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 ((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) (do_utf8 ? 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(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
181 PL_reg_flags |= RF_tainted; \
186 if (do_utf8 && 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, do_utf8)) \
197 : LCFUNC_utf8((U8*)locinput))) \
201 locinput += PL_utf8skip[nextchr]; \
202 nextchr = UCHARAT(locinput); \
205 if (!(OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
207 nextchr = UCHARAT(++locinput); \
210 #define CCC_TRY_NEG(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
212 PL_reg_flags |= RF_tainted; \
215 if (!nextchr && locinput >= PL_regeol) \
217 if (do_utf8 && UTF8_IS_CONTINUED(nextchr)) { \
218 if (!CAT2(PL_utf8_,CLASS)) { \
222 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
226 if ((OP(scan) == NAME \
227 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, do_utf8)) \
228 : LCFUNC_utf8((U8*)locinput))) \
232 locinput += PL_utf8skip[nextchr]; \
233 nextchr = UCHARAT(locinput); \
236 if ((OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
238 nextchr = UCHARAT(++locinput); \
245 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
247 /* for use after a quantifier and before an EXACT-like node -- japhy */
248 /* it would be nice to rework regcomp.sym to generate this stuff. sigh */
249 #define JUMPABLE(rn) ( \
251 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
253 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
254 OP(rn) == PLUS || OP(rn) == MINMOD || \
255 OP(rn) == KEEPS || (PL_regkind[OP(rn)] == VERB) || \
256 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
258 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
260 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
263 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
264 we don't need this definition. */
265 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
266 #define IS_TEXTF(rn) ( OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
267 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
270 /* ... so we use this as its faster. */
271 #define IS_TEXT(rn) ( OP(rn)==EXACT )
272 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
273 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
278 Search for mandatory following text node; for lookahead, the text must
279 follow but for lookbehind (rn->flags != 0) we skip to the next step.
281 #define FIND_NEXT_IMPT(rn) STMT_START { \
282 while (JUMPABLE(rn)) { \
283 const OPCODE type = OP(rn); \
284 if (type == SUSPEND || PL_regkind[type] == CURLY) \
285 rn = NEXTOPER(NEXTOPER(rn)); \
286 else if (type == PLUS) \
288 else if (type == IFMATCH) \
289 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
290 else rn += NEXT_OFF(rn); \
295 static void restore_pos(pTHX_ void *arg);
297 #define REGCP_PAREN_ELEMS 4
298 #define REGCP_OTHER_ELEMS 5
299 #define REGCP_FRAME_ELEMS 1
300 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
301 * are needed for the regexp context stack bookkeeping. */
304 S_regcppush(pTHX_ I32 parenfloor)
307 const int retval = PL_savestack_ix;
308 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
309 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
310 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
312 GET_RE_DEBUG_FLAGS_DECL;
314 if (paren_elems_to_push < 0)
315 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
317 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
318 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
319 " out of range (%lu-%ld)",
320 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
322 SSGROW(total_elems + REGCP_FRAME_ELEMS);
324 for (p = PL_regsize; p > parenfloor; p--) {
325 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
326 SSPUSHINT(PL_regoffs[p].end);
327 SSPUSHINT(PL_regoffs[p].start);
328 SSPUSHPTR(PL_reg_start_tmp[p]);
330 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
331 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
332 (UV)p, (IV)PL_regoffs[p].start,
333 (IV)(PL_reg_start_tmp[p] - PL_bostr),
334 (IV)PL_regoffs[p].end
337 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
338 SSPUSHPTR(PL_regoffs);
339 SSPUSHINT(PL_regsize);
340 SSPUSHINT(*PL_reglastparen);
341 SSPUSHINT(*PL_reglastcloseparen);
342 SSPUSHPTR(PL_reginput);
343 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
348 /* These are needed since we do not localize EVAL nodes: */
349 #define REGCP_SET(cp) \
351 PerlIO_printf(Perl_debug_log, \
352 " Setting an EVAL scope, savestack=%"IVdf"\n", \
353 (IV)PL_savestack_ix)); \
356 #define REGCP_UNWIND(cp) \
358 if (cp != PL_savestack_ix) \
359 PerlIO_printf(Perl_debug_log, \
360 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
361 (IV)(cp), (IV)PL_savestack_ix)); \
365 S_regcppop(pTHX_ const regexp *rex)
370 GET_RE_DEBUG_FLAGS_DECL;
372 PERL_ARGS_ASSERT_REGCPPOP;
374 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
376 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
377 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
378 input = (char *) SSPOPPTR;
379 *PL_reglastcloseparen = SSPOPINT;
380 *PL_reglastparen = SSPOPINT;
381 PL_regsize = SSPOPINT;
382 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
384 i -= REGCP_OTHER_ELEMS;
385 /* Now restore the parentheses context. */
386 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
388 U32 paren = (U32)SSPOPINT;
389 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
390 PL_regoffs[paren].start = SSPOPINT;
392 if (paren <= *PL_reglastparen)
393 PL_regoffs[paren].end = tmps;
395 PerlIO_printf(Perl_debug_log,
396 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
397 (UV)paren, (IV)PL_regoffs[paren].start,
398 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
399 (IV)PL_regoffs[paren].end,
400 (paren > *PL_reglastparen ? "(no)" : ""));
404 if (*PL_reglastparen + 1 <= rex->nparens) {
405 PerlIO_printf(Perl_debug_log,
406 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
407 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
411 /* It would seem that the similar code in regtry()
412 * already takes care of this, and in fact it is in
413 * a better location to since this code can #if 0-ed out
414 * but the code in regtry() is needed or otherwise tests
415 * requiring null fields (pat.t#187 and split.t#{13,14}
416 * (as of patchlevel 7877) will fail. Then again,
417 * this code seems to be necessary or otherwise
418 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
419 * --jhi updated by dapm */
420 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
422 PL_regoffs[i].start = -1;
423 PL_regoffs[i].end = -1;
429 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
432 * pregexec and friends
435 #ifndef PERL_IN_XSUB_RE
437 - pregexec - match a regexp against a string
440 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
441 char *strbeg, I32 minend, SV *screamer, U32 nosave)
442 /* strend: pointer to null at end of string */
443 /* strbeg: real beginning of string */
444 /* minend: end of match must be >=minend after stringarg. */
445 /* nosave: For optimizations. */
447 PERL_ARGS_ASSERT_PREGEXEC;
450 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
451 nosave ? 0 : REXEC_COPY_STR);
456 * Need to implement the following flags for reg_anch:
458 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
460 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
461 * INTUIT_AUTORITATIVE_ML
462 * INTUIT_ONCE_NOML - Intuit can match in one location only.
465 * Another flag for this function: SECOND_TIME (so that float substrs
466 * with giant delta may be not rechecked).
469 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
471 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
472 Otherwise, only SvCUR(sv) is used to get strbeg. */
474 /* XXXX We assume that strpos is strbeg unless sv. */
476 /* XXXX Some places assume that there is a fixed substring.
477 An update may be needed if optimizer marks as "INTUITable"
478 RExen without fixed substrings. Similarly, it is assumed that
479 lengths of all the strings are no more than minlen, thus they
480 cannot come from lookahead.
481 (Or minlen should take into account lookahead.)
482 NOTE: Some of this comment is not correct. minlen does now take account
483 of lookahead/behind. Further research is required. -- demerphq
487 /* A failure to find a constant substring means that there is no need to make
488 an expensive call to REx engine, thus we celebrate a failure. Similarly,
489 finding a substring too deep into the string means that less calls to
490 regtry() should be needed.
492 REx compiler's optimizer found 4 possible hints:
493 a) Anchored substring;
495 c) Whether we are anchored (beginning-of-line or \G);
496 d) First node (of those at offset 0) which may distingush positions;
497 We use a)b)d) and multiline-part of c), and try to find a position in the
498 string which does not contradict any of them.
501 /* Most of decisions we do here should have been done at compile time.
