5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #define RF_tainted 1 /* tainted information used? */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
97 /* Valid for non-utf8 strings only: avoids the reginclass call if there are no
98 * complications: i.e., if everything matchable is straight forward in the
100 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
101 : ANYOF_BITMAP_TEST(p,*(c)))
107 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
108 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
110 #define HOPc(pos,off) \
111 (char *)(PL_reg_match_utf8 \
112 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
114 #define HOPBACKc(pos, off) \
115 (char*)(PL_reg_match_utf8\
116 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
117 : (pos - off >= PL_bostr) \
121 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
122 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
124 /* these are unrolled below in the CCC_TRY_XXX defined */
125 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
126 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
128 /* Doesn't do an assert to verify that is correct */
129 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
130 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)" "); LEAVE; } } STMT_END
132 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
133 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
134 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
136 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
137 LOAD_UTF8_CHARCLASS(X_begin, " "); \
138 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
139 /* These are utf8 constants, and not utf-ebcdic constants, so the \
140 * assert should likely and hopefully fail on an EBCDIC machine */ \
141 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
143 /* No asserts are done for these, in case called on an early \
144 * Unicode version in which they map to nothing */ \
145 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
146 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
147 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
148 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
149 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
150 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
151 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
154 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
155 so that it is possible to override the option here without having to
156 rebuild the entire core. as we are required to do if we change regcomp.h
157 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
159 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
160 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
163 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
164 #define LOAD_UTF8_CHARCLASS_PERL_WORD() LOAD_UTF8_CHARCLASS_ALNUM()
165 #define LOAD_UTF8_CHARCLASS_PERL_SPACE() LOAD_UTF8_CHARCLASS_SPACE()
166 #define LOAD_UTF8_CHARCLASS_POSIX_DIGIT() LOAD_UTF8_CHARCLASS_DIGIT()
167 #define RE_utf8_perl_word PL_utf8_alnum
168 #define RE_utf8_perl_space PL_utf8_space
169 #define RE_utf8_posix_digit PL_utf8_digit
170 #define perl_word alnum
171 #define perl_space space
172 #define posix_digit digit
174 #define LOAD_UTF8_CHARCLASS_PERL_WORD() LOAD_UTF8_CHARCLASS(perl_word,"a")
175 #define LOAD_UTF8_CHARCLASS_PERL_SPACE() LOAD_UTF8_CHARCLASS(perl_space," ")
176 #define LOAD_UTF8_CHARCLASS_POSIX_DIGIT() LOAD_UTF8_CHARCLASS(posix_digit,"0")
177 #define RE_utf8_perl_word PL_utf8_perl_word
178 #define RE_utf8_perl_space PL_utf8_perl_space
179 #define RE_utf8_posix_digit PL_utf8_posix_digit
183 #define _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
185 PL_reg_flags |= RF_tainted; \
190 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
191 if (!CAT2(PL_utf8_,CLASS)) { \
195 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
199 if (!(OP(scan) == NAME \
200 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, utf8_target)) \
201 : LCFUNC_utf8((U8*)locinput))) \
205 locinput += PL_utf8skip[nextchr]; \
206 nextchr = UCHARAT(locinput); \
209 /* Drops through to the macro that calls this one */
211 #define CCC_TRY_AFF(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
212 _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
213 if (!(OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
215 nextchr = UCHARAT(++locinput); \
218 /* Almost identical to the above, but has a case for a node that matches chars
219 * between 128 and 255 using Unicode (latin1) semantics. */
220 #define CCC_TRY_AFF_U(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU,LCFUNC) \
221 _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
222 if (!(OP(scan) == NAMEL ? LCFUNC(nextchr) : (FUNCU(nextchr) && (isASCII(nextchr) || (FLAGS(scan) & USE_UNI))))) \
224 nextchr = UCHARAT(++locinput); \
227 #define _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
229 PL_reg_flags |= RF_tainted; \
232 if (!nextchr && locinput >= PL_regeol) \
234 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
235 if (!CAT2(PL_utf8_,CLASS)) { \
239 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
243 if ((OP(scan) == NAME \
244 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, utf8_target)) \
245 : LCFUNC_utf8((U8*)locinput))) \
249 locinput += PL_utf8skip[nextchr]; \
250 nextchr = UCHARAT(locinput); \
254 #define CCC_TRY_NEG(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
255 _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
256 if ((OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
258 nextchr = UCHARAT(++locinput); \
262 #define CCC_TRY_NEG_U(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU,LCFUNC) \
263 _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU) \
264 if ((OP(scan) == NAMEL ? LCFUNC(nextchr) : (FUNCU(nextchr) && (isASCII(nextchr) || (FLAGS(scan) & USE_UNI))))) \
266 nextchr = UCHARAT(++locinput); \
271 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
273 /* for use after a quantifier and before an EXACT-like node -- japhy */
274 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
276 * NOTE that *nothing* that affects backtracking should be in here, specifically
277 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
278 * node that is in between two EXACT like nodes when ascertaining what the required
279 * "follow" character is. This should probably be moved to regex compile time
280 * although it may be done at run time beause of the REF possibility - more
281 * investigation required. -- demerphq
283 #define JUMPABLE(rn) ( \
285 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
287 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
288 OP(rn) == PLUS || OP(rn) == MINMOD || \
290 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
292 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
294 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
297 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
298 we don't need this definition. */
299 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
300 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
301 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
304 /* ... so we use this as its faster. */
305 #define IS_TEXT(rn) ( OP(rn)==EXACT )
306 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU )
307 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
308 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
313 Search for mandatory following text node; for lookahead, the text must
314 follow but for lookbehind (rn->flags != 0) we skip to the next step.
