674 !is_VERTWS_latin1(s)
1679 is_HORIZWS_latin1(s)
1683 !is_HORIZWS_utf8(s),
1684 !is_HORIZWS_latin1(s)
1690 /* what trie are we using right now */
1692 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1694 = (reg_trie_data*)progi->data->data[ aho->trie ];
1695 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1697 const char *last_start = strend - trie->minlen;
1699 const char *real_start = s;
1701 STRLEN maxlen = trie->maxlen;
1703 U8 **points; /* map of where we were in the input string
1704 when reading a given char. For ASCII this
1705 is unnecessary overhead as the relationship
1706 is always 1:1, but for Unicode, especially
1707 case folded Unicode this is not true. */
1708 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1712 GET_RE_DEBUG_FLAGS_DECL;
1714 /* We can't just allocate points here. We need to wrap it in
1715 * an SV so it gets freed properly if there is a croak while
1716 * running the match */
1719 sv_points=newSV(maxlen * sizeof(U8 *));
1720 SvCUR_set(sv_points,
1721 maxlen * sizeof(U8 *));
1722 SvPOK_on(sv_points);
1723 sv_2mortal(sv_points);
1724 points=(U8**)SvPV_nolen(sv_points );
1725 if ( trie_type != trie_utf8_fold
1726 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1729 bitmap=(U8*)trie->bitmap;
1731 bitmap=(U8*)ANYOF_BITMAP(c);
1733 /* this is the Aho-Corasick algorithm modified a touch
1734 to include special handling for long "unknown char"
1735 sequences. The basic idea being that we use AC as long
1736 as we are dealing with a possible matching char, when
1737 we encounter an unknown char (and we have not encountered
1738 an accepting state) we scan forward until we find a legal
1740 AC matching is basically that of trie matching, except
1741 that when we encounter a failing transition, we fall back
1742 to the current states "fail state", and try the current char
1743 again, a process we repeat until we reach the root state,
1744 state 1, or a legal transition. If we fail on the root state
1745 then we can either terminate if we have reached an accepting
1746 state previously, or restart the entire process from the beginning
1750 while (s <= last_start) {
1751 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1759 U8 *uscan = (U8*)NULL;
1760 U8 *leftmost = NULL;
1762 U32 accepted_word= 0;
1766 while ( state && uc <= (U8*)strend ) {
1768 U32 word = aho->states[ state ].wordnum;
1772 DEBUG_TRIE_EXECUTE_r(
1773 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1774 dump_exec_pos( (char *)uc, c, strend, real_start,
1775 (char *)uc, utf8_target );
1776 PerlIO_printf( Perl_debug_log,
1777 " Scanning for legal start char...\n");
1780 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1785 if (uc >(U8*)last_start) break;
1789 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1790 if (!leftmost || lpos < leftmost) {
1791 DEBUG_r(accepted_word=word);
1797 points[pointpos++ % maxlen]= uc;
1798 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1799 uscan, len, uvc, charid, foldlen,
1801 DEBUG_TRIE_EXECUTE_r({
1802 dump_exec_pos( (char *)uc, c, strend, real_start,
1804 PerlIO_printf(Perl_debug_log,
1805 " Charid:%3u CP:%4"UVxf" ",
1811 word = aho->states[ state ].wordnum;
1813 base = aho->states[ state ].trans.base;
1815 DEBUG_TRIE_EXECUTE_r({
1817 dump_exec_pos( (char *)uc, c, strend, real_start,
1819 PerlIO_printf( Perl_debug_log,
1820 "%sState: %4"UVxf", word=%"UVxf,
1821 failed ? " Fail transition to " : "",
1822 (UV)state, (UV)word);
1828 ( ((offset = base + charid
1829 - 1 - trie->uniquecharcount)) >= 0)
1830 && ((U32)offset < trie->lasttrans)
1831 && trie->trans[offset].check == state
1832 && (tmp=trie->trans[offset].next))
1834 DEBUG_TRIE_EXECUTE_r(
1835 PerlIO_printf( Perl_debug_log," - legal\n"));
1840 DEBUG_TRIE_EXECUTE_r(
1841 PerlIO_printf( Perl_debug_log," - fail\n"));
1843 state = aho->fail[state];
1847 /* we must be accepting here */
1848 DEBUG_TRIE_EXECUTE_r(
1849 PerlIO_printf( Perl_debug_log," - accepting\n"));
1858 if (!state) state = 1;
1861 if ( aho->states[ state ].wordnum ) {
1862 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1863 if (!leftmost || lpos < leftmost) {
1864 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1869 s = (char*)leftmost;
1870 DEBUG_TRIE_EXECUTE_r({
1872 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1873 (UV)accepted_word, (IV)(s - real_start)
1876 if (!reginfo || regtry(reginfo, &s)) {
1882 DEBUG_TRIE_EXECUTE_r({
1883 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1886 DEBUG_TRIE_EXECUTE_r(
1887 PerlIO_printf( Perl_debug_log,"No match.\n"));
1896 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1906 - regexec_flags - match a regexp against a string
1909 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1910 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1911 /* strend: pointer to null at end of string */
1912 /* strbeg: real beginning of string */
1913 /* minend: end of match must be >=minend after stringarg. */
1914 /* data: May be used for some additional optimizations.