502 The nodes of the REx which we used for the search should have been
503 deleted from the finite automaton. */
506 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
507 char *strend, const U32 flags, re_scream_pos_data *data)
510 struct regexp *const prog = (struct regexp *)SvANY(rx);
511 register I32 start_shift = 0;
512 /* Should be nonnegative! */
513 register I32 end_shift = 0;
518 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
520 register char *other_last = NULL; /* other substr checked before this */
521 char *check_at = NULL; /* check substr found at this pos */
522 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
523 RXi_GET_DECL(prog,progi);
525 const char * const i_strpos = strpos;
527 GET_RE_DEBUG_FLAGS_DECL;
529 PERL_ARGS_ASSERT_RE_INTUIT_START;
531 RX_MATCH_UTF8_set(rx,do_utf8);
534 PL_reg_flags |= RF_utf8;
537 debug_start_match(rx, do_utf8, strpos, strend,
538 sv ? "Guessing start of match in sv for"
539 : "Guessing start of match in string for");
542 /* CHR_DIST() would be more correct here but it makes things slow. */
543 if (prog->minlen > strend - strpos) {
544 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
545 "String too short... [re_intuit_start]\n"));
549 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
552 if (!prog->check_utf8 && prog->check_substr)
553 to_utf8_substr(prog);
554 check = prog->check_utf8;
556 if (!prog->check_substr && prog->check_utf8)
557 to_byte_substr(prog);
558 check = prog->check_substr;
560 if (check == &PL_sv_undef) {
561 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
562 "Non-utf8 string cannot match utf8 check string\n"));
565 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
566 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
567 || ( (prog->extflags & RXf_ANCH_BOL)
568 && !multiline ) ); /* Check after \n? */
571 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
572 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
573 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
575 && (strpos != strbeg)) {
576 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
579 if (prog->check_offset_min == prog->check_offset_max &&
580 !(prog->extflags & RXf_CANY_SEEN)) {
581 /* Substring at constant offset from beg-of-str... */
584 s = HOP3c(strpos, prog->check_offset_min, strend);
587 slen = SvCUR(check); /* >= 1 */
589 if ( strend - s > slen || strend - s < slen - 1
590 || (strend - s == slen && strend[-1] != '\n')) {
591 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
594 /* Now should match s[0..slen-2] */
596 if (slen && (*SvPVX_const(check) != *s
598 && memNE(SvPVX_const(check), s, slen)))) {
600 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
604 else if (*SvPVX_const(check) != *s
605 || ((slen = SvCUR(check)) > 1
606 && memNE(SvPVX_const(check), s, slen)))
609 goto success_at_start;
612 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
614 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
615 end_shift = prog->check_end_shift;
618 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
619 - (SvTAIL(check) != 0);
620 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
622 if (end_shift < eshift)
626 else { /* Can match at random position */
629 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
630 end_shift = prog->check_end_shift;
632 /* end shift should be non negative here */
635 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
637 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
638 (IV)end_shift, RX_PRECOMP(prog));
642 /* Find a possible match in the region s..strend by looking for
643 the "check" substring in the region corrected by start/end_shift. */
646 I32 srch_start_shift = start_shift;
647 I32 srch_end_shift = end_shift;
648 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
649 srch_end_shift -= ((strbeg - s) - srch_start_shift);
650 srch_start_shift = strbeg - s;
652 DEBUG_OPTIMISE_MORE_r({
653 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
654 (IV)prog->check_offset_min,
655 (IV)srch_start_shift,
657 (IV)prog->check_end_shift);
660 if (flags & REXEC_SCREAM) {
661 I32 p = -1; /* Internal iterator of scream. */
662 I32 * const pp = data ? data->scream_pos : &p;
664 if (PL_screamfirst[BmRARE(check)] >= 0
665 || ( BmRARE(check) == '\n'
666 && (BmPREVIOUS(check) == SvCUR(check) - 1)
668 s = screaminstr(sv, check,
669 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
672 /* we may be pointing at the wrong string */
673 if (s && RXp_MATCH_COPIED(prog))
674 s = strbeg + (s - SvPVX_const(sv));
676 *data->scream_olds = s;
681 if (prog->extflags & RXf_CANY_SEEN) {
682 start_point= (U8*)(s + srch_start_shift);
683 end_point= (U8*)(strend - srch_end_shift);
685 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
686 end_point= HOP3(strend, -srch_end_shift, strbeg);
688 DEBUG_OPTIMISE_MORE_r({
689 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
690 (int)(end_point - start_point),
691 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
695 s = fbm_instr( start_point, end_point,
696 check, multiline ? FBMrf_MULTILINE : 0);
699 /* Update the count-of-usability, remove useless subpatterns,
703 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
704 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
705 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
706 (s ? "Found" : "Did not find"),
707 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
708 ? "anchored" : "floating"),
711 (s ? " at offset " : "...\n") );
716 /* Finish the diagnostic message */
717 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
719 /* XXX dmq: first branch is for positive lookbehind...
720 Our check string is offset from the beginning of the pattern.
721 So we need to do any stclass tests offset forward from that
730 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
731 Start with the other substr.
732 XXXX no SCREAM optimization yet - and a very coarse implementation
733 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
734 *always* match. Probably should be marked during compile...
735 Probably it is right to do no SCREAM here...
738 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
739 : (prog->float_substr && prog->anchored_substr))
741 /* Take into account the "other" substring. */
742 /* XXXX May be hopelessly wrong for UTF... */
745 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
748 char * const last = HOP3c(s, -start_shift, strbeg);
750 char * const saved_s = s;
753 t = s - prog->check_offset_max;
754 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
756 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
761 t = HOP3c(t, prog->anchored_offset, strend);
762 if (t < other_last) /* These positions already checked */
764 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
767 /* XXXX It is not documented what units *_offsets are in.
768 We assume bytes, but this is clearly wrong.
769 Meaning this code needs to be carefully reviewed for errors.
773 /* On end-of-str: see comment below. */
774 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
775 if (must == &PL_sv_undef) {
777 DEBUG_r(must = prog->anchored_utf8); /* for debug */
782 HOP3(HOP3(last1, prog->anchored_offset, strend)
783 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
785 multiline ? FBMrf_MULTILINE : 0
788 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
789 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
790 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
791 (s ? "Found" : "Contradicts"),
792 quoted, RE_SV_TAIL(must));
797 if (last1 >= last2) {
798 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
799 ", giving up...\n"));
802 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
803 ", trying floating at offset %ld...\n",
804 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
805 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
806 s = HOP3c(last, 1, strend);
810 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
811 (long)(s - i_strpos)));
812 t = HOP3c(s, -prog->anchored_offset, strbeg);
813 other_last = HOP3c(s, 1, strend);
821 else { /* Take into account the floating substring. */
823 char * const saved_s = s;
826 t = HOP3c(s, -start_shift, strbeg);
828 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
829 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
830 last = HOP3c(t, prog->float_max_offset, strend);
831 s = HOP3c(t, prog->float_min_offset, strend);
834 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
835 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
836 /* fbm_instr() takes into account exact value of end-of-str
837 if the check is SvTAIL(ed). Since false positives are OK,
838 and end-of-str is not later than strend we are OK. */
839 if (must == &PL_sv_undef) {
841 DEBUG_r(must = prog->float_utf8); /* for debug message */
844 s = fbm_instr((unsigned char*)s,
845 (unsigned char*)last + SvCUR(must)
847 must, multiline ? FBMrf_MULTILINE : 0);
849 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
850 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
851 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
852 (s ? "Found" : "Contradicts"),
853 quoted, RE_SV_TAIL(must));
857 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
858 ", giving up...\n"));
861 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
862 ", trying anchored starting at offset %ld...\n",
863 (long)(saved_s + 1 - i_strpos)));
865 s = HOP3c(t, 1, strend);
869 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
870 (long)(s - i_strpos)));
871 other_last = s; /* Fix this later. --Hugo */
881 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
883 DEBUG_OPTIMISE_MORE_r(
884 PerlIO_printf(Perl_debug_log,
885 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
886 (IV)prog->check_offset_min,
887 (IV)prog->check_offset_max,
895 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
897 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
900 /* Fixed substring is found far enough so that the match
901 cannot start at strpos. */
903 if (ml_anch && t[-1] != '\n') {
904 /* Eventually fbm_*() should handle this, but often
905 anchored_offset is not 0, so this check will not be wasted. */
906 /* XXXX In the code below we prefer to look for "^" even in
907 presence of anchored substrings. And we search even
908 beyond the found float position. These pessimizations
909 are historical artefacts only. */
911 while (t < strend - prog->minlen) {
913 if (t < check_at - prog->check_offset_min) {
914 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
915 /* Since we moved from the found position,
916 we definitely contradict the found anchored
917 substr. Due to the above check we do not
918 contradict "check" substr.
919 Thus we can arrive here only if check substr
920 is float. Redo checking for "other"=="fixed".
923 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
924 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
925 goto do_other_anchored;
927 /* We don't contradict the found floating substring. */
928 /* XXXX Why not check for STCLASS? */
930 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
931 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
934 /* Position contradicts check-string */
935 /* XXXX probably better to look for check-string
936 than for "\n", so one should lower the limit for t? */
937 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
938 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
939 other_last = strpos = s = t + 1;
944 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
945 PL_colors[0], PL_colors[1]));
949 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
950 PL_colors[0], PL_colors[1]));
954 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
957 /* The found string does not prohibit matching at strpos,
958 - no optimization of calling REx engine can be performed,
959 unless it was an MBOL and we are not after MBOL,
960 or a future STCLASS check will fail this. */
962 /* Even in this situation we may use MBOL flag if strpos is offset
963 wrt the start of the string. */
964 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
965 && (strpos != strbeg) && strpos[-1] != '\n'
966 /* May be due to an implicit anchor of m{.*foo} */
967 && !(prog->intflags & PREGf_IMPLICIT))
972 DEBUG_EXECUTE_r( if (ml_anch)
973 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
974 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
977 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
979 prog->check_utf8 /* Could be deleted already */
980 && --BmUSEFUL(prog->check_utf8) < 0
981 && (prog->check_utf8 == prog->float_utf8)
983 prog->check_substr /* Could be deleted already */
984 && --BmUSEFUL(prog->check_substr) < 0
985 && (prog->check_substr == prog->float_substr)
988 /* If flags & SOMETHING - do not do it many times on the same match */
989 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
990 /* XXX Does the destruction order has to change with do_utf8? */
991 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
992 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
993 prog->check_substr = prog->check_utf8 = NULL; /* disable */
994 prog->float_substr = prog->float_utf8 = NULL; /* clear */
995 check = NULL; /* abort */
997 /* XXXX This is a remnant of the old implementation. It
998 looks wasteful, since now INTUIT can use many
1000 prog->extflags &= ~RXf_USE_INTUIT;
1006 /* Last resort... */
1007 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1008 /* trie stclasses are too expensive to use here, we are better off to
1009 leave it to regmatch itself */
1010 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1011 /* minlen == 0 is possible if regstclass is \b or \B,
1012 and the fixed substr is ''$.