316 #define FIND_NEXT_IMPT(rn) STMT_START { \
317 while (JUMPABLE(rn)) { \
318 const OPCODE type = OP(rn); \
319 if (type == SUSPEND || PL_regkind[type] == CURLY) \
320 rn = NEXTOPER(NEXTOPER(rn)); \
321 else if (type == PLUS) \
323 else if (type == IFMATCH) \
324 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
325 else rn += NEXT_OFF(rn); \
330 static void restore_pos(pTHX_ void *arg);
332 #define REGCP_PAREN_ELEMS 4
333 #define REGCP_OTHER_ELEMS 5
334 #define REGCP_FRAME_ELEMS 1
335 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
336 * are needed for the regexp context stack bookkeeping. */
339 S_regcppush(pTHX_ I32 parenfloor)
342 const int retval = PL_savestack_ix;
343 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
344 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
345 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
347 GET_RE_DEBUG_FLAGS_DECL;
349 if (paren_elems_to_push < 0)
350 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
352 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
353 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
354 " out of range (%lu-%ld)",
355 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
357 SSGROW(total_elems + REGCP_FRAME_ELEMS);
359 for (p = PL_regsize; p > parenfloor; p--) {
360 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
361 SSPUSHINT(PL_regoffs[p].end);
362 SSPUSHINT(PL_regoffs[p].start);
363 SSPUSHPTR(PL_reg_start_tmp[p]);
365 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
366 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
367 (UV)p, (IV)PL_regoffs[p].start,
368 (IV)(PL_reg_start_tmp[p] - PL_bostr),
369 (IV)PL_regoffs[p].end
372 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
373 SSPUSHPTR(PL_regoffs);
374 SSPUSHINT(PL_regsize);
375 SSPUSHINT(*PL_reglastparen);
376 SSPUSHINT(*PL_reglastcloseparen);
377 SSPUSHPTR(PL_reginput);
378 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
383 /* These are needed since we do not localize EVAL nodes: */
384 #define REGCP_SET(cp) \
386 PerlIO_printf(Perl_debug_log, \
387 " Setting an EVAL scope, savestack=%"IVdf"\n", \
388 (IV)PL_savestack_ix)); \
391 #define REGCP_UNWIND(cp) \
393 if (cp != PL_savestack_ix) \
394 PerlIO_printf(Perl_debug_log, \
395 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
396 (IV)(cp), (IV)PL_savestack_ix)); \
400 S_regcppop(pTHX_ const regexp *rex)
405 GET_RE_DEBUG_FLAGS_DECL;
407 PERL_ARGS_ASSERT_REGCPPOP;
409 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
411 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
412 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
413 input = (char *) SSPOPPTR;
414 *PL_reglastcloseparen = SSPOPINT;
415 *PL_reglastparen = SSPOPINT;
416 PL_regsize = SSPOPINT;
417 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
419 i -= REGCP_OTHER_ELEMS;
420 /* Now restore the parentheses context. */
421 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
423 U32 paren = (U32)SSPOPINT;
424 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
425 PL_regoffs[paren].start = SSPOPINT;
427 if (paren <= *PL_reglastparen)
428 PL_regoffs[paren].end = tmps;
430 PerlIO_printf(Perl_debug_log,
431 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
432 (UV)paren, (IV)PL_regoffs[paren].start,
433 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
434 (IV)PL_regoffs[paren].end,
435 (paren > *PL_reglastparen ? "(no)" : ""));
439 if (*PL_reglastparen + 1 <= rex->nparens) {
440 PerlIO_printf(Perl_debug_log,
441 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
442 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
446 /* It would seem that the similar code in regtry()
447 * already takes care of this, and in fact it is in
448 * a better location to since this code can #if 0-ed out
449 * but the code in regtry() is needed or otherwise tests
450 * requiring null fields (pat.t#187 and split.t#{13,14}
451 * (as of patchlevel 7877) will fail. Then again,
452 * this code seems to be necessary or otherwise
453 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
454 * --jhi updated by dapm */
455 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
457 PL_regoffs[i].start = -1;
458 PL_regoffs[i].end = -1;
464 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
467 * pregexec and friends
470 #ifndef PERL_IN_XSUB_RE
472 - pregexec - match a regexp against a string
475 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
476 char *strbeg, I32 minend, SV *screamer, U32 nosave)
477 /* strend: pointer to null at end of string */
478 /* strbeg: real beginning of string */
479 /* minend: end of match must be >=minend after stringarg. */
480 /* nosave: For optimizations. */
482 PERL_ARGS_ASSERT_PREGEXEC;
485 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
486 nosave ? 0 : REXEC_COPY_STR);
491 * Need to implement the following flags for reg_anch:
493 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
495 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
496 * INTUIT_AUTORITATIVE_ML
497 * INTUIT_ONCE_NOML - Intuit can match in one location only.
500 * Another flag for this function: SECOND_TIME (so that float substrs
501 * with giant delta may be not rechecked).
504 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
506 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
507 Otherwise, only SvCUR(sv) is used to get strbeg. */
509 /* XXXX We assume that strpos is strbeg unless sv. */
511 /* XXXX Some places assume that there is a fixed substring.
512 An update may be needed if optimizer marks as "INTUITable"
513 RExen without fixed substrings. Similarly, it is assumed that
514 lengths of all the strings are no more than minlen, thus they
515 cannot come from lookahead.