1915 Currently its only used, with a U32 cast, for transmitting
1916 the ganch offset when doing a /g match. This will change */
1917 /* nosave: For optimizations. */
1920 struct regexp *const prog = (struct regexp *)SvANY(rx);
1921 /*register*/ char *s;
1922 register regnode *c;
1923 /*register*/ char *startpos = stringarg;
1924 I32 minlen; /* must match at least this many chars */
1925 I32 dontbother = 0; /* how many characters not to try at end */
1926 I32 end_shift = 0; /* Same for the end. */ /* CC */
1927 I32 scream_pos = -1; /* Internal iterator of scream. */
1928 char *scream_olds = NULL;
1929 const bool utf8_target = cBOOL(DO_UTF8(sv));
1931 RXi_GET_DECL(prog,progi);
1932 regmatch_info reginfo; /* create some info to pass to regtry etc */
1933 regexp_paren_pair *swap = NULL;
1934 GET_RE_DEBUG_FLAGS_DECL;
1936 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1937 PERL_UNUSED_ARG(data);
1939 /* Be paranoid... */
1940 if (prog == NULL || startpos == NULL) {
1941 Perl_croak(aTHX_ "NULL regexp parameter");
1945 multiline = prog->extflags & RXf_PMf_MULTILINE;
1946 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
1948 RX_MATCH_UTF8_set(rx, utf8_target);
1950 debug_start_match(rx, utf8_target, startpos, strend,
1954 minlen = prog->minlen;
1956 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1957 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1958 "String too short [regexec_flags]...\n"));
1963 /* Check validity of program. */
1964 if (UCHARAT(progi->program) != REG_MAGIC) {
1965 Perl_croak(aTHX_ "corrupted regexp program");
1969 PL_reg_eval_set = 0;
1973 PL_reg_flags |= RF_utf8;
1975 /* Mark beginning of line for ^ and lookbehind. */
1976 reginfo.bol = startpos; /* XXX not used ??? */
1980 /* Mark end of line for $ (and such) */
1983 /* see how far we have to get to not match where we matched before */
1984 reginfo.till = startpos+minend;
1986 /* If there is a "must appear" string, look for it. */
1989 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1991 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
1992 reginfo.ganch = startpos + prog->gofs;
1993 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1994 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
1995 } else if (sv && SvTYPE(sv) >= SVt_PVMG
1997 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1998 && mg->mg_len >= 0) {
1999 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2000 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2001 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2003 if (prog->extflags & RXf_ANCH_GPOS) {
2004 if (s > reginfo.ganch)
2006 s = reginfo.ganch - prog->gofs;
2007 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2008 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2014 reginfo.ganch = strbeg + PTR2UV(data);
2015 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2016 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2018 } else { /* pos() not defined */
2019 reginfo.ganch = strbeg;
2020 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2021 "GPOS: reginfo.ganch = strbeg\n"));
2024 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2025 /* We have to be careful. If the previous successful match
2026 was from this regex we don't want a subsequent partially
2027 successful match to clobber the old results.
2028 So when we detect this possibility we add a swap buffer
2029 to the re, and switch the buffer each match. If we fail
2030 we switch it back, otherwise we leave it swapped.
2033 /* do we need a save destructor here for eval dies? */
2034 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2036 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2037 re_scream_pos_data d;
2039 d.scream_olds = &scream_olds;
2040 d.scream_pos = &scream_pos;
2041 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2043 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2044 goto phooey; /* not present */
2050 /* Simplest case: anchored match need be tried only once. */
2051 /* [unless only anchor is BOL and multiline is set] */
2052 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2053 if (s == startpos && regtry(®info, &startpos))
2055 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2056 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2061 dontbother = minlen - 1;
2062 end = HOP3c(strend, -dontbother, strbeg) - 1;
2063 /* for multiline we only have to try after newlines */
2064 if (prog->check_substr || prog->check_utf8) {
2065 /* because of the goto we can not easily reuse the macros for bifurcating the
2066 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2069 goto after_try_utf8;
2071 if (regtry(®info, &s)) {
2078 if (prog->extflags & RXf_USE_INTUIT) {
2079 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2088 } /* end search for check string in unicode */
2090 if (s == startpos) {
2091 goto after_try_latin;
2094 if (regtry(®info, &s)) {
2101 if (prog->extflags & RXf_USE_INTUIT) {
2102 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2111 } /* end search for check string in latin*/
2112 } /* end search for check string */
2113 else { /* search for newline */
2115 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2118 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2120 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2121 if (regtry(®info, &s))
2125 } /* end search for newline */
2126 } /* end anchored/multiline check string search */
2128 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2130 /* the warning about reginfo.ganch being used without intialization
2131 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2132 and we only enter this block when the same bit is set. */
2133 char *tmp_s = reginfo.ganch - prog->gofs;
2135 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2140 /* Messy cases: unanchored match. */
2141 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2142 /* we have /x+whatever/ */
2143 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2148 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2149 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2150 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2155 DEBUG_EXECUTE_r( did_match = 1 );
2156 if (regtry(®info, &s)) goto got_it;
2158 while (s < strend && *s == ch)
2166 DEBUG_EXECUTE_r( did_match = 1 );
2167 if (regtry(®info, &s)) goto got_it;
2169 while (s < strend && *s == ch)
2174 DEBUG_EXECUTE_r(if (!did_match)
2175 PerlIO_printf(Perl_debug_log,
2176 "Did not find anchored character...\n")
2179 else if (prog->anchored_substr != NULL
2180 || prog->anchored_utf8 != NULL
2181 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2182 && prog->float_max_offset < strend - s)) {
2187 char *last1; /* Last position checked before */
2191 if (prog->anchored_substr || prog->anchored_utf8) {
2192 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2193 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2194 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2195 back_max = back_min = prog->anchored_offset;
2197 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2198 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2199 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2200 back_max = prog->float_max_offset;
2201 back_min = prog->float_min_offset;
2205 if (must == &PL_sv_undef)
2206 /* could not downgrade utf8 check substring, so must fail */
2212 last = HOP3c(strend, /* Cannot start after this */
2213 -(I32)(CHR_SVLEN(must)
2214 - (SvTAIL(must) != 0) + back_min), strbeg);
2217 last1 = HOPc(s, -1);
2219 last1 = s - 1; /* bogus */
2221 /* XXXX check_substr already used to find "s", can optimize if
2222 check_substr==must. */
2224 dontbother = end_shift;
2225 strend = HOPc(strend, -dontbother);
2226 while ( (s <= last) &&
2227 ((flags & REXEC_SCREAM)
2228 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2229 end_shift, &scream_pos, 0))
2230 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2231 (unsigned char*)strend, must,