1013 Since minlen is already taken into account, s+1 is before strend;
1014 accidentally, minlen >= 1 guaranties no false positives at s + 1
1015 even for \b or \B. But (minlen? 1 : 0) below assumes that
1016 regstclass does not come from lookahead... */
1017 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1018 This leaves EXACTF only, which is dealt with in find_byclass(). */
1019 const U8* const str = (U8*)STRING(progi->regstclass);
1020 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1021 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1024 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1025 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1026 else if (prog->float_substr || prog->float_utf8)
1027 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1031 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1032 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1035 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1038 const char *what = NULL;
1040 if (endpos == strend) {
1041 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1042 "Could not match STCLASS...\n") );
1045 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1046 "This position contradicts STCLASS...\n") );
1047 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1049 /* Contradict one of substrings */
1050 if (prog->anchored_substr || prog->anchored_utf8) {
1051 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1052 DEBUG_EXECUTE_r( what = "anchored" );
1054 s = HOP3c(t, 1, strend);
1055 if (s + start_shift + end_shift > strend) {
1056 /* XXXX Should be taken into account earlier? */
1057 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1058 "Could not match STCLASS...\n") );
1063 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1064 "Looking for %s substr starting at offset %ld...\n",
1065 what, (long)(s + start_shift - i_strpos)) );
1068 /* Have both, check_string is floating */
1069 if (t + start_shift >= check_at) /* Contradicts floating=check */
1070 goto retry_floating_check;
1071 /* Recheck anchored substring, but not floating... */
1075 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1076 "Looking for anchored substr starting at offset %ld...\n",
1077 (long)(other_last - i_strpos)) );
1078 goto do_other_anchored;
1080 /* Another way we could have checked stclass at the
1081 current position only: */
1086 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1087 "Looking for /%s^%s/m starting at offset %ld...\n",
1088 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1091 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1093 /* Check is floating subtring. */
1094 retry_floating_check:
1095 t = check_at - start_shift;
1096 DEBUG_EXECUTE_r( what = "floating" );
1097 goto hop_and_restart;
1100 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1101 "By STCLASS: moving %ld --> %ld\n",
1102 (long)(t - i_strpos), (long)(s - i_strpos))
1106 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1107 "Does not contradict STCLASS...\n");
1112 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1113 PL_colors[4], (check ? "Guessed" : "Giving up"),
1114 PL_colors[5], (long)(s - i_strpos)) );
1117 fail_finish: /* Substring not found */
1118 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1119 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1121 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1122 PL_colors[4], PL_colors[5]));
1126 #define DECL_TRIE_TYPE(scan) \
1127 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1128 trie_type = (scan->flags != EXACT) \
1129 ? (do_utf8 ? trie_utf8_fold : (UTF ? trie_latin_utf8_fold : trie_plain)) \
1130 : (do_utf8 ? trie_utf8 : trie_plain)
1132 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1133 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1134 switch (trie_type) { \
1135 case trie_utf8_fold: \
1136 if ( foldlen>0 ) { \
1137 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1142 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1143 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1144 foldlen -= UNISKIP( uvc ); \
1145 uscan = foldbuf + UNISKIP( uvc ); \
1148 case trie_latin_utf8_fold: \
1149 if ( foldlen>0 ) { \
1150 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1156 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1157 foldlen -= UNISKIP( uvc ); \
1158 uscan = foldbuf + UNISKIP( uvc ); \
1162 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1169 charid = trie->charmap[ uvc ]; \
1173 if (widecharmap) { \
1174 SV** const svpp = hv_fetch(widecharmap, \
1175 (char*)&uvc, sizeof(UV), 0); \
1177 charid = (U16)SvIV(*svpp); \
1182 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1184 char *my_strend= (char *)strend; \
1187 foldEQ_utf8(s, &my_strend, 0, do_utf8, \
1188 m, NULL, ln, cBOOL(UTF))) \
1189 && (!reginfo || regtry(reginfo, &s)) ) \
1192 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1193 uvchr_to_utf8(tmpbuf, c); \
1194 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1196 && (f == c1 || f == c2) \
1198 foldEQ_utf8(s, &my_strend, 0, do_utf8,\
1199 m, NULL, ln, cBOOL(UTF)))\
1200 && (!reginfo || regtry(reginfo, &s)) ) \
1206 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1210 && (ln == 1 || (OP(c) == EXACTF \
1211 ? foldEQ(s, m, ln) \
1212 : foldEQ_locale(s, m, ln))) \
1213 && (!reginfo || regtry(reginfo, &s)) ) \
1219 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1221 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1227 #define REXEC_FBC_SCAN(CoDe) \
1229 while (s < strend) { \
1235 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1236 REXEC_FBC_UTF8_SCAN( \
1238 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1247 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1250 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1259 #define REXEC_FBC_TRYIT \
1260 if ((!reginfo || regtry(reginfo, &s))) \
1263 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1265 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1268 REXEC_FBC_CLASS_SCAN(CoNd); \
1272 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1275 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1278 REXEC_FBC_CLASS_SCAN(CoNd); \
1282 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1283 PL_reg_flags |= RF_tainted; \
1285 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1288 REXEC_FBC_CLASS_SCAN(CoNd); \
1292 #define DUMP_EXEC_POS(li,s,doutf8) \
1293 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1295 /* We know what class REx starts with. Try to find this position... */
1296 /* if reginfo is NULL, its a dryrun */
1297 /* annoyingly all the vars in this routine have different names from their counterparts
1298 in regmatch. /grrr */
1301 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1302 const char *strend, regmatch_info *reginfo)
1305 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1309 register STRLEN uskip;
1313 register I32 tmp = 1; /* Scratch variable? */
1314 register const bool do_utf8 = PL_reg_match_utf8;
1315 RXi_GET_DECL(prog,progi);
1317 PERL_ARGS_ASSERT_FIND_BYCLASS;
1319 /* We know what class it must start with. */
1323 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1324 !UTF8_IS_INVARIANT((U8)s[0]) ?
1325 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1326 REGINCLASS(prog, c, (U8*)s));
1329 while (s < strend) {
1332 if (REGINCLASS(prog, c, (U8*)s) ||
1333 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1334 /* The assignment of 2 is intentional:
1335 * for the folded sharp s, the skip is 2. */
1336 (skip = SHARP_S_SKIP))) {
1337 if (tmp && (!reginfo || regtry(reginfo, &s)))
1350 if (tmp && (!reginfo || regtry(reginfo, &s)))
1358 ln = STR_LEN(c); /* length to match in octets/bytes */
1359 lnc = (I32) ln; /* length to match in characters */
1361 STRLEN ulen1, ulen2;
1363 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1364 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1365 /* used by commented-out code below */
1366 /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
1368 /* XXX: Since the node will be case folded at compile
1369 time this logic is a little odd, although im not
1370 sure that its actually wrong. --dmq */
1372 c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1373 c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1375 /* XXX: This is kinda strange. to_utf8_XYZ returns the
1376 codepoint of the first character in the converted
1377 form, yet originally we did the extra step.
1378 No tests fail by commenting this code out however
1379 so Ive left it out. -- dmq.
1381 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1383 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1388 while (sm < ((U8 *) m + ln)) {
1403 c2 = PL_fold_locale[c1];
1405 e = HOP3c(strend, -((I32)lnc), s);
1407 if (!reginfo && e < s)
1408 e = s; /* Due to minlen logic of intuit() */
1410 /* The idea in the EXACTF* cases is to first find the
1411 * first character of the EXACTF* node and then, if
1412 * necessary, case-insensitively compare the full
1413 * text of the node. The c1 and c2 are the first
1414 * characters (though in Unicode it gets a bit
1415 * more complicated because there are more cases
1416 * than just upper and lower: one needs to use
1417 * the so-called folding case for case-insensitive
1418 * matching (called "loose matching" in Unicode).
1419 * foldEQ_utf8() will do just that. */
1421 if (do_utf8 || UTF) {
1423 U8 tmpbuf [UTF8_MAXBYTES+1];
1426 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1428 /* Upper and lower of 1st char are equal -
1429 * probably not a "letter". */
1432 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1437 REXEC_FBC_EXACTISH_CHECK(c == c1);
1443 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1449 /* Handle some of the three Greek sigmas cases.
1450 * Note that not all the possible combinations
1451 * are handled here: some of them are handled
1452 * by the standard folding rules, and some of
1453 * them (the character class or ANYOF cases)
1454 * are handled during compiletime in
1455 * regexec.c:S_regclass(). */
1456 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1457 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1458 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1460 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1465 /* Neither pattern nor string are UTF8 */
1467 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1469 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1473 PL_reg_flags |= RF_tainted;
1480 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1481 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1483 tmp = ((OP(c) == BOUND ?
1484 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1485 LOAD_UTF8_CHARCLASS_ALNUM();
1486 REXEC_FBC_UTF8_SCAN(
1487 if (tmp == !(OP(c) == BOUND ?
1488 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) :
1489 isALNUM_LC_utf8((U8*)s)))
1497 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1498 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1501 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1507 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1511 PL_reg_flags |= RF_tainted;
1518 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1519 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1521 tmp = ((OP(c) == NBOUND ?
1522 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1523 LOAD_UTF8_CHARCLASS_ALNUM();
1524 REXEC_FBC_UTF8_SCAN(
1525 if (tmp == !(OP(c) == NBOUND ?
1526 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) :
1527 isALNUM_LC_utf8((U8*)s)))
1529 else REXEC_FBC_TRYIT;
1533 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1534 tmp = ((OP(c) == NBOUND ?