516 (Or minlen should take into account lookahead.)
517 NOTE: Some of this comment is not correct. minlen does now take account
518 of lookahead/behind. Further research is required. -- demerphq
522 /* A failure to find a constant substring means that there is no need to make
523 an expensive call to REx engine, thus we celebrate a failure. Similarly,
524 finding a substring too deep into the string means that less calls to
525 regtry() should be needed.
527 REx compiler's optimizer found 4 possible hints:
528 a) Anchored substring;
530 c) Whether we are anchored (beginning-of-line or \G);
531 d) First node (of those at offset 0) which may distingush positions;
532 We use a)b)d) and multiline-part of c), and try to find a position in the
533 string which does not contradict any of them.
536 /* Most of decisions we do here should have been done at compile time.
537 The nodes of the REx which we used for the search should have been
538 deleted from the finite automaton. */
541 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
542 char *strend, const U32 flags, re_scream_pos_data *data)
545 struct regexp *const prog = (struct regexp *)SvANY(rx);
546 register I32 start_shift = 0;
547 /* Should be nonnegative! */
548 register I32 end_shift = 0;
553 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
555 register char *other_last = NULL; /* other substr checked before this */
556 char *check_at = NULL; /* check substr found at this pos */
557 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
558 RXi_GET_DECL(prog,progi);
560 const char * const i_strpos = strpos;
562 GET_RE_DEBUG_FLAGS_DECL;
564 PERL_ARGS_ASSERT_RE_INTUIT_START;
566 RX_MATCH_UTF8_set(rx,utf8_target);
569 PL_reg_flags |= RF_utf8;
572 debug_start_match(rx, utf8_target, strpos, strend,
573 sv ? "Guessing start of match in sv for"
574 : "Guessing start of match in string for");
577 /* CHR_DIST() would be more correct here but it makes things slow. */
578 if (prog->minlen > strend - strpos) {
579 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
580 "String too short... [re_intuit_start]\n"));
584 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
587 if (!prog->check_utf8 && prog->check_substr)
588 to_utf8_substr(prog);
589 check = prog->check_utf8;
591 if (!prog->check_substr && prog->check_utf8)
592 to_byte_substr(prog);
593 check = prog->check_substr;
595 if (check == &PL_sv_undef) {
596 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
597 "Non-utf8 string cannot match utf8 check string\n"));
600 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
601 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
602 || ( (prog->extflags & RXf_ANCH_BOL)
603 && !multiline ) ); /* Check after \n? */
606 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
607 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
608 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
610 && (strpos != strbeg)) {
611 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
614 if (prog->check_offset_min == prog->check_offset_max &&
615 !(prog->extflags & RXf_CANY_SEEN)) {
616 /* Substring at constant offset from beg-of-str... */
619 s = HOP3c(strpos, prog->check_offset_min, strend);
622 slen = SvCUR(check); /* >= 1 */
624 if ( strend - s > slen || strend - s < slen - 1
625 || (strend - s == slen && strend[-1] != '\n')) {
626 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
629 /* Now should match s[0..slen-2] */
631 if (slen && (*SvPVX_const(check) != *s
633 && memNE(SvPVX_const(check), s, slen)))) {
635 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
639 else if (*SvPVX_const(check) != *s
640 || ((slen = SvCUR(check)) > 1
641 && memNE(SvPVX_const(check), s, slen)))
644 goto success_at_start;
647 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
649 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
650 end_shift = prog->check_end_shift;
653 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
654 - (SvTAIL(check) != 0);
655 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
657 if (end_shift < eshift)
661 else { /* Can match at random position */
664 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
665 end_shift = prog->check_end_shift;
667 /* end shift should be non negative here */
670 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
672 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
673 (IV)end_shift, RX_PRECOMP(prog));
677 /* Find a possible match in the region s..strend by looking for
678 the "check" substring in the region corrected by start/end_shift. */
681 I32 srch_start_shift = start_shift;
682 I32 srch_end_shift = end_shift;
683 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
684 srch_end_shift -= ((strbeg - s) - srch_start_shift);
685 srch_start_shift = strbeg - s;
687 DEBUG_OPTIMISE_MORE_r({
688 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
689 (IV)prog->check_offset_min,
690 (IV)srch_start_shift,
692 (IV)prog->check_end_shift);
695 if (flags & REXEC_SCREAM) {
696 I32 p = -1; /* Internal iterator of scream. */
697 I32 * const pp = data ? data->scream_pos : &p;
699 if (PL_screamfirst[BmRARE(check)] >= 0
700 || ( BmRARE(check) == '\n'
701 && (BmPREVIOUS(check) == SvCUR(check) - 1)
703 s = screaminstr(sv, check,
704 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
707 /* we may be pointing at the wrong string */
708 if (s && RXp_MATCH_COPIED(prog))
709 s = strbeg + (s - SvPVX_const(sv));
711 *data->scream_olds = s;
716 if (prog->extflags & RXf_CANY_SEEN) {
717 start_point= (U8*)(s + srch_start_shift);
718 end_point= (U8*)(strend - srch_end_shift);
720 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
721 end_point= HOP3(strend, -srch_end_shift, strbeg);
723 DEBUG_OPTIMISE_MORE_r({
724 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
725 (int)(end_point - start_point),
726 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
730 s = fbm_instr( start_point, end_point,
731 check, multiline ? FBMrf_MULTILINE : 0);
734 /* Update the count-of-usability, remove useless subpatterns,
738 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
739 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
740 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
741 (s ? "Found" : "Did not find"),
742 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
743 ? "anchored" : "floating"),
746 (s ? " at offset " : "...\n") );
751 /* Finish the diagnostic message */
752 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
754 /* XXX dmq: first branch is for positive lookbehind...