2232 multiline ? FBMrf_MULTILINE : 0))) ) {
2233 /* we may be pointing at the wrong string */
2234 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2235 s = strbeg + (s - SvPVX_const(sv));
2236 DEBUG_EXECUTE_r( did_match = 1 );
2237 if (HOPc(s, -back_max) > last1) {
2238 last1 = HOPc(s, -back_min);
2239 s = HOPc(s, -back_max);
2242 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2244 last1 = HOPc(s, -back_min);
2248 while (s <= last1) {
2249 if (regtry(®info, &s))
2255 while (s <= last1) {
2256 if (regtry(®info, &s))
2262 DEBUG_EXECUTE_r(if (!did_match) {
2263 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2264 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2265 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2266 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2267 ? "anchored" : "floating"),
2268 quoted, RE_SV_TAIL(must));
2272 else if ( (c = progi->regstclass) ) {
2274 const OPCODE op = OP(progi->regstclass);
2275 /* don't bother with what can't match */
2276 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2277 strend = HOPc(strend, -(minlen - 1));
2280 SV * const prop = sv_newmortal();
2281 regprop(prog, prop, c);
2283 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2285 PerlIO_printf(Perl_debug_log,
2286 "Matching stclass %.*s against %s (%d bytes)\n",
2287 (int)SvCUR(prop), SvPVX_const(prop),
2288 quoted, (int)(strend - s));
2291 if (find_byclass(prog, c, s, strend, ®info))
2293 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2297 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2302 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2303 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2304 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2306 if (flags & REXEC_SCREAM) {
2307 last = screaminstr(sv, float_real, s - strbeg,
2308 end_shift, &scream_pos, 1); /* last one */
2310 last = scream_olds; /* Only one occurrence. */
2311 /* we may be pointing at the wrong string */
2312 else if (RXp_MATCH_COPIED(prog))
2313 s = strbeg + (s - SvPVX_const(sv));
2317 const char * const little = SvPV_const(float_real, len);
2319 if (SvTAIL(float_real)) {
2320 if (memEQ(strend - len + 1, little, len - 1))
2321 last = strend - len + 1;
2322 else if (!multiline)
2323 last = memEQ(strend - len, little, len)
2324 ? strend - len : NULL;
2330 last = rninstr(s, strend, little, little + len);
2332 last = strend; /* matching "$" */
2337 PerlIO_printf(Perl_debug_log,
2338 "%sCan't trim the tail, match fails (should not happen)%s\n",
2339 PL_colors[4], PL_colors[5]));
2340 goto phooey; /* Should not happen! */
2342 dontbother = strend - last + prog->float_min_offset;
2344 if (minlen && (dontbother < minlen))
2345 dontbother = minlen - 1;
2346 strend -= dontbother; /* this one's always in bytes! */
2347 /* We don't know much -- general case. */
2350 if (regtry(®info, &s))
2359 if (regtry(®info, &s))
2361 } while (s++ < strend);
2370 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2372 if (PL_reg_eval_set)
2373 restore_pos(aTHX_ prog);
2374 if (RXp_PAREN_NAMES(prog))
2375 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2377 /* make sure $`, $&, $', and $digit will work later */
2378 if ( !(flags & REXEC_NOT_FIRST) ) {
2379 RX_MATCH_COPY_FREE(rx);
2380 if (flags & REXEC_COPY_STR) {
2381 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2382 #ifdef PERL_OLD_COPY_ON_WRITE
2384 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2386 PerlIO_printf(Perl_debug_log,
2387 "Copy on write: regexp capture, type %d\n",
2390 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2391 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2392 assert (SvPOKp(prog->saved_copy));
2396 RX_MATCH_COPIED_on(rx);
2397 s = savepvn(strbeg, i);
2403 prog->subbeg = strbeg;
2404 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2411 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2412 PL_colors[4], PL_colors[5]));
2413 if (PL_reg_eval_set)
2414 restore_pos(aTHX_ prog);
2416 /* we failed :-( roll it back */
2417 Safefree(prog->offs);
2426 - regtry - try match at specific point
2428 STATIC I32 /* 0 failure, 1 success */
2429 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2433 REGEXP *const rx = reginfo->prog;
2434 regexp *const prog = (struct regexp *)SvANY(rx);
2435 RXi_GET_DECL(prog,progi);
2436 GET_RE_DEBUG_FLAGS_DECL;
2438 PERL_ARGS_ASSERT_REGTRY;
2440 reginfo->cutpoint=NULL;
2442 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2445 PL_reg_eval_set = RS_init;
2446 DEBUG_EXECUTE_r(DEBUG_s(
2447 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2448 (IV)(PL_stack_sp - PL_stack_base));
2451 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2452 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2454 /* Apparently this is not needed, judging by wantarray. */
2455 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2456 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2459 /* Make $_ available to executed code. */
2460 if (reginfo->sv != DEFSV) {
2462 DEFSV_set(reginfo->sv);
2465 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2466 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2467 /* prepare for quick setting of pos */
2468 #ifdef PERL_OLD_COPY_ON_WRITE
2469 if (SvIsCOW(reginfo->sv))
2470 sv_force_normal_flags(reginfo->sv, 0);
2472 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2473 &PL_vtbl_mglob, NULL, 0);
2477 PL_reg_oldpos = mg->mg_len;
2478 SAVEDESTRUCTOR_X(restore_pos, prog);
2480 if (!PL_reg_curpm) {
2481 Newxz(PL_reg_curpm, 1, PMOP);
2484 SV* const repointer = &PL_sv_undef;
2485 /* this regexp is also owned by the new PL_reg_curpm, which
2486 will try to free it. */
2487 av_push(PL_regex_padav, repointer);
2488 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2489 PL_regex_pad = AvARRAY(PL_regex_padav);
2494 /* It seems that non-ithreads works both with and without this code.
2495 So for efficiency reasons it seems best not to have the code
2496 compiled when it is not needed. */
2497 /* This is safe against NULLs: */
2498 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2499 /* PM_reg_curpm owns a reference to this regexp. */
2502 PM_SETRE(PL_reg_curpm, rx);
2503 PL_reg_oldcurpm = PL_curpm;
2504 PL_curpm = PL_reg_curpm;
2505 if (RXp_MATCH_COPIED(prog)) {
2506 /* Here is a serious problem: we cannot rewrite subbeg,
2507 since it may be needed if this match fails. Thus
2508 $` inside (?{}) could fail... */
2509 PL_reg_oldsaved = prog->subbeg;
2510 PL_reg_oldsavedlen = prog->sublen;
2511 #ifdef PERL_OLD_COPY_ON_WRITE
2512 PL_nrs = prog->saved_copy;
2514 RXp_MATCH_COPIED_off(prog);
2517 PL_reg_oldsaved = NULL;
2518 prog->subbeg = PL_bostr;
2519 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2521 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2522 prog->offs[0].start = *startpos - PL_bostr;
2523 PL_reginput = *startpos;
2524 PL_reglastparen = &prog->lastparen;
2525 PL_reglastcloseparen = &prog->lastcloseparen;
2526 prog->lastparen = 0;
2527 prog->lastcloseparen = 0;
2529 PL_regoffs = prog->offs;
2530 if (PL_reg_start_tmpl <= prog->nparens) {
2531 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2532 if(PL_reg_start_tmp)
2533 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2535 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2538 /* XXXX What this code is doing here?!!! There should be no need
2539 to do this again and again, PL_reglastparen should take care of
2542 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2543 * Actually, the code in regcppop() (which Ilya may be meaning by
2544 * PL_reglastparen), is not needed at all by the test suite
2545 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2546 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2547 * Meanwhile, this code *is* needed for the
2548 * above-mentioned test suite tests to succeed. The common theme
2549 * on those tests seems to be returning null fields from matches.