1535 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1538 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1540 else REXEC_FBC_TRYIT;
1543 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1547 REXEC_FBC_CSCAN_PRELOAD(
1548 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1549 swash_fetch(RE_utf8_perl_word, (U8*)s, do_utf8),
1553 REXEC_FBC_CSCAN_TAINT(
1554 isALNUM_LC_utf8((U8*)s),
1558 REXEC_FBC_CSCAN_PRELOAD(
1559 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1560 !swash_fetch(RE_utf8_perl_word, (U8*)s, do_utf8),
1564 REXEC_FBC_CSCAN_TAINT(
1565 !isALNUM_LC_utf8((U8*)s),
1569 REXEC_FBC_CSCAN_PRELOAD(
1570 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1571 *s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, do_utf8),
1575 REXEC_FBC_CSCAN_TAINT(
1576 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1580 REXEC_FBC_CSCAN_PRELOAD(
1581 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1582 !(*s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, do_utf8)),
1586 REXEC_FBC_CSCAN_TAINT(
1587 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1591 REXEC_FBC_CSCAN_PRELOAD(
1592 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1593 swash_fetch(RE_utf8_posix_digit,(U8*)s, do_utf8),
1597 REXEC_FBC_CSCAN_TAINT(
1598 isDIGIT_LC_utf8((U8*)s),
1602 REXEC_FBC_CSCAN_PRELOAD(
1603 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1604 !swash_fetch(RE_utf8_posix_digit,(U8*)s, do_utf8),
1608 REXEC_FBC_CSCAN_TAINT(
1609 !isDIGIT_LC_utf8((U8*)s),
1615 is_LNBREAK_latin1(s)
1625 !is_VERTWS_latin1(s)
1630 is_HORIZWS_latin1(s)
1634 !is_HORIZWS_utf8(s),
1635 !is_HORIZWS_latin1(s)
1641 /* what trie are we using right now */
1643 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1645 = (reg_trie_data*)progi->data->data[ aho->trie ];
1646 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1648 const char *last_start = strend - trie->minlen;
1650 const char *real_start = s;
1652 STRLEN maxlen = trie->maxlen;
1654 U8 **points; /* map of where we were in the input string
1655 when reading a given char. For ASCII this
1656 is unnecessary overhead as the relationship
1657 is always 1:1, but for Unicode, especially
1658 case folded Unicode this is not true. */
1659 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1663 GET_RE_DEBUG_FLAGS_DECL;
1665 /* We can't just allocate points here. We need to wrap it in
1666 * an SV so it gets freed properly if there is a croak while
1667 * running the match */
1670 sv_points=newSV(maxlen * sizeof(U8 *));
1671 SvCUR_set(sv_points,
1672 maxlen * sizeof(U8 *));
1673 SvPOK_on(sv_points);
1674 sv_2mortal(sv_points);
1675 points=(U8**)SvPV_nolen(sv_points );
1676 if ( trie_type != trie_utf8_fold
1677 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1680 bitmap=(U8*)trie->bitmap;
1682 bitmap=(U8*)ANYOF_BITMAP(c);
1684 /* this is the Aho-Corasick algorithm modified a touch
1685 to include special handling for long "unknown char"
1686 sequences. The basic idea being that we use AC as long
1687 as we are dealing with a possible matching char, when
1688 we encounter an unknown char (and we have not encountered
1689 an accepting state) we scan forward until we find a legal
1691 AC matching is basically that of trie matching, except
1692 that when we encounter a failing transition, we fall back
1693 to the current states "fail state", and try the current char
1694 again, a process we repeat until we reach the root state,
1695 state 1, or a legal transition. If we fail on the root state
1696 then we can either terminate if we have reached an accepting
1697 state previously, or restart the entire process from the beginning
1701 while (s <= last_start) {
1702 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1710 U8 *uscan = (U8*)NULL;
1711 U8 *leftmost = NULL;
1713 U32 accepted_word= 0;
1717 while ( state && uc <= (U8*)strend ) {
1719 U32 word = aho->states[ state ].wordnum;
1723 DEBUG_TRIE_EXECUTE_r(
1724 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1725 dump_exec_pos( (char *)uc, c, strend, real_start,
1726 (char *)uc, do_utf8 );
1727 PerlIO_printf( Perl_debug_log,
1728 " Scanning for legal start char...\n");
1731 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1736 if (uc >(U8*)last_start) break;
1740 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1741 if (!leftmost || lpos < leftmost) {
1742 DEBUG_r(accepted_word=word);
1748 points[pointpos++ % maxlen]= uc;
1749 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1750 uscan, len, uvc, charid, foldlen,
1752 DEBUG_TRIE_EXECUTE_r({
1753 dump_exec_pos( (char *)uc, c, strend, real_start,
1755 PerlIO_printf(Perl_debug_log,
1756 " Charid:%3u CP:%4"UVxf" ",
1762 word = aho->states[ state ].wordnum;
1764 base = aho->states[ state ].trans.base;
1766 DEBUG_TRIE_EXECUTE_r({
1768 dump_exec_pos( (char *)uc, c, strend, real_start,
1770 PerlIO_printf( Perl_debug_log,
1771 "%sState: %4"UVxf", word=%"UVxf,
1772 failed ? " Fail transition to " : "",
1773 (UV)state, (UV)word);
1778 (base + charid > trie->uniquecharcount )
1779 && (base + charid - 1 - trie->uniquecharcount
1781 && trie->trans[base + charid - 1 -
1782 trie->uniquecharcount].check == state
1783 && (tmp=trie->trans[base + charid - 1 -
1784 trie->uniquecharcount ].next))
1786 DEBUG_TRIE_EXECUTE_r(
1787 PerlIO_printf( Perl_debug_log," - legal\n"));
1792 DEBUG_TRIE_EXECUTE_r(
1793 PerlIO_printf( Perl_debug_log," - fail\n"));
1795 state = aho->fail[state];
1799 /* we must be accepting here */
1800 DEBUG_TRIE_EXECUTE_r(
1801 PerlIO_printf( Perl_debug_log," - accepting\n"));
1810 if (!state) state = 1;
1813 if ( aho->states[ state ].wordnum ) {
1814 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1815 if (!leftmost || lpos < leftmost) {
1816 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1821 s = (char*)leftmost;
1822 DEBUG_TRIE_EXECUTE_r({
1824 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1825 (UV)accepted_word, (IV)(s - real_start)
1828 if (!reginfo || regtry(reginfo, &s)) {
1834 DEBUG_TRIE_EXECUTE_r({
1835 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1838 DEBUG_TRIE_EXECUTE_r(
1839 PerlIO_printf( Perl_debug_log,"No match.\n"));
1848 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1858 - regexec_flags - match a regexp against a string
1861 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1862 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1863 /* strend: pointer to null at end of string */
1864 /* strbeg: real beginning of string */
1865 /* minend: end of match must be >=minend after stringarg. */
1866 /* data: May be used for some additional optimizations.
1867 Currently its only used, with a U32 cast, for transmitting
1868 the ganch offset when doing a /g match. This will change */
1869 /* nosave: For optimizations. */
1872 struct regexp *const prog = (struct regexp *)SvANY(rx);
1873 /*register*/ char *s;
1874 register regnode *c;
1875 /*register*/ char *startpos = stringarg;
1876 I32 minlen; /* must match at least this many chars */
1877 I32 dontbother = 0; /* how many characters not to try at end */
1878 I32 end_shift = 0; /* Same for the end. */ /* CC */
1879 I32 scream_pos = -1; /* Internal iterator of scream. */
1880 char *scream_olds = NULL;
1881 const bool do_utf8 = cBOOL(DO_UTF8(sv));
1883 RXi_GET_DECL(prog,progi);
1884 regmatch_info reginfo; /* create some info to pass to regtry etc */
1885 regexp_paren_pair *swap = NULL;
1886 GET_RE_DEBUG_FLAGS_DECL;
1888 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1889 PERL_UNUSED_ARG(data);
1891 /* Be paranoid... */
1892 if (prog == NULL || startpos == NULL) {
1893 Perl_croak(aTHX_ "NULL regexp parameter");
1897 multiline = prog->extflags & RXf_PMf_MULTILINE;
1898 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
1900 RX_MATCH_UTF8_set(rx, do_utf8);
1902 debug_start_match(rx, do_utf8, startpos, strend,
1906 minlen = prog->minlen;
1908 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1909 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1910 "String too short [regexec_flags]...\n"));
1915 /* Check validity of program. */
1916 if (UCHARAT(progi->program) != REG_MAGIC) {
1917 Perl_croak(aTHX_ "corrupted regexp program");
1921 PL_reg_eval_set = 0;
1925 PL_reg_flags |= RF_utf8;
1927 /* Mark beginning of line for ^ and lookbehind. */
1928 reginfo.bol = startpos; /* XXX not used ??? */
1932 /* Mark end of line for $ (and such) */
1935 /* see how far we have to get to not match where we matched before */
1936 reginfo.till = startpos+minend;
1938 /* If there is a "must appear" string, look for it. */
1941 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1943 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
1944 reginfo.ganch = startpos + prog->gofs;
1945 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1946 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
1947 } else if (sv && SvTYPE(sv) >= SVt_PVMG
1949 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1950 && mg->mg_len >= 0) {
1951 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1952 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1953 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
1955 if (prog->extflags & RXf_ANCH_GPOS) {
1956 if (s > reginfo.ganch)
1958 s = reginfo.ganch - prog->gofs;
1959 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1960 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
1966 reginfo.ganch = strbeg + PTR2UV(data);
1967 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1968 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
1970 } else { /* pos() not defined */
1971 reginfo.ganch = strbeg;
1972 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1973 "GPOS: reginfo.ganch = strbeg\n"));
1976 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
1977 /* We have to be careful. If the previous successful match
1978 was from this regex we don't want a subsequent partially
1979 successful match to clobber the old results.
1980 So when we detect this possibility we add a swap buffer
1981 to the re, and switch the buffer each match. If we fail
1982 we switch it back, otherwise we leave it swapped.