755 Our check string is offset from the beginning of the pattern.
756 So we need to do any stclass tests offset forward from that
765 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
766 Start with the other substr.
767 XXXX no SCREAM optimization yet - and a very coarse implementation
768 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
769 *always* match. Probably should be marked during compile...
770 Probably it is right to do no SCREAM here...
773 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
774 : (prog->float_substr && prog->anchored_substr))
776 /* Take into account the "other" substring. */
777 /* XXXX May be hopelessly wrong for UTF... */
780 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
783 char * const last = HOP3c(s, -start_shift, strbeg);
785 char * const saved_s = s;
788 t = s - prog->check_offset_max;
789 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
791 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
796 t = HOP3c(t, prog->anchored_offset, strend);
797 if (t < other_last) /* These positions already checked */
799 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
802 /* XXXX It is not documented what units *_offsets are in.
803 We assume bytes, but this is clearly wrong.
804 Meaning this code needs to be carefully reviewed for errors.
808 /* On end-of-str: see comment below. */
809 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
810 if (must == &PL_sv_undef) {
812 DEBUG_r(must = prog->anchored_utf8); /* for debug */
817 HOP3(HOP3(last1, prog->anchored_offset, strend)
818 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
820 multiline ? FBMrf_MULTILINE : 0
823 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
824 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
825 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
826 (s ? "Found" : "Contradicts"),
827 quoted, RE_SV_TAIL(must));
832 if (last1 >= last2) {
833 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
834 ", giving up...\n"));
837 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
838 ", trying floating at offset %ld...\n",
839 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
840 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
841 s = HOP3c(last, 1, strend);
845 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
846 (long)(s - i_strpos)));
847 t = HOP3c(s, -prog->anchored_offset, strbeg);
848 other_last = HOP3c(s, 1, strend);
856 else { /* Take into account the floating substring. */
858 char * const saved_s = s;
861 t = HOP3c(s, -start_shift, strbeg);
863 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
864 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
865 last = HOP3c(t, prog->float_max_offset, strend);
866 s = HOP3c(t, prog->float_min_offset, strend);
869 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
870 must = utf8_target ? prog->float_utf8 : prog->float_substr;
871 /* fbm_instr() takes into account exact value of end-of-str
872 if the check is SvTAIL(ed). Since false positives are OK,
873 and end-of-str is not later than strend we are OK. */
874 if (must == &PL_sv_undef) {
876 DEBUG_r(must = prog->float_utf8); /* for debug message */
879 s = fbm_instr((unsigned char*)s,
880 (unsigned char*)last + SvCUR(must)
882 must, multiline ? FBMrf_MULTILINE : 0);
884 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
885 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
886 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
887 (s ? "Found" : "Contradicts"),
888 quoted, RE_SV_TAIL(must));
892 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
893 ", giving up...\n"));
896 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
897 ", trying anchored starting at offset %ld...\n",
898 (long)(saved_s + 1 - i_strpos)));
900 s = HOP3c(t, 1, strend);
904 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
905 (long)(s - i_strpos)));
906 other_last = s; /* Fix this later. --Hugo */
916 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
918 DEBUG_OPTIMISE_MORE_r(
919 PerlIO_printf(Perl_debug_log,
920 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
921 (IV)prog->check_offset_min,
922 (IV)prog->check_offset_max,
930 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
932 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
935 /* Fixed substring is found far enough so that the match
936 cannot start at strpos. */
938 if (ml_anch && t[-1] != '\n') {
939 /* Eventually fbm_*() should handle this, but often
940 anchored_offset is not 0, so this check will not be wasted. */
941 /* XXXX In the code below we prefer to look for "^" even in
942 presence of anchored substrings. And we search even
943 beyond the found float position. These pessimizations
944 are historical artefacts only. */
946 while (t < strend - prog->minlen) {
948 if (t < check_at - prog->check_offset_min) {
949 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
950 /* Since we moved from the found position,
951 we definitely contradict the found anchored
952 substr. Due to the above check we do not
953 contradict "check" substr.
954 Thus we can arrive here only if check substr
955 is float. Redo checking for "other"=="fixed".