2550 * --jhi updated by dapm */
2552 if (prog->nparens) {
2553 regexp_paren_pair *pp = PL_regoffs;
2555 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2563 if (regmatch(reginfo, progi->program + 1)) {
2564 PL_regoffs[0].end = PL_reginput - PL_bostr;
2567 if (reginfo->cutpoint)
2568 *startpos= reginfo->cutpoint;
2569 REGCP_UNWIND(lastcp);
2574 #define sayYES goto yes
2575 #define sayNO goto no
2576 #define sayNO_SILENT goto no_silent
2578 /* we dont use STMT_START/END here because it leads to
2579 "unreachable code" warnings, which are bogus, but distracting. */
2580 #define CACHEsayNO \
2581 if (ST.cache_mask) \
2582 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2585 /* this is used to determine how far from the left messages like
2586 'failed...' are printed. It should be set such that messages
2587 are inline with the regop output that created them.
2589 #define REPORT_CODE_OFF 32
2592 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2593 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2595 #define SLAB_FIRST(s) (&(s)->states[0])
2596 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2598 /* grab a new slab and return the first slot in it */
2600 STATIC regmatch_state *
2603 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2606 regmatch_slab *s = PL_regmatch_slab->next;
2608 Newx(s, 1, regmatch_slab);
2609 s->prev = PL_regmatch_slab;
2611 PL_regmatch_slab->next = s;
2613 PL_regmatch_slab = s;
2614 return SLAB_FIRST(s);
2618 /* push a new state then goto it */
2620 #define PUSH_STATE_GOTO(state, node) \
2622 st->resume_state = state; \
2625 /* push a new state with success backtracking, then goto it */
2627 #define PUSH_YES_STATE_GOTO(state, node) \
2629 st->resume_state = state; \
2630 goto push_yes_state;
2636 regmatch() - main matching routine
2638 This is basically one big switch statement in a loop. We execute an op,
2639 set 'next' to point the next op, and continue. If we come to a point which
2640 we may need to backtrack to on failure such as (A|B|C), we push a
2641 backtrack state onto the backtrack stack. On failure, we pop the top
2642 state, and re-enter the loop at the state indicated. If there are no more
2643 states to pop, we return failure.
2645 Sometimes we also need to backtrack on success; for example /A+/, where
2646 after successfully matching one A, we need to go back and try to
2647 match another one; similarly for lookahead assertions: if the assertion
2648 completes successfully, we backtrack to the state just before the assertion
2649 and then carry on. In these cases, the pushed state is marked as
2650 'backtrack on success too'. This marking is in fact done by a chain of
2651 pointers, each pointing to the previous 'yes' state. On success, we pop to
2652 the nearest yes state, discarding any intermediate failure-only states.
2653 Sometimes a yes state is pushed just to force some cleanup code to be
2654 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2655 it to free the inner regex.
2657 Note that failure backtracking rewinds the cursor position, while
2658 success backtracking leaves it alone.
2660 A pattern is complete when the END op is executed, while a subpattern
2661 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2662 ops trigger the "pop to last yes state if any, otherwise return true"
2665 A common convention in this function is to use A and B to refer to the two
2666 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2667 the subpattern to be matched possibly multiple times, while B is the entire
2668 rest of the pattern. Variable and state names reflect this convention.
2670 The states in the main switch are the union of ops and failure/success of
2671 substates associated with with that op. For example, IFMATCH is the op
2672 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2673 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2674 successfully matched A and IFMATCH_A_fail is a state saying that we have
2675 just failed to match A. Resume states always come in pairs. The backtrack
2676 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2677 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2678 on success or failure.
2680 The struct that holds a backtracking state is actually a big union, with
2681 one variant for each major type of op. The variable st points to the
2682 top-most backtrack struct. To make the code clearer, within each
2683 block of code we #define ST to alias the relevant union.
2685 Here's a concrete example of a (vastly oversimplified) IFMATCH
2691 #define ST st->u.ifmatch
2693 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2694 ST.foo = ...; // some state we wish to save
2696 // push a yes backtrack state with a resume value of
2697 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2699 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2702 case IFMATCH_A: // we have successfully executed A; now continue with B
2704 bar = ST.foo; // do something with the preserved value
2707 case IFMATCH_A_fail: // A failed, so the assertion failed
2708 ...; // do some housekeeping, then ...
2709 sayNO; // propagate the failure
2716 For any old-timers reading this who are familiar with the old recursive
2717 approach, the code above is equivalent to:
2719 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2728 ...; // do some housekeeping, then ...
2729 sayNO; // propagate the failure
2732 The topmost backtrack state, pointed to by st, is usually free. If you
2733 want to claim it, populate any ST.foo fields in it with values you wish to
2734 save, then do one of
2736 PUSH_STATE_GOTO(resume_state, node);
2737 PUSH_YES_STATE_GOTO(resume_state, node);
2739 which sets that backtrack state's resume value to 'resume_state', pushes a
2740 new free entry to the top of the backtrack stack, then goes to 'node'.
2741 On backtracking, the free slot is popped, and the saved state becomes the
2742 new free state. An ST.foo field in this new top state can be temporarily
2743 accessed to retrieve values, but once the main loop is re-entered, it
2744 becomes available for reuse.
2746 Note that the depth of the backtrack stack constantly increases during the
2747 left-to-right execution of the pattern, rather than going up and down with
2748 the pattern nesting. For example the stack is at its maximum at Z at the
2749 end of the pattern, rather than at X in the following:
2751 /(((X)+)+)+....(Y)+....Z/
2753 The only exceptions to this are lookahead/behind assertions and the cut,
2754 (?>A), which pop all the backtrack states associated with A before
2757 Bascktrack state structs are allocated in slabs of about 4K in size.
2758 PL_regmatch_state and st always point to the currently active state,
2759 and PL_regmatch_slab points to the slab currently containing
2760 PL_regmatch_state. The first time regmatch() is called, the first slab is
2761 allocated, and is never freed until interpreter destruction. When the slab
2762 is full, a new one is allocated and chained to the end. At exit from
2763 regmatch(), slabs allocated since entry are freed.