1985 /* do we need a save destructor here for eval dies? */
1986 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
1988 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1989 re_scream_pos_data d;
1991 d.scream_olds = &scream_olds;
1992 d.scream_pos = &scream_pos;
1993 s = re_intuit_start(rx, sv, s, strend, flags, &d);
1995 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1996 goto phooey; /* not present */
2002 /* Simplest case: anchored match need be tried only once. */
2003 /* [unless only anchor is BOL and multiline is set] */
2004 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2005 if (s == startpos && regtry(®info, &startpos))
2007 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2008 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2013 dontbother = minlen - 1;
2014 end = HOP3c(strend, -dontbother, strbeg) - 1;
2015 /* for multiline we only have to try after newlines */
2016 if (prog->check_substr || prog->check_utf8) {
2020 if (regtry(®info, &s))
2025 if (prog->extflags & RXf_USE_INTUIT) {
2026 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2037 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2038 if (regtry(®info, &s))
2045 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2047 /* the warning about reginfo.ganch being used without intialization
2048 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2049 and we only enter this block when the same bit is set. */
2050 char *tmp_s = reginfo.ganch - prog->gofs;
2052 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2057 /* Messy cases: unanchored match. */
2058 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2059 /* we have /x+whatever/ */
2060 /* it must be a one character string (XXXX Except UTF?) */
2065 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
2066 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2067 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
2072 DEBUG_EXECUTE_r( did_match = 1 );
2073 if (regtry(®info, &s)) goto got_it;
2075 while (s < strend && *s == ch)
2083 DEBUG_EXECUTE_r( did_match = 1 );
2084 if (regtry(®info, &s)) goto got_it;
2086 while (s < strend && *s == ch)
2091 DEBUG_EXECUTE_r(if (!did_match)
2092 PerlIO_printf(Perl_debug_log,
2093 "Did not find anchored character...\n")
2096 else if (prog->anchored_substr != NULL
2097 || prog->anchored_utf8 != NULL
2098 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2099 && prog->float_max_offset < strend - s)) {
2104 char *last1; /* Last position checked before */
2108 if (prog->anchored_substr || prog->anchored_utf8) {
2109 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
2110 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2111 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
2112 back_max = back_min = prog->anchored_offset;
2114 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
2115 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2116 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
2117 back_max = prog->float_max_offset;
2118 back_min = prog->float_min_offset;
2122 if (must == &PL_sv_undef)
2123 /* could not downgrade utf8 check substring, so must fail */
2129 last = HOP3c(strend, /* Cannot start after this */
2130 -(I32)(CHR_SVLEN(must)
2131 - (SvTAIL(must) != 0) + back_min), strbeg);
2134 last1 = HOPc(s, -1);
2136 last1 = s - 1; /* bogus */
2138 /* XXXX check_substr already used to find "s", can optimize if
2139 check_substr==must. */
2141 dontbother = end_shift;
2142 strend = HOPc(strend, -dontbother);
2143 while ( (s <= last) &&
2144 ((flags & REXEC_SCREAM)
2145 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2146 end_shift, &scream_pos, 0))
2147 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2148 (unsigned char*)strend, must,
2149 multiline ? FBMrf_MULTILINE : 0))) ) {
2150 /* we may be pointing at the wrong string */
2151 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2152 s = strbeg + (s - SvPVX_const(sv));
2153 DEBUG_EXECUTE_r( did_match = 1 );
2154 if (HOPc(s, -back_max) > last1) {
2155 last1 = HOPc(s, -back_min);
2156 s = HOPc(s, -back_max);
2159 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2161 last1 = HOPc(s, -back_min);
2165 while (s <= last1) {
2166 if (regtry(®info, &s))
2172 while (s <= last1) {
2173 if (regtry(®info, &s))
2179 DEBUG_EXECUTE_r(if (!did_match) {
2180 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
2181 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2182 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2183 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2184 ? "anchored" : "floating"),
2185 quoted, RE_SV_TAIL(must));
2189 else if ( (c = progi->regstclass) ) {
2191 const OPCODE op = OP(progi->regstclass);
2192 /* don't bother with what can't match */
2193 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2194 strend = HOPc(strend, -(minlen - 1));
2197 SV * const prop = sv_newmortal();
2198 regprop(prog, prop, c);
2200 RE_PV_QUOTED_DECL(quoted,do_utf8,PERL_DEBUG_PAD_ZERO(1),
2202 PerlIO_printf(Perl_debug_log,
2203 "Matching stclass %.*s against %s (%d bytes)\n",
2204 (int)SvCUR(prop), SvPVX_const(prop),
2205 quoted, (int)(strend - s));
2208 if (find_byclass(prog, c, s, strend, ®info))
2210 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2214 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2219 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
2220 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2221 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
2223 if (flags & REXEC_SCREAM) {
2224 last = screaminstr(sv, float_real, s - strbeg,
2225 end_shift, &scream_pos, 1); /* last one */
2227 last = scream_olds; /* Only one occurrence. */
2228 /* we may be pointing at the wrong string */
2229 else if (RXp_MATCH_COPIED(prog))
2230 s = strbeg + (s - SvPVX_const(sv));
2234 const char * const little = SvPV_const(float_real, len);
2236 if (SvTAIL(float_real)) {
2237 if (memEQ(strend - len + 1, little, len - 1))
2238 last = strend - len + 1;
2239 else if (!multiline)
2240 last = memEQ(strend - len, little, len)
2241 ? strend - len : NULL;
2247 last = rninstr(s, strend, little, little + len);
2249 last = strend; /* matching "$" */
2254 PerlIO_printf(Perl_debug_log,
2255 "%sCan't trim the tail, match fails (should not happen)%s\n",
2256 PL_colors[4], PL_colors[5]));
2257 goto phooey; /* Should not happen! */
2259 dontbother = strend - last + prog->float_min_offset;
2261 if (minlen && (dontbother < minlen))
2262 dontbother = minlen - 1;
2263 strend -= dontbother; /* this one's always in bytes! */
2264 /* We don't know much -- general case. */
2267 if (regtry(®info, &s))
2276 if (regtry(®info, &s))
2278 } while (s++ < strend);
2287 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2289 if (PL_reg_eval_set)
2290 restore_pos(aTHX_ prog);
2291 if (RXp_PAREN_NAMES(prog))
2292 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2294 /* make sure $`, $&, $', and $digit will work later */
2295 if ( !(flags & REXEC_NOT_FIRST) ) {
2296 RX_MATCH_COPY_FREE(rx);
2297 if (flags & REXEC_COPY_STR) {
2298 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2299 #ifdef PERL_OLD_COPY_ON_WRITE
2301 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2303 PerlIO_printf(Perl_debug_log,
2304 "Copy on write: regexp capture, type %d\n",
2307 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2308 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2309 assert (SvPOKp(prog->saved_copy));
2313 RX_MATCH_COPIED_on(rx);
2314 s = savepvn(strbeg, i);
2320 prog->subbeg = strbeg;
2321 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2328 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2329 PL_colors[4], PL_colors[5]));
2330 if (PL_reg_eval_set)
2331 restore_pos(aTHX_ prog);
2333 /* we failed :-( roll it back */
2334 Safefree(prog->offs);
2343 - regtry - try match at specific point
2345 STATIC I32 /* 0 failure, 1 success */
2346 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2350 REGEXP *const rx = reginfo->prog;
2351 regexp *const prog = (struct regexp *)SvANY(rx);
2352 RXi_GET_DECL(prog,progi);
2353 GET_RE_DEBUG_FLAGS_DECL;
2355 PERL_ARGS_ASSERT_REGTRY;
2357 reginfo->cutpoint=NULL;
2359 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2362 PL_reg_eval_set = RS_init;
2363 DEBUG_EXECUTE_r(DEBUG_s(
2364 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2365 (IV)(PL_stack_sp - PL_stack_base));
2368 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2369 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2371 /* Apparently this is not needed, judging by wantarray. */
2372 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2373 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2376 /* Make $_ available to executed code. */
2377 if (reginfo->sv != DEFSV) {
2379 DEFSV_set(reginfo->sv);
2382 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2383 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2384 /* prepare for quick setting of pos */
2385 #ifdef PERL_OLD_COPY_ON_WRITE
2386 if (SvIsCOW(reginfo->sv))
2387 sv_force_normal_flags(reginfo->sv, 0);
2389 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2390 &PL_vtbl_mglob, NULL, 0);
2394 PL_reg_oldpos = mg->mg_len;
2395 SAVEDESTRUCTOR_X(restore_pos, prog);
2397 if (!PL_reg_curpm) {
2398 Newxz(PL_reg_curpm, 1, PMOP);
2401 SV* const repointer = &PL_sv_undef;
2402 /* this regexp is also owned by the new PL_reg_curpm, which
2403 will try to free it. */
2404 av_push(PL_regex_padav, repointer);
2405 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2406 PL_regex_pad = AvARRAY(PL_regex_padav);
2411 /* It seems that non-ithreads works both with and without this code.
2412 So for efficiency reasons it seems best not to have the code
2413 compiled when it is not needed. */
2414 /* This is safe against NULLs: */
2415 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2416 /* PM_reg_curpm owns a reference to this regexp. */
2419 PM_SETRE(PL_reg_curpm, rx);
2420 PL_reg_oldcurpm = PL_curpm;
2421 PL_curpm = PL_reg_curpm;
2422 if (RXp_MATCH_COPIED(prog)) {
2423 /* Here is a serious problem: we cannot rewrite subbeg,
2424 since it may be needed if this match fails. Thus
2425 $` inside (?{}) could fail... */
2426 PL_reg_oldsaved = prog->subbeg;
2427 PL_reg_oldsavedlen = prog->sublen;
2428 #ifdef PERL_OLD_COPY_ON_WRITE
2429 PL_nrs = prog->saved_copy;
2431 RXp_MATCH_COPIED_off(prog);
2434 PL_reg_oldsaved = NULL;
2435 prog->subbeg = PL_bostr;
2436 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2438 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2439 prog->offs[0].start = *startpos - PL_bostr;
2440 PL_reginput = *startpos;
2441 PL_reglastparen = &prog->lastparen;
2442 PL_reglastcloseparen = &prog->lastcloseparen;
2443 prog->lastparen = 0;
2444 prog->lastcloseparen = 0;
2446 PL_regoffs = prog->offs;
2447 if (PL_reg_start_tmpl <= prog->nparens) {
2448 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2449 if(PL_reg_start_tmp)
2450 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2452 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2455 /* XXXX What this code is doing here?!!! There should be no need
2456 to do this again and again, PL_reglastparen should take care of
2459 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2460 * Actually, the code in regcppop() (which Ilya may be meaning by
2461 * PL_reglastparen), is not needed at all by the test suite
2462 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2463 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2464 * Meanwhile, this code *is* needed for the
2465 * above-mentioned test suite tests to succeed. The common theme
2466 * on those tests seems to be returning null fields from matches.
2467 * --jhi updated by dapm */
2469 if (prog->nparens) {
2470 regexp_paren_pair *pp = PL_regoffs;
2472 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2480 if (regmatch(reginfo, progi->program + 1)) {
2481 PL_regoffs[0].end = PL_reginput - PL_bostr;
2484 if (reginfo->cutpoint)
2485 *startpos= reginfo->cutpoint;
2486 REGCP_UNWIND(lastcp);
2491 #define sayYES goto yes
2492 #define sayNO goto no
2493 #define sayNO_SILENT goto no_silent
2495 /* we dont use STMT_START/END here because it leads to
2496 "unreachable code" warnings, which are bogus, but distracting. */
2497 #define CACHEsayNO \
2498 if (ST.cache_mask) \
2499 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2502 /* this is used to determine how far from the left messages like
2503 'failed...' are printed. It should be set such that messages
2504 are inline with the regop output that created them.
2506 #define REPORT_CODE_OFF 32
2509 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2510 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2512 #define SLAB_FIRST(s) (&(s)->states[0])
2513 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2515 /* grab a new slab and return the first slot in it */
2517 STATIC regmatch_state *
2520 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2523 regmatch_slab *s = PL_regmatch_slab->next;
2525 Newx(s, 1, regmatch_slab);
2526 s->prev = PL_regmatch_slab;
2528 PL_regmatch_slab->next = s;
2530 PL_regmatch_slab = s;
2531 return SLAB_FIRST(s);
2535 /* push a new state then goto it */
2537 #define PUSH_STATE_GOTO(state, node) \
2539 st->resume_state = state; \
2542 /* push a new state with success backtracking, then goto it */
2544 #define PUSH_YES_STATE_GOTO(state, node) \
2546 st->resume_state = state; \
2547 goto push_yes_state;
2553 regmatch() - main matching routine
2555 This is basically one big switch statement in a loop. We execute an op,
2556 set 'next' to point the next op, and continue. If we come to a point which
2557 we may need to backtrack to on failure such as (A|B|C), we push a
2558 backtrack state onto the backtrack stack. On failure, we pop the top
2559 state, and re-enter the loop at the state indicated. If there are no more
2560 states to pop, we return failure.