958 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
959 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
960 goto do_other_anchored;
962 /* We don't contradict the found floating substring. */
963 /* XXXX Why not check for STCLASS? */
965 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
966 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
969 /* Position contradicts check-string */
970 /* XXXX probably better to look for check-string
971 than for "\n", so one should lower the limit for t? */
972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
973 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
974 other_last = strpos = s = t + 1;
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
980 PL_colors[0], PL_colors[1]));
984 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
985 PL_colors[0], PL_colors[1]));
989 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
992 /* The found string does not prohibit matching at strpos,
993 - no optimization of calling REx engine can be performed,
994 unless it was an MBOL and we are not after MBOL,
995 or a future STCLASS check will fail this. */
997 /* Even in this situation we may use MBOL flag if strpos is offset
998 wrt the start of the string. */
999 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1000 && (strpos != strbeg) && strpos[-1] != '\n'
1001 /* May be due to an implicit anchor of m{.*foo} */
1002 && !(prog->intflags & PREGf_IMPLICIT))
1007 DEBUG_EXECUTE_r( if (ml_anch)
1008 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1009 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1012 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1014 prog->check_utf8 /* Could be deleted already */
1015 && --BmUSEFUL(prog->check_utf8) < 0
1016 && (prog->check_utf8 == prog->float_utf8)
1018 prog->check_substr /* Could be deleted already */
1019 && --BmUSEFUL(prog->check_substr) < 0
1020 && (prog->check_substr == prog->float_substr)
1023 /* If flags & SOMETHING - do not do it many times on the same match */
1024 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1025 /* XXX Does the destruction order has to change with utf8_target? */
1026 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1027 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1028 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1029 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1030 check = NULL; /* abort */
1032 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevent flag
1033 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1034 if (prog->intflags & PREGf_IMPLICIT)
1035 prog->extflags &= ~RXf_ANCH_MBOL;
1036 /* XXXX This is a remnant of the old implementation. It
1037 looks wasteful, since now INTUIT can use many
1038 other heuristics. */
1039 prog->extflags &= ~RXf_USE_INTUIT;
1040 /* XXXX What other flags might need to be cleared in this branch? */
1046 /* Last resort... */
1047 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1048 /* trie stclasses are too expensive to use here, we are better off to
1049 leave it to regmatch itself */
1050 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1051 /* minlen == 0 is possible if regstclass is \b or \B,
1052 and the fixed substr is ''$.
1053 Since minlen is already taken into account, s+1 is before strend;
1054 accidentally, minlen >= 1 guaranties no false positives at s + 1
1055 even for \b or \B. But (minlen? 1 : 0) below assumes that
1056 regstclass does not come from lookahead... */
1057 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1058 This leaves EXACTF, EXACTFU only, which are dealt with in find_byclass(). */
1059 const U8* const str = (U8*)STRING(progi->regstclass);
1060 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1061 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1064 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1065 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1066 else if (prog->float_substr || prog->float_utf8)
1067 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1071 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1072 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1075 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1078 const char *what = NULL;
1080 if (endpos == strend) {
1081 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1082 "Could not match STCLASS...\n") );
1085 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1086 "This position contradicts STCLASS...\n") );
1087 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1089 /* Contradict one of substrings */
1090 if (prog->anchored_substr || prog->anchored_utf8) {
1091 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1092 DEBUG_EXECUTE_r( what = "anchored" );
1094 s = HOP3c(t, 1, strend);
1095 if (s + start_shift + end_shift > strend) {
1096 /* XXXX Should be taken into account earlier? */
1097 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1098 "Could not match STCLASS...\n") );
1103 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1104 "Looking for %s substr starting at offset %ld...\n",
1105 what, (long)(s + start_shift - i_strpos)) );
1108 /* Have both, check_string is floating */
1109 if (t + start_shift >= check_at) /* Contradicts floating=check */
1110 goto retry_floating_check;
1111 /* Recheck anchored substring, but not floating... */
1115 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1116 "Looking for anchored substr starting at offset %ld...\n",
1117 (long)(other_last - i_strpos)) );
1118 goto do_other_anchored;
1120 /* Another way we could have checked stclass at the
1121 current position only: */
1126 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1127 "Looking for /%s^%s/m starting at offset %ld...\n",
1128 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1131 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1133 /* Check is floating subtring. */
1134 retry_floating_check:
1135 t = check_at - start_shift;
1136 DEBUG_EXECUTE_r( what = "floating" );
1137 goto hop_and_restart;
1140 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1141 "By STCLASS: moving %ld --> %ld\n",
1142 (long)(t - i_strpos), (long)(s - i_strpos))
1146 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1147 "Does not contradict STCLASS...\n");
1152 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1153 PL_colors[4], (check ? "Guessed" : "Giving up"),
1154 PL_colors[5], (long)(s - i_strpos)) );
1157 fail_finish: /* Substring not found */
1158 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1159 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1161 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1162 PL_colors[4], PL_colors[5]));
1166 #define DECL_TRIE_TYPE(scan) \
1167 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1168 trie_type = (scan->flags != EXACT) \
1169 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1170 : (utf8_target ? trie_utf8 : trie_plain)
1172 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1173 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1174 switch (trie_type) { \
1175 case trie_utf8_fold: \
1176 if ( foldlen>0 ) { \
1177 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1182 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1183 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1184 foldlen -= UNISKIP( uvc ); \