2768 #define DEBUG_STATE_pp(pp) \
2770 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2771 PerlIO_printf(Perl_debug_log, \
2772 " %*s"pp" %s%s%s%s%s\n", \
2774 PL_reg_name[st->resume_state], \
2775 ((st==yes_state||st==mark_state) ? "[" : ""), \
2776 ((st==yes_state) ? "Y" : ""), \
2777 ((st==mark_state) ? "M" : ""), \
2778 ((st==yes_state||st==mark_state) ? "]" : "") \
2783 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2788 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2789 const char *start, const char *end, const char *blurb)
2791 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2793 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2798 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2799 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2801 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2802 start, end - start, 60);
2804 PerlIO_printf(Perl_debug_log,
2805 "%s%s REx%s %s against %s\n",
2806 PL_colors[4], blurb, PL_colors[5], s0, s1);
2808 if (utf8_target||utf8_pat)
2809 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2810 utf8_pat ? "pattern" : "",
2811 utf8_pat && utf8_target ? " and " : "",
2812 utf8_target ? "string" : ""
2818 S_dump_exec_pos(pTHX_ const char *locinput,
2819 const regnode *scan,
2820 const char *loc_regeol,
2821 const char *loc_bostr,
2822 const char *loc_reg_starttry,
2823 const bool utf8_target)
2825 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2826 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2827 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2828 /* The part of the string before starttry has one color
2829 (pref0_len chars), between starttry and current
2830 position another one (pref_len - pref0_len chars),
2831 after the current position the third one.
2832 We assume that pref0_len <= pref_len, otherwise we
2833 decrease pref0_len. */
2834 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2835 ? (5 + taill) - l : locinput - loc_bostr;
2838 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2840 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2842 pref0_len = pref_len - (locinput - loc_reg_starttry);
2843 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2844 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2845 ? (5 + taill) - pref_len : loc_regeol - locinput);
2846 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2850 if (pref0_len > pref_len)
2851 pref0_len = pref_len;
2853 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2855 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2856 (locinput - pref_len),pref0_len, 60, 4, 5);
2858 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2859 (locinput - pref_len + pref0_len),
2860 pref_len - pref0_len, 60, 2, 3);
2862 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2863 locinput, loc_regeol - locinput, 10, 0, 1);
2865 const STRLEN tlen=len0+len1+len2;
2866 PerlIO_printf(Perl_debug_log,
2867 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2868 (IV)(locinput - loc_bostr),
2871 (docolor ? "" : "> <"),
2873 (int)(tlen > 19 ? 0 : 19 - tlen),
2880 /* reg_check_named_buff_matched()
2881 * Checks to see if a named buffer has matched. The data array of
2882 * buffer numbers corresponding to the buffer is expected to reside
2883 * in the regexp->data->data array in the slot stored in the ARG() of
2884 * node involved. Note that this routine doesn't actually care about the
2885 * name, that information is not preserved from compilation to execution.
2886 * Returns the index of the leftmost defined buffer with the given name
2887 * or 0 if non of the buffers matched.
2890 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2893 RXi_GET_DECL(rex,rexi);
2894 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2895 I32 *nums=(I32*)SvPVX(sv_dat);
2897 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2899 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2900 if ((I32)*PL_reglastparen >= nums[n] &&
2901 PL_regoffs[nums[n]].end != -1)
2910 /* free all slabs above current one - called during LEAVE_SCOPE */
2913 S_clear_backtrack_stack(pTHX_ void *p)
2915 regmatch_slab *s = PL_regmatch_slab->next;
2920 PL_regmatch_slab->next = NULL;
2922 regmatch_slab * const osl = s;
2929 #define SETREX(Re1,Re2) \
2930 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2933 STATIC I32 /* 0 failure, 1 success */
2934 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2936 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2940 register const bool utf8_target = PL_reg_match_utf8;
2941 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2942 REGEXP *rex_sv = reginfo->prog;
2943 regexp *rex = (struct regexp *)SvANY(rex_sv);
2944 RXi_GET_DECL(rex,rexi);
2946 /* the current state. This is a cached copy of PL_regmatch_state */
2947 register regmatch_state *st;
2948 /* cache heavy used fields of st in registers */
2949 register regnode *scan;
2950 register regnode *next;
2951 register U32 n = 0; /* general value; init to avoid compiler warning */
2952 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2953 register char *locinput = PL_reginput;
2954 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2956 bool result = 0; /* return value of S_regmatch */
2957 int depth = 0; /* depth of backtrack stack */
2958 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2959 const U32 max_nochange_depth =
2960 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2961 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2962 regmatch_state *yes_state = NULL; /* state to pop to on success of
2964 /* mark_state piggy backs on the yes_state logic so that when we unwind
2965 the stack on success we can update the mark_state as we go */
2966 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2967 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2968 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2970 bool no_final = 0; /* prevent failure from backtracking? */
2971 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2972 char *startpoint = PL_reginput;
2973 SV *popmark = NULL; /* are we looking for a mark? */
2974 SV *sv_commit = NULL; /* last mark name seen in failure */
2975 SV *sv_yes_mark = NULL; /* last mark name we have seen
2976 during a successfull match */
2977 U32 lastopen = 0; /* last open we saw */
2978 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2979 SV* const oreplsv = GvSV(PL_replgv);
2980 /* these three flags are set by various ops to signal information to
2981 * the very next op. They have a useful lifetime of exactly one loop
2982 * iteration, and are not preserved or restored by state pushes/pops
2984 bool sw = 0; /* the condition value in (?(cond)a|b) */
2985 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2986 int logical = 0; /* the following EVAL is:
2990 or the following IFMATCH/UNLESSM is:
2991 false: plain (?=foo)
2992 true: used as a condition: (?(?=foo))
2995 GET_RE_DEBUG_FLAGS_DECL;
2998 PERL_ARGS_ASSERT_REGMATCH;
3000 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3001 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3003 /* on first ever call to regmatch, allocate first slab */
3004 if (!PL_regmatch_slab) {
3005 Newx(PL_regmatch_slab, 1, regmatch_slab);
3006 PL_regmatch_slab->prev = NULL;
3007 PL_regmatch_slab->next = NULL;
3008 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3011 oldsave = PL_savestack_ix;
3012 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3013 SAVEVPTR(PL_regmatch_slab);
3014 SAVEVPTR(PL_regmatch_state);
3016 /* grab next free state slot */
3017 st = ++PL_regmatch_state;
3018 if (st > SLAB_LAST(PL_regmatch_slab))
3019 st = PL_regmatch_state = S_push_slab(aTHX);
3021 /* Note that nextchr is a byte even in UTF */
3022 nextchr = UCHARAT(locinput);
3024 while (scan != NULL) {
3027 SV * const prop = sv_newmortal();
3028 regnode *rnext=regnext(scan);
3029 DUMP_EXEC_POS( locinput, scan, utf8_target );
3030 regprop(rex, prop, scan);
3032 PerlIO_printf(Perl_debug_log,
3033 "%3"IVdf":%*s%s(%"IVdf")\n",
3034 (IV)(scan - rexi->program), depth*2, "",
3036 (PL_regkind[OP(scan)] == END || !rnext) ?