2562 Sometimes we also need to backtrack on success; for example /A+/, where
2563 after successfully matching one A, we need to go back and try to
2564 match another one; similarly for lookahead assertions: if the assertion
2565 completes successfully, we backtrack to the state just before the assertion
2566 and then carry on. In these cases, the pushed state is marked as
2567 'backtrack on success too'. This marking is in fact done by a chain of
2568 pointers, each pointing to the previous 'yes' state. On success, we pop to
2569 the nearest yes state, discarding any intermediate failure-only states.
2570 Sometimes a yes state is pushed just to force some cleanup code to be
2571 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2572 it to free the inner regex.
2574 Note that failure backtracking rewinds the cursor position, while
2575 success backtracking leaves it alone.
2577 A pattern is complete when the END op is executed, while a subpattern
2578 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2579 ops trigger the "pop to last yes state if any, otherwise return true"
2582 A common convention in this function is to use A and B to refer to the two
2583 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2584 the subpattern to be matched possibly multiple times, while B is the entire
2585 rest of the pattern. Variable and state names reflect this convention.
2587 The states in the main switch are the union of ops and failure/success of
2588 substates associated with with that op. For example, IFMATCH is the op
2589 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2590 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2591 successfully matched A and IFMATCH_A_fail is a state saying that we have
2592 just failed to match A. Resume states always come in pairs. The backtrack
2593 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2594 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2595 on success or failure.
2597 The struct that holds a backtracking state is actually a big union, with
2598 one variant for each major type of op. The variable st points to the
2599 top-most backtrack struct. To make the code clearer, within each
2600 block of code we #define ST to alias the relevant union.
2602 Here's a concrete example of a (vastly oversimplified) IFMATCH
2608 #define ST st->u.ifmatch
2610 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2611 ST.foo = ...; // some state we wish to save
2613 // push a yes backtrack state with a resume value of
2614 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2616 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2619 case IFMATCH_A: // we have successfully executed A; now continue with B
2621 bar = ST.foo; // do something with the preserved value
2624 case IFMATCH_A_fail: // A failed, so the assertion failed
2625 ...; // do some housekeeping, then ...
2626 sayNO; // propagate the failure
2633 For any old-timers reading this who are familiar with the old recursive
2634 approach, the code above is equivalent to:
2636 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2645 ...; // do some housekeeping, then ...
2646 sayNO; // propagate the failure
2649 The topmost backtrack state, pointed to by st, is usually free. If you
2650 want to claim it, populate any ST.foo fields in it with values you wish to
2651 save, then do one of
2653 PUSH_STATE_GOTO(resume_state, node);
2654 PUSH_YES_STATE_GOTO(resume_state, node);
2656 which sets that backtrack state's resume value to 'resume_state', pushes a
2657 new free entry to the top of the backtrack stack, then goes to 'node'.
2658 On backtracking, the free slot is popped, and the saved state becomes the
2659 new free state. An ST.foo field in this new top state can be temporarily
2660 accessed to retrieve values, but once the main loop is re-entered, it
2661 becomes available for reuse.
2663 Note that the depth of the backtrack stack constantly increases during the
2664 left-to-right execution of the pattern, rather than going up and down with
2665 the pattern nesting. For example the stack is at its maximum at Z at the
2666 end of the pattern, rather than at X in the following:
2668 /(((X)+)+)+....(Y)+....Z/
2670 The only exceptions to this are lookahead/behind assertions and the cut,
2671 (?>A), which pop all the backtrack states associated with A before
2674 Bascktrack state structs are allocated in slabs of about 4K in size.
2675 PL_regmatch_state and st always point to the currently active state,
2676 and PL_regmatch_slab points to the slab currently containing
2677 PL_regmatch_state. The first time regmatch() is called, the first slab is
2678 allocated, and is never freed until interpreter destruction. When the slab
2679 is full, a new one is allocated and chained to the end. At exit from
2680 regmatch(), slabs allocated since entry are freed.
2685 #define DEBUG_STATE_pp(pp) \
2687 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2688 PerlIO_printf(Perl_debug_log, \
2689 " %*s"pp" %s%s%s%s%s\n", \
2691 PL_reg_name[st->resume_state], \
2692 ((st==yes_state||st==mark_state) ? "[" : ""), \
2693 ((st==yes_state) ? "Y" : ""), \
2694 ((st==mark_state) ? "M" : ""), \
2695 ((st==yes_state||st==mark_state) ? "]" : "") \
2700 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2705 S_debug_start_match(pTHX_ const REGEXP *prog, const bool do_utf8,
2706 const char *start, const char *end, const char *blurb)
2708 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2710 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2715 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2716 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2718 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2719 start, end - start, 60);
2721 PerlIO_printf(Perl_debug_log,
2722 "%s%s REx%s %s against %s\n",
2723 PL_colors[4], blurb, PL_colors[5], s0, s1);
2725 if (do_utf8||utf8_pat)
2726 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2727 utf8_pat ? "pattern" : "",
2728 utf8_pat && do_utf8 ? " and " : "",
2729 do_utf8 ? "string" : ""
2735 S_dump_exec_pos(pTHX_ const char *locinput,
2736 const regnode *scan,
2737 const char *loc_regeol,
2738 const char *loc_bostr,
2739 const char *loc_reg_starttry,
2742 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2743 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2744 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2745 /* The part of the string before starttry has one color
2746 (pref0_len chars), between starttry and current
2747 position another one (pref_len - pref0_len chars),
2748 after the current position the third one.
2749 We assume that pref0_len <= pref_len, otherwise we
2750 decrease pref0_len. */
2751 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2752 ? (5 + taill) - l : locinput - loc_bostr;
2755 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2757 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2759 pref0_len = pref_len - (locinput - loc_reg_starttry);
2760 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2761 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2762 ? (5 + taill) - pref_len : loc_regeol - locinput);
2763 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2767 if (pref0_len > pref_len)
2768 pref0_len = pref_len;
2770 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2772 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2773 (locinput - pref_len),pref0_len, 60, 4, 5);
2775 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2776 (locinput - pref_len + pref0_len),
2777 pref_len - pref0_len, 60, 2, 3);
2779 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2780 locinput, loc_regeol - locinput, 10, 0, 1);
2782 const STRLEN tlen=len0+len1+len2;
2783 PerlIO_printf(Perl_debug_log,
2784 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2785 (IV)(locinput - loc_bostr),
2788 (docolor ? "" : "> <"),
2790 (int)(tlen > 19 ? 0 : 19 - tlen),
2797 /* reg_check_named_buff_matched()
2798 * Checks to see if a named buffer has matched. The data array of
2799 * buffer numbers corresponding to the buffer is expected to reside
2800 * in the regexp->data->data array in the slot stored in the ARG() of
2801 * node involved. Note that this routine doesn't actually care about the
2802 * name, that information is not preserved from compilation to execution.
2803 * Returns the index of the leftmost defined buffer with the given name
2804 * or 0 if non of the buffers matched.
2807 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2810 RXi_GET_DECL(rex,rexi);
2811 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2812 I32 *nums=(I32*)SvPVX(sv_dat);
2814 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2816 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2817 if ((I32)*PL_reglastparen >= nums[n] &&
2818 PL_regoffs[nums[n]].end != -1)
2827 /* free all slabs above current one - called during LEAVE_SCOPE */
2830 S_clear_backtrack_stack(pTHX_ void *p)
2832 regmatch_slab *s = PL_regmatch_slab->next;
2837 PL_regmatch_slab->next = NULL;
2839 regmatch_slab * const osl = s;
2846 #define SETREX(Re1,Re2) \
2847 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2850 STATIC I32 /* 0 failure, 1 success */
2851 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2853 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2857 register const bool do_utf8 = PL_reg_match_utf8;
2858 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2859 REGEXP *rex_sv = reginfo->prog;
2860 regexp *rex = (struct regexp *)SvANY(rex_sv);
2861 RXi_GET_DECL(rex,rexi);
2863 /* the current state. This is a cached copy of PL_regmatch_state */
2864 register regmatch_state *st;
2865 /* cache heavy used fields of st in registers */
2866 register regnode *scan;
2867 register regnode *next;
2868 register U32 n = 0; /* general value; init to avoid compiler warning */
2869 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2870 register char *locinput = PL_reginput;
2871 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2873 bool result = 0; /* return value of S_regmatch */
2874 int depth = 0; /* depth of backtrack stack */
2875 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2876 const U32 max_nochange_depth =
2877 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2878 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2879 regmatch_state *yes_state = NULL; /* state to pop to on success of
2881 /* mark_state piggy backs on the yes_state logic so that when we unwind
2882 the stack on success we can update the mark_state as we go */
2883 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2884 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2885 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2887 bool no_final = 0; /* prevent failure from backtracking? */
2888 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2889 char *startpoint = PL_reginput;
2890 SV *popmark = NULL; /* are we looking for a mark? */
2891 SV *sv_commit = NULL; /* last mark name seen in failure */
2892 SV *sv_yes_mark = NULL; /* last mark name we have seen
2893 during a successfull match */
2894 U32 lastopen = 0; /* last open we saw */
2895 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2896 SV* const oreplsv = GvSV(PL_replgv);
2897 /* these three flags are set by various ops to signal information to
2898 * the very next op. They have a useful lifetime of exactly one loop
2899 * iteration, and are not preserved or restored by state pushes/pops
2901 bool sw = 0; /* the condition value in (?(cond)a|b) */
2902 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2903 int logical = 0; /* the following EVAL is:
2907 or the following IFMATCH/UNLESSM is:
2908 false: plain (?=foo)
2909 true: used as a condition: (?(?=foo))
2912 GET_RE_DEBUG_FLAGS_DECL;
2915 PERL_ARGS_ASSERT_REGMATCH;
2917 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
2918 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2920 /* on first ever call to regmatch, allocate first slab */
2921 if (!PL_regmatch_slab) {
2922 Newx(PL_regmatch_slab, 1, regmatch_slab);
2923 PL_regmatch_slab->prev = NULL;
2924 PL_regmatch_slab->next = NULL;
2925 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2928 oldsave = PL_savestack_ix;
2929 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
2930 SAVEVPTR(PL_regmatch_slab);
2931 SAVEVPTR(PL_regmatch_state);
2933 /* grab next free state slot */
2934 st = ++PL_regmatch_state;
2935 if (st > SLAB_LAST(PL_regmatch_slab))
2936 st = PL_regmatch_state = S_push_slab(aTHX);
2938 /* Note that nextchr is a byte even in UTF */
2939 nextchr = UCHARAT(locinput);
2941 while (scan != NULL) {
2944 SV * const prop = sv_newmortal();
2945 regnode *rnext=regnext(scan);
2946 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2947 regprop(rex, prop, scan);
2949 PerlIO_printf(Perl_debug_log,
2950 "%3"IVdf":%*s%s(%"IVdf")\n",
2951 (IV)(scan - rexi->program), depth*2, "",
2953 (PL_regkind[OP(scan)] == END || !rnext) ?