1185 uscan = foldbuf + UNISKIP( uvc ); \
1188 case trie_latin_utf8_fold: \
1189 if ( foldlen>0 ) { \
1190 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1196 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1197 foldlen -= UNISKIP( uvc ); \
1198 uscan = foldbuf + UNISKIP( uvc ); \
1202 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1209 charid = trie->charmap[ uvc ]; \
1213 if (widecharmap) { \
1214 SV** const svpp = hv_fetch(widecharmap, \
1215 (char*)&uvc, sizeof(UV), 0); \
1217 charid = (U16)SvIV(*svpp); \
1222 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1224 char *my_strend= (char *)strend; \
1227 foldEQ_utf8(s, &my_strend, 0, utf8_target, \
1228 m, NULL, ln, cBOOL(UTF_PATTERN))) \
1229 && (!reginfo || regtry(reginfo, &s)) ) \
1232 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1233 uvchr_to_utf8(tmpbuf, c); \
1234 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1236 && (f == c1 || f == c2) \
1238 foldEQ_utf8(s, &my_strend, 0, utf8_target,\
1239 m, NULL, ln, cBOOL(UTF_PATTERN)))\
1240 && (!reginfo || regtry(reginfo, &s)) ) \
1246 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1250 case EXACTFU: folder = foldEQ_latin1; break; \
1251 case EXACTFL: folder = foldEQ_locale; break; \
1252 case EXACTF: folder = foldEQ; break; \
1254 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(c)); \
1258 && (ln == 1 || folder(s, m, ln)) \
1259 && (!reginfo || regtry(reginfo, &s)) ) \
1265 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1267 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1273 #define REXEC_FBC_SCAN(CoDe) \
1275 while (s < strend) { \
1281 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1282 REXEC_FBC_UTF8_SCAN( \
1284 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1293 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1296 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1305 #define REXEC_FBC_TRYIT \
1306 if ((!reginfo || regtry(reginfo, &s))) \
1309 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1310 if (utf8_target) { \
1311 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1314 REXEC_FBC_CLASS_SCAN(CoNd); \
1318 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1319 if (utf8_target) { \
1321 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1324 REXEC_FBC_CLASS_SCAN(CoNd); \
1328 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1329 PL_reg_flags |= RF_tainted; \
1330 if (utf8_target) { \
1331 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1334 REXEC_FBC_CLASS_SCAN(CoNd); \
1338 #define DUMP_EXEC_POS(li,s,doutf8) \
1339 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1341 /* We know what class REx starts with. Try to find this position... */
1342 /* if reginfo is NULL, its a dryrun */
1343 /* annoyingly all the vars in this routine have different names from their counterparts
1344 in regmatch. /grrr */
1347 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1348 const char *strend, regmatch_info *reginfo)
1351 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1355 register STRLEN uskip;
1359 register I32 tmp = 1; /* Scratch variable? */
1360 register const bool utf8_target = PL_reg_match_utf8;
1361 RXi_GET_DECL(prog,progi);
1363 PERL_ARGS_ASSERT_FIND_BYCLASS;
1365 /* We know what class it must start with. */
1369 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_NONBITMAP) ||
1370 !UTF8_IS_INVARIANT((U8)s[0]) ?
1371 reginclass(prog, c, (U8*)s, 0, utf8_target) :
1372 REGINCLASS(prog, c, (U8*)s));
1375 while (s < strend) {
1378 if (REGINCLASS(prog, c, (U8*)s) ||
1379 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1380 /* The assignment of 2 is intentional:
1381 * for the folded sharp s, the skip is 2. */
1382 (skip = SHARP_S_SKIP))) {
1383 if (tmp && (!reginfo || regtry(reginfo, &s)))
1396 if (tmp && (!reginfo || regtry(reginfo, &s)))
1405 ln = STR_LEN(c); /* length to match in octets/bytes */
1406 lnc = (I32) ln; /* length to match in characters */
1408 STRLEN ulen1, ulen2;
1410 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1411 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1412 /* used by commented-out code below */
1413 /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
1415 /* XXX: Since the node will be case folded at compile
1416 time this logic is a little odd, although im not
1417 sure that its actually wrong. --dmq */
1419 c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1420 c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1422 /* XXX: This is kinda strange. to_utf8_XYZ returns the
1423 codepoint of the first character in the converted
1424 form, yet originally we did the extra step.
1425 No tests fail by commenting this code out however
1426 so Ive left it out. -- dmq.
1428 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1430 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1435 while (sm < ((U8 *) m + ln)) {
1442 if (utf8_target || OP(c) == EXACTFU) {
1444 /* Micro sign folds to GREEK SMALL LETTER MU;
1445 LATIN_SMALL_LETTER_SHARP_S folds to 'ss', and this sets
1446 c2 to the first 's' of the pair, and the code below will
1448 c2 = (c1 == MICRO_SIGN)
1449 ? GREEK_SMALL_LETTER_MU
1450 : (c1 == LATIN_SMALL_LETTER_SHARP_S)
1452 : PL_fold_latin1[c1];
1453 } else c2 = PL_fold[c1];
1461 c2 = PL_fold_locale[c1];
1463 e = HOP3c(strend, -((I32)lnc), s);
1465 if (!reginfo && e < s)
1466 e = s; /* Due to minlen logic of intuit() */
1468 /* The idea in the EXACTF* cases is to first find the
1469 * first character of the EXACTF* node and then, if
1470 * necessary, case-insensitively compare the full
1471 * text of the node. The c1 and c2 are the first
1472 * characters (though in Unicode it gets a bit
1473 * more complicated because there are more cases
1474 * than just upper and lower: one needs to use
1475 * the so-called folding case for case-insensitive
1476 * matching (called "loose matching" in Unicode).
1477 * foldEQ_utf8() will do just that. */
1479 if (utf8_target || UTF_PATTERN) {
1481 U8 tmpbuf [UTF8_MAXBYTES+1];
1484 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1486 /* Upper and lower of 1st char are equal -
1487 * probably not a "letter". */
1490 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1495 REXEC_FBC_EXACTISH_CHECK(c == c1);
1501 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1507 /* Handle some of the three Greek sigmas cases.
1508 * Note that not all the possible combinations
1509 * are handled here: some of them are handled
1510 * by the standard folding rules, and some of
1511 * them (the character class or ANYOF cases)
1512 * are handled during compiletime in
1513 * regexec.c:S_regclass(). */
1514 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1515 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1516 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1518 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1523 /* Neither pattern nor string are UTF8 */
1525 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1527 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1531 PL_reg_flags |= RF_tainted;
1538 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1539 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1541 tmp = ((OP(c) == BOUND ?