3037 0 : (IV)(rnext - rexi->program));
3040 next = scan + NEXT_OFF(scan);
3043 state_num = OP(scan);
3047 assert(PL_reglastparen == &rex->lastparen);
3048 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3049 assert(PL_regoffs == rex->offs);
3051 switch (state_num) {
3053 if (locinput == PL_bostr)
3055 /* reginfo->till = reginfo->bol; */
3060 if (locinput == PL_bostr ||
3061 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3067 if (locinput == PL_bostr)
3071 if (locinput == reginfo->ganch)
3076 /* update the startpoint */
3077 st->u.keeper.val = PL_regoffs[0].start;
3078 PL_reginput = locinput;
3079 PL_regoffs[0].start = locinput - PL_bostr;
3080 PUSH_STATE_GOTO(KEEPS_next, next);
3082 case KEEPS_next_fail:
3083 /* rollback the start point change */
3084 PL_regoffs[0].start = st->u.keeper.val;
3090 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3095 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3097 if (PL_regeol - locinput > 1)
3101 if (PL_regeol != locinput)
3105 if (!nextchr && locinput >= PL_regeol)
3108 locinput += PL_utf8skip[nextchr];
3109 if (locinput > PL_regeol)
3111 nextchr = UCHARAT(locinput);
3114 nextchr = UCHARAT(++locinput);
3117 if (!nextchr && locinput >= PL_regeol)
3119 nextchr = UCHARAT(++locinput);
3122 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3125 locinput += PL_utf8skip[nextchr];
3126 if (locinput > PL_regeol)
3128 nextchr = UCHARAT(locinput);
3131 nextchr = UCHARAT(++locinput);
3135 #define ST st->u.trie
3137 /* In this case the charclass data is available inline so
3138 we can fail fast without a lot of extra overhead.
3140 if (scan->flags == EXACT || !utf8_target) {
3141 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3143 PerlIO_printf(Perl_debug_log,
3144 "%*s %sfailed to match trie start class...%s\n",
3145 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3153 /* the basic plan of execution of the trie is:
3154 * At the beginning, run though all the states, and
3155 * find the longest-matching word. Also remember the position
3156 * of the shortest matching word. For example, this pattern:
3159 * when matched against the string "abcde", will generate
3160 * accept states for all words except 3, with the longest
3161 * matching word being 4, and the shortest being 1 (with
3162 * the position being after char 1 of the string).
3164 * Then for each matching word, in word order (i.e. 1,2,4,5),
3165 * we run the remainder of the pattern; on each try setting
3166 * the current position to the character following the word,
3167 * returning to try the next word on failure.
3169 * We avoid having to build a list of words at runtime by
3170 * using a compile-time structure, wordinfo[].prev, which
3171 * gives, for each word, the previous accepting word (if any).
3172 * In the case above it would contain the mappings 1->2, 2->0,
3173 * 3->0, 4->5, 5->1. We can use this table to generate, from
3174 * the longest word (4 above), a list of all words, by
3175 * following the list of prev pointers; this gives us the
3176 * unordered list 4,5,1,2. Then given the current word we have
3177 * just tried, we can go through the list and find the
3178 * next-biggest word to try (so if we just failed on word 2,
3179 * the next in the list is 4).
3181 * Since at runtime we don't record the matching position in
3182 * the string for each word, we have to work that out for
3183 * each word we're about to process. The wordinfo table holds
3184 * the character length of each word; given that we recorded
3185 * at the start: the position of the shortest word and its
3186 * length in chars, we just need to move the pointer the
3187 * difference between the two char lengths. Depending on
3188 * Unicode status and folding, that's cheap or expensive.
3190 * This algorithm is optimised for the case where are only a
3191 * small number of accept states, i.e. 0,1, or maybe 2.
3192 * With lots of accepts states, and having to try all of them,
3193 * it becomes quadratic on number of accept states to find all
3198 /* what type of TRIE am I? (utf8 makes this contextual) */
3199 DECL_TRIE_TYPE(scan);
3201 /* what trie are we using right now */
3202 reg_trie_data * const trie
3203 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3204 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3205 U32 state = trie->startstate;
3207 if (trie->bitmap && trie_type != trie_utf8_fold &&
3208 !TRIE_BITMAP_TEST(trie,*locinput)
3210 if (trie->states[ state ].wordnum) {
3212 PerlIO_printf(Perl_debug_log,
3213 "%*s %smatched empty string...%s\n",
3214 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3219 PerlIO_printf(Perl_debug_log,
3220 "%*s %sfailed to match trie start class...%s\n",
3221 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3228 U8 *uc = ( U8* )locinput;
3232 U8 *uscan = (U8*)NULL;
3233 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3234 U32 charcount = 0; /* how many input chars we have matched */
3235 U32 accepted = 0; /* have we seen any accepting states? */
3238 ST.jump = trie->jump;
3241 ST.longfold = FALSE; /* char longer if folded => it's harder */
3244 /* fully traverse the TRIE; note the position of the
3245 shortest accept state and the wordnum of the longest
3248 while ( state && uc <= (U8*)PL_regeol ) {
3249 U32 base = trie->states[ state ].trans.base;
3253 wordnum = trie->states[ state ].wordnum;
3255 if (wordnum) { /* it's an accept state */
3258 /* record first match position */
3260 ST.firstpos = (U8*)locinput;
3265 ST.firstchars = charcount;
3268 if (!ST.nextword || wordnum < ST.nextword)
3269 ST.nextword = wordnum;
3270 ST.topword = wordnum;
3273 DEBUG_TRIE_EXECUTE_r({
3274 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3275 PerlIO_printf( Perl_debug_log,