2954 0 : (IV)(rnext - rexi->program));
2957 next = scan + NEXT_OFF(scan);
2960 state_num = OP(scan);
2964 assert(PL_reglastparen == &rex->lastparen);
2965 assert(PL_reglastcloseparen == &rex->lastcloseparen);
2966 assert(PL_regoffs == rex->offs);
2968 switch (state_num) {
2970 if (locinput == PL_bostr)
2972 /* reginfo->till = reginfo->bol; */
2977 if (locinput == PL_bostr ||
2978 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2984 if (locinput == PL_bostr)
2988 if (locinput == reginfo->ganch)
2993 /* update the startpoint */
2994 st->u.keeper.val = PL_regoffs[0].start;
2995 PL_reginput = locinput;
2996 PL_regoffs[0].start = locinput - PL_bostr;
2997 PUSH_STATE_GOTO(KEEPS_next, next);
2999 case KEEPS_next_fail:
3000 /* rollback the start point change */
3001 PL_regoffs[0].start = st->u.keeper.val;
3007 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3012 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3014 if (PL_regeol - locinput > 1)
3018 if (PL_regeol != locinput)
3022 if (!nextchr && locinput >= PL_regeol)
3025 locinput += PL_utf8skip[nextchr];
3026 if (locinput > PL_regeol)
3028 nextchr = UCHARAT(locinput);
3031 nextchr = UCHARAT(++locinput);
3034 if (!nextchr && locinput >= PL_regeol)
3036 nextchr = UCHARAT(++locinput);
3039 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3042 locinput += PL_utf8skip[nextchr];
3043 if (locinput > PL_regeol)
3045 nextchr = UCHARAT(locinput);
3048 nextchr = UCHARAT(++locinput);
3052 #define ST st->u.trie
3054 /* In this case the charclass data is available inline so
3055 we can fail fast without a lot of extra overhead.
3057 if (scan->flags == EXACT || !do_utf8) {
3058 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3060 PerlIO_printf(Perl_debug_log,
3061 "%*s %sfailed to match trie start class...%s\n",
3062 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3070 /* the basic plan of execution of the trie is:
3071 * At the beginning, run though all the states, and
3072 * find the longest-matching word. Also remember the position
3073 * of the shortest matching word. For example, this pattern:
3076 * when matched against the string "abcde", will generate
3077 * accept states for all words except 3, with the longest
3078 * matching word being 4, and the shortest being 1 (with
3079 * the position being after char 1 of the string).
3081 * Then for each matching word, in word order (i.e. 1,2,4,5),
3082 * we run the remainder of the pattern; on each try setting
3083 * the current position to the character following the word,
3084 * returning to try the next word on failure.
3086 * We avoid having to build a list of words at runtime by
3087 * using a compile-time structure, wordinfo[].prev, which
3088 * gives, for each word, the previous accepting word (if any).
3089 * In the case above it would contain the mappings 1->2, 2->0,
3090 * 3->0, 4->5, 5->1. We can use this table to generate, from
3091 * the longest word (4 above), a list of all words, by
3092 * following the list of prev pointers; this gives us the
3093 * unordered list 4,5,1,2. Then given the current word we have
3094 * just tried, we can go through the list and find the
3095 * next-biggest word to try (so if we just failed on word 2,
3096 * the next in the list is 4).
3098 * Since at runtime we don't record the matching position in
3099 * the string for each word, we have to work that out for
3100 * each word we're about to process. The wordinfo table holds
3101 * the character length of each word; given that we recorded
3102 * at the start: the position of the shortest word and its
3103 * length in chars, we just need to move the pointer the
3104 * difference between the two char lengths. Depending on
3105 * Unicode status and folding, that's cheap or expensive.
3107 * This algorithm is optimised for the case where are only a
3108 * small number of accept states, i.e. 0,1, or maybe 2.
3109 * With lots of accepts states, and having to try all of them,
3110 * it becomes quadratic on number of accept states to find all
3115 /* what type of TRIE am I? (utf8 makes this contextual) */
3116 DECL_TRIE_TYPE(scan);
3118 /* what trie are we using right now */
3119 reg_trie_data * const trie
3120 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3121 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3122 U32 state = trie->startstate;
3124 if (trie->bitmap && trie_type != trie_utf8_fold &&
3125 !TRIE_BITMAP_TEST(trie,*locinput)
3127 if (trie->states[ state ].wordnum) {
3129 PerlIO_printf(Perl_debug_log,
3130 "%*s %smatched empty string...%s\n",
3131 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3136 PerlIO_printf(Perl_debug_log,
3137 "%*s %sfailed to match trie start class...%s\n",
3138 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3145 U8 *uc = ( U8* )locinput;
3149 U8 *uscan = (U8*)NULL;
3150 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3151 U32 charcount = 0; /* how many input chars we have matched */
3152 U32 accepted = 0; /* have we seen any accepting states? */
3155 ST.jump = trie->jump;
3158 ST.longfold = FALSE; /* char longer if folded => it's harder */
3161 /* fully traverse the TRIE; note the position of the
3162 shortest accept state and the wordnum of the longest
3165 while ( state && uc <= (U8*)PL_regeol ) {
3166 U32 base = trie->states[ state ].trans.base;
3170 wordnum = trie->states[ state ].wordnum;
3172 if (wordnum) { /* it's an accept state */
3175 /* record first match position */
3177 ST.firstpos = (U8*)locinput;
3182 ST.firstchars = charcount;
3185 if (!ST.nextword || wordnum < ST.nextword)
3186 ST.nextword = wordnum;
3187 ST.topword = wordnum;
3190 DEBUG_TRIE_EXECUTE_r({
3191 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
3192 PerlIO_printf( Perl_debug_log,
3193 "%*s %sState: %4"UVxf" Accepted: %c ",
3194 2+depth * 2, "", PL_colors[4],
3195 (UV)state, (accepted ? 'Y' : 'N'));
3198 /* read a char and goto next state */
3200 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3201 uscan, len, uvc, charid, foldlen,
3207 (base + charid > trie->uniquecharcount )
3208 && (base + charid - 1 - trie->uniquecharcount
3210 && trie->trans[base + charid - 1 -
3211 trie->uniquecharcount].check == state)
3213 state = trie->trans[base + charid - 1 -
3214 trie->uniquecharcount ].next;
3225 DEBUG_TRIE_EXECUTE_r(
3226 PerlIO_printf( Perl_debug_log,
3227 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3228 charid, uvc, (UV)state, PL_colors[5] );
3234 /* calculate total number of accept states */
3239 w = trie->wordinfo[w].prev;
3242 ST.accepted = accepted;
3246 PerlIO_printf( Perl_debug_log,
3247 "%*s %sgot %"IVdf" possible matches%s\n",
3248 REPORT_CODE_OFF + depth * 2, "",
3249 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3251 goto trie_first_try; /* jump into the fail handler */
3255 case TRIE_next_fail: /* we failed - try next alternative */
3257 REGCP_UNWIND(ST.cp);
3258 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3259 PL_regoffs[n].end = -1;
3260 *PL_reglastparen = n;
3262 if (!--ST.accepted) {
3264 PerlIO_printf( Perl_debug_log,
3265 "%*s %sTRIE failed...%s\n",
3266 REPORT_CODE_OFF+depth*2, "",
3273 /* Find next-highest word to process. Note that this code
3274 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3275 register U16 min = 0;
3277 register U16 const nextword = ST.nextword;
3278 register reg_trie_wordinfo * const wordinfo
3279 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3280 for (word=ST.topword; word; word=wordinfo[word].prev) {
3281 if (word > nextword && (!min || word < min))
3294 ST.lastparen = *PL_reglastparen;
3298 /* find start char of end of current word */
3300 U32 chars; /* how many chars to skip */
3301 U8 *uc = ST.firstpos;
3302 reg_trie_data * const trie
3303 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3305 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3307 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3311 /* the hard option - fold each char in turn and find
3312 * its folded length (which may be different */
3313 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3321 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3329 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3334 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3348 PL_reginput = (char *)uc;
3351 scan = (ST.jump && ST.jump[ST.nextword])
3352 ? ST.me + ST.jump[ST.nextword]
3356 PerlIO_printf( Perl_debug_log,
3357 "%*s %sTRIE matched word #%d, continuing%s\n",
3358 REPORT_CODE_OFF+depth*2, "",
3365 if (ST.accepted > 1 || has_cutgroup) {
3366 PUSH_STATE_GOTO(TRIE_next, scan);
3369 /* only one choice left - just continue */
3371 AV *const trie_words
3372 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3373 SV ** const tmp = av_fetch( trie_words,
3375 SV *sv= tmp ? sv_newmortal() : NULL;
3377 PerlIO_printf( Perl_debug_log,
3378 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3379 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3381 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3382 PL_colors[0], PL_colors[1],
3383 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3385 : "not compiled under -Dr",
3389 locinput = PL_reginput;
3390 nextchr = UCHARAT(locinput);
3391 continue; /* execute rest of RE */
3396 char *s = STRING(scan);
3398 if (do_utf8 != UTF) {
3399 /* The target and the pattern have differing utf8ness. */
3401 const char * const e = s + ln;
3404 /* The target is utf8, the pattern is not utf8. */
3409 if (NATIVE_TO_UNI(*(U8*)s) !=
3410 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3418 /* The target is not utf8, the pattern is utf8. */
3423 if (NATIVE_TO_UNI(*((U8*)l)) !=
3424 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3432 nextchr = UCHARAT(locinput);
3435 /* The target and the pattern have the same utf8ness. */
3436 /* Inline the first character, for speed. */
3437 if (UCHARAT(s) != nextchr)
3439 if (PL_regeol - locinput < ln)
3441 if (ln > 1 && memNE(s, locinput, ln))
3444 nextchr = UCHARAT(locinput);
3448 PL_reg_flags |= RF_tainted;
3451 char * const s = STRING(scan);
3454 if (do_utf8 || UTF) {
3455 /* Either target or the pattern are utf8. */
3456 const char * const l = locinput;
3457 char *e = PL_regeol;
3459 if (! foldEQ_utf8(s, 0, ln, cBOOL(UTF),
3460 l, &e, 0, do_utf8)) {
3461 /* One more case for the sharp s:
3462 * pack("U0U*", 0xDF) =~ /ss/i,
3463 * the 0xC3 0x9F are the UTF-8
3464 * byte sequence for the U+00DF. */
3467 toLOWER(s[0]) == 's' &&
3469 toLOWER(s[1]) == 's' &&
3476 nextchr = UCHARAT(locinput);
3480 /* Neither the target and the pattern are utf8. */
3482 /* Inline the first character, for speed. */
3483 if (UCHARAT(s) != nextchr &&
3484 UCHARAT(s) != ((OP(scan) == EXACTF)
3485 ? PL_fold : PL_fold_locale)[nextchr])
3487 if (PL_regeol - locinput < ln)
3489 if (ln > 1 && (OP(scan) == EXACTF
3490 ? ! foldEQ(s, locinput, ln)
3491 : ! foldEQ_locale(s, locinput, ln)))
3494 nextchr = UCHARAT(locinput);
3499 PL_reg_flags |= RF_tainted;
3503 /* was last char in word? */
3505 if (locinput == PL_bostr)
3508 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3510 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3512 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3513 ln = isALNUM_uni(ln);
3514 LOAD_UTF8_CHARCLASS_ALNUM();
3515 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3518 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3519 n = isALNUM_LC_utf8((U8*)locinput);
3523 ln = (locinput != PL_bostr) ?