1542 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1543 LOAD_UTF8_CHARCLASS_ALNUM();
1544 REXEC_FBC_UTF8_SCAN(
1545 if (tmp == !(OP(c) == BOUND ?
1546 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)) :
1547 isALNUM_LC_utf8((U8*)s)))
1554 else { /* Not utf8 */
1555 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1556 tmp = cBOOL((OP(c) == BOUNDL)
1558 : (isWORDCHAR_L1(tmp)
1559 && (isASCII(tmp) || (FLAGS(c) & USE_UNI))));
1564 : (isWORDCHAR_L1((U8) *s)
1565 && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))))
1572 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1576 PL_reg_flags |= RF_tainted;
1583 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1584 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1586 tmp = ((OP(c) == NBOUND ?
1587 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1588 LOAD_UTF8_CHARCLASS_ALNUM();
1589 REXEC_FBC_UTF8_SCAN(
1590 if (tmp == !(OP(c) == NBOUND ?
1591 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)) :
1592 isALNUM_LC_utf8((U8*)s)))
1594 else REXEC_FBC_TRYIT;
1598 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1599 tmp = cBOOL((OP(c) == NBOUNDL)
1601 : (isWORDCHAR_L1(tmp)
1602 && (isASCII(tmp) || (FLAGS(c) & USE_UNI))));
1607 : (isWORDCHAR_L1((U8) *s)
1608 && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))))
1612 else REXEC_FBC_TRYIT;
1615 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1619 REXEC_FBC_CSCAN_PRELOAD(
1620 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1621 swash_fetch(RE_utf8_perl_word, (U8*)s, utf8_target),
1622 (FLAGS(c) & USE_UNI) ? isWORDCHAR_L1((U8) *s) : isALNUM(*s)
1625 REXEC_FBC_CSCAN_TAINT(
1626 isALNUM_LC_utf8((U8*)s),
1630 REXEC_FBC_CSCAN_PRELOAD(
1631 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1632 !swash_fetch(RE_utf8_perl_word, (U8*)s, utf8_target),
1633 ! ((FLAGS(c) & USE_UNI) ? isWORDCHAR_L1((U8) *s) : isALNUM(*s))
1636 REXEC_FBC_CSCAN_TAINT(
1637 !isALNUM_LC_utf8((U8*)s),
1641 REXEC_FBC_CSCAN_PRELOAD(
1642 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1643 *s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, utf8_target),
1644 isSPACE_L1((U8) *s) && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI))
1647 REXEC_FBC_CSCAN_TAINT(
1648 isSPACE_LC_utf8((U8*)s),
1652 REXEC_FBC_CSCAN_PRELOAD(
1653 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1654 !(*s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, utf8_target)),
1655 !(isSPACE_L1((U8) *s) && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))
1658 REXEC_FBC_CSCAN_TAINT(
1659 !isSPACE_LC_utf8((U8*)s),
1663 REXEC_FBC_CSCAN_PRELOAD(
1664 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1665 swash_fetch(RE_utf8_posix_digit,(U8*)s, utf8_target),
1669 REXEC_FBC_CSCAN_TAINT(
1670 isDIGIT_LC_utf8((U8*)s),
1674 REXEC_FBC_CSCAN_PRELOAD(
1675 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1676 !swash_fetch(RE_utf8_posix_digit,(U8*)s, utf8_target),
1680 REXEC_FBC_CSCAN_TAINT(
1681 !isDIGIT_LC_utf8((U8*)s),
1687 is_LNBREAK_latin1(s)
1697 !is_VERTWS_latin1(s)
1702 is_HORIZWS_latin1(s)
1706 !is_HORIZWS_utf8(s),
1707 !is_HORIZWS_latin1(s)
1713 /* what trie are we using right now */
1715 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1717 = (reg_trie_data*)progi->data->data[ aho->trie ];
1718 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1720 const char *last_start = strend - trie->minlen;
1722 const char *real_start = s;
1724 STRLEN maxlen = trie->maxlen;
1726 U8 **points; /* map of where we were in the input string
1727 when reading a given char. For ASCII this
1728 is unnecessary overhead as the relationship
1729 is always 1:1, but for Unicode, especially
1730 case folded Unicode this is not true. */
1731 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1735 GET_RE_DEBUG_FLAGS_DECL;
1737 /* We can't just allocate points here. We need to wrap it in
1738 * an SV so it gets freed properly if there is a croak while
1739 * running the match */
1742 sv_points=newSV(maxlen * sizeof(U8 *));
1743 SvCUR_set(sv_points,
1744 maxlen * sizeof(U8 *));
1745 SvPOK_on(sv_points);
1746 sv_2mortal(sv_points);
1747 points=(U8**)SvPV_nolen(sv_points );
1748 if ( trie_type != trie_utf8_fold
1749 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1752 bitmap=(U8*)trie->bitmap;
1754 bitmap=(U8*)ANYOF_BITMAP(c);
1756 /* this is the Aho-Corasick algorithm modified a touch
1757 to include special handling for long "unknown char"
1758 sequences. The basic idea being that we use AC as long
1759 as we are dealing with a possible matching char, when
1760 we encounter an unknown char (and we have not encountered
1761 an accepting state) we scan forward until we find a legal
1763 AC matching is basically that of trie matching, except
1764 that when we encounter a failing transition, we fall back
1765 to the current states "fail state", and try the current char
1766 again, a process we repeat until we reach the root state,
1767 state 1, or a legal transition. If we fail on the root state
1768 then we can either terminate if we have reached an accepting
1769 state previously, or restart the entire process from the beginning
1773 while (s <= last_start) {