3276 "%*s %sState: %4"UVxf" Accepted: %c ",
3277 2+depth * 2, "", PL_colors[4],
3278 (UV)state, (accepted ? 'Y' : 'N'));
3281 /* read a char and goto next state */
3284 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3285 uscan, len, uvc, charid, foldlen,
3292 base + charid - 1 - trie->uniquecharcount)) >= 0)
3294 && ((U32)offset < trie->lasttrans)
3295 && trie->trans[offset].check == state)
3297 state = trie->trans[offset].next;
3308 DEBUG_TRIE_EXECUTE_r(
3309 PerlIO_printf( Perl_debug_log,
3310 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3311 charid, uvc, (UV)state, PL_colors[5] );
3317 /* calculate total number of accept states */
3322 w = trie->wordinfo[w].prev;
3325 ST.accepted = accepted;
3329 PerlIO_printf( Perl_debug_log,
3330 "%*s %sgot %"IVdf" possible matches%s\n",
3331 REPORT_CODE_OFF + depth * 2, "",
3332 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3334 goto trie_first_try; /* jump into the fail handler */
3338 case TRIE_next_fail: /* we failed - try next alternative */
3340 REGCP_UNWIND(ST.cp);
3341 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3342 PL_regoffs[n].end = -1;
3343 *PL_reglastparen = n;
3345 if (!--ST.accepted) {
3347 PerlIO_printf( Perl_debug_log,
3348 "%*s %sTRIE failed...%s\n",
3349 REPORT_CODE_OFF+depth*2, "",
3356 /* Find next-highest word to process. Note that this code
3357 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3358 register U16 min = 0;
3360 register U16 const nextword = ST.nextword;
3361 register reg_trie_wordinfo * const wordinfo
3362 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3363 for (word=ST.topword; word; word=wordinfo[word].prev) {
3364 if (word > nextword && (!min || word < min))
3377 ST.lastparen = *PL_reglastparen;
3381 /* find start char of end of current word */
3383 U32 chars; /* how many chars to skip */
3384 U8 *uc = ST.firstpos;
3385 reg_trie_data * const trie
3386 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3388 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3390 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3394 /* the hard option - fold each char in turn and find
3395 * its folded length (which may be different */
3396 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3404 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3412 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3417 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3431 PL_reginput = (char *)uc;
3434 scan = (ST.jump && ST.jump[ST.nextword])
3435 ? ST.me + ST.jump[ST.nextword]
3439 PerlIO_printf( Perl_debug_log,
3440 "%*s %sTRIE matched word #%d, continuing%s\n",
3441 REPORT_CODE_OFF+depth*2, "",
3448 if (ST.accepted > 1 || has_cutgroup) {
3449 PUSH_STATE_GOTO(TRIE_next, scan);
3452 /* only one choice left - just continue */
3454 AV *const trie_words
3455 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3456 SV ** const tmp = av_fetch( trie_words,
3458 SV *sv= tmp ? sv_newmortal() : NULL;
3460 PerlIO_printf( Perl_debug_log,
3461 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3462 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3464 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3465 PL_colors[0], PL_colors[1],
3466 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3468 : "not compiled under -Dr",
3472 locinput = PL_reginput;
3473 nextchr = UCHARAT(locinput);
3474 continue; /* execute rest of RE */
3479 char *s = STRING(scan);
3481 if (utf8_target != UTF_PATTERN) {
3482 /* The target and the pattern have differing utf8ness. */
3484 const char * const e = s + ln;
3487 /* The target is utf8, the pattern is not utf8. */
3492 if (NATIVE_TO_UNI(*(U8*)s) !=
3493 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3501 /* The target is not utf8, the pattern is utf8. */
3506 if (NATIVE_TO_UNI(*((U8*)l)) !=
3507 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3515 nextchr = UCHARAT(locinput);
3518 /* The target and the pattern have the same utf8ness. */
3519 /* Inline the first character, for speed. */
3520 if (UCHARAT(s) != nextchr)
3522 if (PL_regeol - locinput < ln)
3524 if (ln > 1 && memNE(s, locinput, ln))
3527 nextchr = UCHARAT(locinput);
3531 PL_reg_flags |= RF_tainted;
3534 char * const s = STRING(scan);
3537 if (utf8_target || UTF_PATTERN) {
3538 /* Either target or the pattern are utf8. */
3539 const char * const l = locinput;
3540 char *e = PL_regeol;
3542 if (! foldEQ_utf8(s, 0, ln, cBOOL(UTF_PATTERN),
3543 l, &e, 0, utf8_target)) {
3544 /* One more case for the sharp s:
3545 * pack("U0U*", 0xDF) =~ /ss/i,
3546 * the 0xC3 0x9F are the UTF-8
3547 * byte sequence for the U+00DF. */
3549 if (!(utf8_target &&
3550 toLOWER(s[0]) == 's' &&
3552 toLOWER(s[1]) == 's' &&
3559 nextchr = UCHARAT(locinput);
3563 /* Neither the target and the pattern are utf8. */
3565 /* Inline the first character, for speed. */
3566 if (UCHARAT(s) != nextchr &&
3567 UCHARAT(s) != ((OP(scan) == EXACTF)
3568 ? PL_fold : PL_fold_locale)[nextchr])
3570 if (PL_regeol - locinput < ln)
3572 if (ln > 1 && (OP(scan) == EXACTF
3573 ? ! foldEQ(s, locinput, ln)
3574 : ! foldEQ_locale(s, locinput, ln)))
3577 nextchr = UCHARAT(locinput);
3582 PL_reg_flags |= RF_tainted;
3586 /* was last char in word? */
3588 if (locinput == PL_bostr)
3591 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3593 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3595 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3596 ln = isALNUM_uni(ln);
3597 LOAD_UTF8_CHARCLASS_ALNUM();
3598 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3601 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3602 n = isALNUM_LC_utf8((U8*)locinput);
3606 ln = (locinput != PL_bostr) ?