3524 UCHARAT(locinput - 1) : '\n';
3525 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3527 n = isALNUM(nextchr);
3530 ln = isALNUM_LC(ln);
3531 n = isALNUM_LC(nextchr);
3534 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3535 OP(scan) == BOUNDL))
3540 STRLEN inclasslen = PL_regeol - locinput;
3542 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3544 if (locinput >= PL_regeol)
3546 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3547 nextchr = UCHARAT(locinput);
3552 nextchr = UCHARAT(locinput);
3553 if (!REGINCLASS(rex, scan, (U8*)locinput))
3555 if (!nextchr && locinput >= PL_regeol)
3557 nextchr = UCHARAT(++locinput);
3561 /* If we might have the case of the German sharp s
3562 * in a casefolding Unicode character class. */
3564 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3565 locinput += SHARP_S_SKIP;
3566 nextchr = UCHARAT(locinput);
3571 /* Special char classes - The defines start on line 129 or so */
3572 CCC_TRY_AFF( ALNUM, ALNUML, perl_word, "a", isALNUM_LC_utf8, isALNUM, isALNUM_LC);
3573 CCC_TRY_NEG(NALNUM, NALNUML, perl_word, "a", isALNUM_LC_utf8, isALNUM, isALNUM_LC);
3575 CCC_TRY_AFF( SPACE, SPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE, isSPACE_LC);
3576 CCC_TRY_NEG(NSPACE, NSPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE, isSPACE_LC);
3578 CCC_TRY_AFF( DIGIT, DIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3579 CCC_TRY_NEG(NDIGIT, NDIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3581 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3582 a Unicode extended Grapheme Cluster */
3583 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3584 extended Grapheme Cluster is:
3587 | Prepend* Begin Extend*
3590 Begin is (Hangul-syllable | ! Control)
3591 Extend is (Grapheme_Extend | Spacing_Mark)
3592 Control is [ GCB_Control CR LF ]
3594 The discussion below shows how the code for CLUMP is derived
3595 from this regex. Note that most of these concepts are from
3596 property values of the Grapheme Cluster Boundary (GCB) property.
3597 No code point can have multiple property values for a given
3598 property. Thus a code point in Prepend can't be in Control, but
3599 it must be in !Control. This is why Control above includes
3600 GCB_Control plus CR plus LF. The latter two are used in the GCB
3601 property separately, and so can't be in GCB_Control, even though
3602 they logically are controls. Control is not the same as gc=cc,
3603 but includes format and other characters as well.
3605 The Unicode definition of Hangul-syllable is:
3607 | (L* ( ( V | LV ) V* | LVT ) T*)
3610 Each of these is a value for the GCB property, and hence must be
3611 disjoint, so the order they are tested is immaterial, so the
3612 above can safely be changed to
3615 | (L* ( LVT | ( V | LV ) V*) T*)
3617 The last two terms can be combined like this:
3619 | (( LVT | ( V | LV ) V*) T*))
3621 And refactored into this:
3622 L* (L | LVT T* | V V* T* | LV V* T*)
3624 That means that if we have seen any L's at all we can quit
3625 there, but if the next character is a LVT, a V or and LV we
3628 There is a subtlety with Prepend* which showed up in testing.
3629 Note that the Begin, and only the Begin is required in:
3630 | Prepend* Begin Extend*
3631 Also, Begin contains '! Control'. A Prepend must be a '!
3632 Control', which means it must be a Begin. What it comes down to
3633 is that if we match Prepend* and then find no suitable Begin
3634 afterwards, that if we backtrack the last Prepend, that one will
3635 be a suitable Begin.
3638 if (locinput >= PL_regeol)
3642 /* Match either CR LF or '.', as all the other possibilities
3644 locinput++; /* Match the . or CR */
3646 && locinput < PL_regeol
3647 && UCHARAT(locinput) == '\n') locinput++;
3651 /* Utf8: See if is ( CR LF ); already know that locinput <
3652 * PL_regeol, so locinput+1 is in bounds */
3653 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3657 /* In case have to backtrack to beginning, then match '.' */
3658 char *starting = locinput;
3660 /* In case have to backtrack the last prepend */
3661 char *previous_prepend = 0;
3663 LOAD_UTF8_CHARCLASS_GCB();
3665 /* Match (prepend)* */
3666 while (locinput < PL_regeol
3667 && swash_fetch(PL_utf8_X_prepend,
3668 (U8*)locinput, do_utf8))
3670 previous_prepend = locinput;
3671 locinput += UTF8SKIP(locinput);
3674 /* As noted above, if we matched a prepend character, but
3675 * the next thing won't match, back off the last prepend we
3676 * matched, as it is guaranteed to match the begin */
3677 if (previous_prepend
3678 && (locinput >= PL_regeol
3679 || ! swash_fetch(PL_utf8_X_begin,
3680 (U8*)locinput, do_utf8)))
3682 locinput = previous_prepend;
3685 /* Note that here we know PL_regeol > locinput, as we
3686 * tested that upon input to this switch case, and if we
3687 * moved locinput forward, we tested the result just above
3688 * and it either passed, or we backed off so that it will
3690 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, do_utf8)) {
3692 /* Here did not match the required 'Begin' in the
3693 * second term. So just match the very first
3694 * character, the '.' of the final term of the regex */
3695 locinput = starting + UTF8SKIP(starting);
3698 /* Here is the beginning of a character that can have
3699 * an extender. It is either a hangul syllable, or a
3701 if (swash_fetch(PL_utf8_X_non_hangul,
3702 (U8*)locinput, do_utf8))
3705 /* Here not a Hangul syllable, must be a
3706 * ('! * Control') */
3707 locinput += UTF8SKIP(locinput);
3710 /* Here is a Hangul syllable. It can be composed
3711 * of several individual characters. One
3712 * possibility is T+ */
3713 if (swash_fetch(PL_utf8_X_T,
3714 (U8*)locinput, do_utf8))
3716 while (locinput < PL_regeol
3717 && swash_fetch(PL_utf8_X_T,
3718 (U8*)locinput, do_utf8))
3720 locinput += UTF8SKIP(locinput);
3724 /* Here, not T+, but is a Hangul. That means
3725 * it is one of the others: L, LV, LVT or V,
3727 * L* (L | LVT T* | V V* T* | LV V* T*) */
3730 while (locinput < PL_regeol
3731 && swash_fetch(PL_utf8_X_L,
3732 (U8*)locinput, do_utf8))
3734 locinput += UTF8SKIP(locinput);
3737 /* Here, have exhausted L*. If the next
3738 * character is not an LV, LVT nor V, it means
3739 * we had to have at least one L, so matches L+
3740 * in the original equation, we have a complete
3741 * hangul syllable. Are done. */
3743 if (locinput < PL_regeol
3744 && swash_fetch(PL_utf8_X_LV_LVT_V,
3745 (U8*)locinput, do_utf8))
3748 /* Otherwise keep going. Must be LV, LVT
3749 * or V. See if LVT */
3750 if (swash_fetch(PL_utf8_X_LVT,
3751 (U8*)locinput, do_utf8))
3753 locinput += UTF8SKIP(locinput);
3756 /* Must be V or LV. Take it, then
3758 locinput += UTF8SKIP(locinput);
3759 while (locinput < PL_regeol
3760 && swash_fetch(PL_utf8_X_V,
3761 (U8*)locinput, do_utf8))
3763 locinput += UTF8SKIP(locinput);
3767 /* And any of LV, LVT, or V can be followed
3769 while (locinput < PL_regeol
3770 && swash_fetch(PL_utf8_X_T,
3774 locinput += UTF8SKIP(locinput);
3780 /* Match any extender */
3781 while (locinput < PL_regeol
3782 && swash_fetch(PL_utf8_X_extend,
3783 (U8*)locinput, do_utf8))
3785 locinput += UTF8SKIP(locinput);
3789 if (locinput > PL_regeol) sayNO;
3791 nextchr = UCHARAT(locinput);
3798 PL_reg_flags |= RF_tainted;
3803 n = reg_check_named_buff_matched(rex,scan);
3806 type = REF + ( type - NREF );
3813 PL_reg_flags |= RF_tainted;
3817 n = ARG(scan); /* which paren pair */
3820 ln = PL_regoffs[n].start;
3821 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3822 if (*PL_reglastparen < n || ln == -1)
3823 sayNO; /* Do not match unless seen CLOSEn. */
3824 if (ln == PL_regoffs[n].end)
3828 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3830 const char *e = PL_bostr + PL_regoffs[n].end;