3607 UCHARAT(locinput - 1) : '\n';
3608 if (FLAGS(scan) & USE_UNI) {
3610 /* Here, can't be BOUNDL or NBOUNDL because they never set
3611 * the flags to USE_UNI */
3612 ln = isWORDCHAR_L1(ln);
3613 n = isWORDCHAR_L1(nextchr);
3615 else if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3617 n = isALNUM(nextchr);
3620 ln = isALNUM_LC(ln);
3621 n = isALNUM_LC(nextchr);
3624 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3625 OP(scan) == BOUNDL))
3630 STRLEN inclasslen = PL_regeol - locinput;
3632 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3634 if (locinput >= PL_regeol)
3636 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3637 nextchr = UCHARAT(locinput);
3642 nextchr = UCHARAT(locinput);
3643 if (!REGINCLASS(rex, scan, (U8*)locinput))
3645 if (!nextchr && locinput >= PL_regeol)
3647 nextchr = UCHARAT(++locinput);
3651 /* If we might have the case of the German sharp s
3652 * in a casefolding Unicode character class. */
3654 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3655 locinput += SHARP_S_SKIP;
3656 nextchr = UCHARAT(locinput);
3661 /* Special char classes - The defines start on line 129 or so */
3662 CCC_TRY_AFF_U( ALNUM, ALNUML, perl_word, "a", isALNUM_LC_utf8, isWORDCHAR_L1, isALNUM_LC);
3663 CCC_TRY_NEG_U(NALNUM, NALNUML, perl_word, "a", isALNUM_LC_utf8, isWORDCHAR_L1, isALNUM_LC);
3665 CCC_TRY_AFF_U( SPACE, SPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE_L1, isSPACE_LC);
3666 CCC_TRY_NEG_U(NSPACE, NSPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE_L1, isSPACE_LC);
3668 CCC_TRY_AFF( DIGIT, DIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3669 CCC_TRY_NEG(NDIGIT, NDIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3671 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3672 a Unicode extended Grapheme Cluster */
3673 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3674 extended Grapheme Cluster is:
3677 | Prepend* Begin Extend*
3680 Begin is (Hangul-syllable | ! Control)
3681 Extend is (Grapheme_Extend | Spacing_Mark)
3682 Control is [ GCB_Control CR LF ]
3684 The discussion below shows how the code for CLUMP is derived
3685 from this regex. Note that most of these concepts are from
3686 property values of the Grapheme Cluster Boundary (GCB) property.
3687 No code point can have multiple property values for a given
3688 property. Thus a code point in Prepend can't be in Control, but
3689 it must be in !Control. This is why Control above includes
3690 GCB_Control plus CR plus LF. The latter two are used in the GCB
3691 property separately, and so can't be in GCB_Control, even though
3692 they logically are controls. Control is not the same as gc=cc,
3693 but includes format and other characters as well.
3695 The Unicode definition of Hangul-syllable is:
3697 | (L* ( ( V | LV ) V* | LVT ) T*)
3700 Each of these is a value for the GCB property, and hence must be
3701 disjoint, so the order they are tested is immaterial, so the
3702 above can safely be changed to
3705 | (L* ( LVT | ( V | LV ) V*) T*)
3707 The last two terms can be combined like this:
3709 | (( LVT | ( V | LV ) V*) T*))
3711 And refactored into this:
3712 L* (L | LVT T* | V V* T* | LV V* T*)
3714 That means that if we have seen any L's at all we can quit
3715 there, but if the next character is a LVT, a V or and LV we
3718 There is a subtlety with Prepend* which showed up in testing.
3719 Note that the Begin, and only the Begin is required in:
3720 | Prepend* Begin Extend*
3721 Also, Begin contains '! Control'. A Prepend must be a '!
3722 Control', which means it must be a Begin. What it comes down to
3723 is that if we match Prepend* and then find no suitable Begin
3724 afterwards, that if we backtrack the last Prepend, that one will
3725 be a suitable Begin.
3728 if (locinput >= PL_regeol)
3730 if (! utf8_target) {
3732 /* Match either CR LF or '.', as all the other possibilities
3734 locinput++; /* Match the . or CR */
3736 && locinput < PL_regeol
3737 && UCHARAT(locinput) == '\n') locinput++;
3741 /* Utf8: See if is ( CR LF ); already know that locinput <
3742 * PL_regeol, so locinput+1 is in bounds */
3743 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3747 /* In case have to backtrack to beginning, then match '.' */
3748 char *starting = locinput;
3750 /* In case have to backtrack the last prepend */
3751 char *previous_prepend = 0;
3753 LOAD_UTF8_CHARCLASS_GCB();
3755 /* Match (prepend)* */
3756 while (locinput < PL_regeol
3757 && swash_fetch(PL_utf8_X_prepend,
3758 (U8*)locinput, utf8_target))
3760 previous_prepend = locinput;
3761 locinput += UTF8SKIP(locinput);
3764 /* As noted above, if we matched a prepend character, but
3765 * the next thing won't match, back off the last prepend we
3766 * matched, as it is guaranteed to match the begin */
3767 if (previous_prepend
3768 && (locinput >= PL_regeol
3769 || ! swash_fetch(PL_utf8_X_begin,
3770 (U8*)locinput, utf8_target)))
3772 locinput = previous_prepend;
3775 /* Note that here we know PL_regeol > locinput, as we
3776 * tested that upon input to this switch case, and if we
3777 * moved locinput forward, we tested the result just above
3778 * and it either passed, or we backed off so that it will
3780 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3782 /* Here did not match the required 'Begin' in the
3783 * second term. So just match the very first
3784 * character, the '.' of the final term of the regex */
3785 locinput = starting + UTF8SKIP(starting);
3788 /* Here is the beginning of a character that can have
3789 * an extender. It is either a hangul syllable, or a
3791 if (swash_fetch(PL_utf8_X_non_hangul,
3792 (U8*)locinput, utf8_target))
3795 /* Here not a Hangul syllable, must be a
3796 * ('! * Control') */
3797 locinput += UTF8SKIP(locinput);
3800 /* Here is a Hangul syllable. It can be composed
3801 * of several individual characters. One
3802 * possibility is T+ */
3803 if (swash_fetch(PL_utf8_X_T,
3804 (U8*)locinput, utf8_target))
3806 while (locinput < PL_regeol
3807 && swash_fetch(PL_utf8_X_T,
3808 (U8*)locinput, utf8_target))
3810 locinput += UTF8SKIP(locinput);
3814 /* Here, not T+, but is a Hangul. That means
3815 * it is one of the others: L, LV, LVT or V,
3817 * L* (L | LVT T* | V V* T* | LV V* T*) */
3820 while (locinput < PL_regeol
3821 && swash_fetch(PL_utf8_X_L,
3822 (U8*)locinput, utf8_target))
3824 locinput += UTF8SKIP(locinput);
3827 /* Here, have exhausted L*. If the next
3828 * character is not an LV, LVT nor V, it means
3829 * we had to have at least one L, so matches L+
3830 * in the original equation, we have a complete
3831 * hangul syllable. Are done. */
3833 if (locinput < PL_regeol
3834 && swash_fetch(PL_utf8_X_LV_LVT_V,
3835 (U8*)locinput, utf8_target))
3838 /* Otherwise keep going. Must be LV, LVT
3839 * or V. See if LVT */
3840 if (swash_fetch(PL_utf8_X_LVT,