1774 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1782 U8 *uscan = (U8*)NULL;
1783 U8 *leftmost = NULL;
1785 U32 accepted_word= 0;
1789 while ( state && uc <= (U8*)strend ) {
1791 U32 word = aho->states[ state ].wordnum;
1795 DEBUG_TRIE_EXECUTE_r(
1796 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1797 dump_exec_pos( (char *)uc, c, strend, real_start,
1798 (char *)uc, utf8_target );
1799 PerlIO_printf( Perl_debug_log,
1800 " Scanning for legal start char...\n");
1804 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1808 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1814 if (uc >(U8*)last_start) break;
1818 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1819 if (!leftmost || lpos < leftmost) {
1820 DEBUG_r(accepted_word=word);
1826 points[pointpos++ % maxlen]= uc;
1827 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1828 uscan, len, uvc, charid, foldlen,
1830 DEBUG_TRIE_EXECUTE_r({
1831 dump_exec_pos( (char *)uc, c, strend, real_start,
1833 PerlIO_printf(Perl_debug_log,
1834 " Charid:%3u CP:%4"UVxf" ",
1840 word = aho->states[ state ].wordnum;
1842 base = aho->states[ state ].trans.base;
1844 DEBUG_TRIE_EXECUTE_r({
1846 dump_exec_pos( (char *)uc, c, strend, real_start,
1848 PerlIO_printf( Perl_debug_log,
1849 "%sState: %4"UVxf", word=%"UVxf,
1850 failed ? " Fail transition to " : "",
1851 (UV)state, (UV)word);
1857 ( ((offset = base + charid
1858 - 1 - trie->uniquecharcount)) >= 0)
1859 && ((U32)offset < trie->lasttrans)
1860 && trie->trans[offset].check == state
1861 && (tmp=trie->trans[offset].next))
1863 DEBUG_TRIE_EXECUTE_r(
1864 PerlIO_printf( Perl_debug_log," - legal\n"));
1869 DEBUG_TRIE_EXECUTE_r(
1870 PerlIO_printf( Perl_debug_log," - fail\n"));
1872 state = aho->fail[state];
1876 /* we must be accepting here */
1877 DEBUG_TRIE_EXECUTE_r(
1878 PerlIO_printf( Perl_debug_log," - accepting\n"));
1887 if (!state) state = 1;
1890 if ( aho->states[ state ].wordnum ) {
1891 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1892 if (!leftmost || lpos < leftmost) {
1893 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1898 s = (char*)leftmost;
1899 DEBUG_TRIE_EXECUTE_r({
1901 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1902 (UV)accepted_word, (IV)(s - real_start)
1905 if (!reginfo || regtry(reginfo, &s)) {
1911 DEBUG_TRIE_EXECUTE_r({
1912 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1915 DEBUG_TRIE_EXECUTE_r(
1916 PerlIO_printf( Perl_debug_log,"No match.\n"));
1925 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1935 - regexec_flags - match a regexp against a string
1938 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1939 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1940 /* strend: pointer to null at end of string */
1941 /* strbeg: real beginning of string */
1942 /* minend: end of match must be >=minend after stringarg. */
1943 /* data: May be used for some additional optimizations.
1944 Currently its only used, with a U32 cast, for transmitting
1945 the ganch offset when doing a /g match. This will change */
1946 /* nosave: For optimizations. */
1949 struct regexp *const prog = (struct regexp *)SvANY(rx);
1950 /*register*/ char *s;
1951 register regnode *c;
1952 /*register*/ char *startpos = stringarg;
1953 I32 minlen; /* must match at least this many chars */
1954 I32 dontbother = 0; /* how many characters not to try at end */
1955 I32 end_shift = 0; /* Same for the end. */ /* CC */
1956 I32 scream_pos = -1; /* Internal iterator of scream. */
1957 char *scream_olds = NULL;
1958 const bool utf8_target = cBOOL(DO_UTF8(sv));
1960 RXi_GET_DECL(prog,progi);
1961 regmatch_info reginfo; /* create some info to pass to regtry etc */
1962 regexp_paren_pair *swap = NULL;
1963 GET_RE_DEBUG_FLAGS_DECL;
1965 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1966 PERL_UNUSED_ARG(data);
1968 /* Be paranoid... */
1969 if (prog == NULL || startpos == NULL) {
1970 Perl_croak(aTHX_ "NULL regexp parameter");
1974 multiline = prog->extflags & RXf_PMf_MULTILINE;
1975 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
1977 RX_MATCH_UTF8_set(rx, utf8_target);
1979 debug_start_match(rx, utf8_target, startpos, strend,
1983 minlen = prog->minlen;
1985 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1986 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1987 "String too short [regexec_flags]...\n"));
1992 /* Check validity of program. */
1993 if (UCHARAT(progi->program) != REG_MAGIC) {
1994 Perl_croak(aTHX_ "corrupted regexp program");
1998 PL_reg_eval_set = 0;
2002 PL_reg_flags |= RF_utf8;
2004 /* Mark beginning of line for ^ and lookbehind. */
2005 reginfo.bol = startpos; /* XXX not used ??? */
2009 /* Mark end of line for $ (and such) */
2012 /* see how far we have to get to not match where we matched before */
2013 reginfo.till = startpos+minend;
2015 /* If there is a "must appear" string, look for it. */
2018 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2020 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2021 reginfo.ganch = startpos + prog->gofs;
2022 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2023 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2024 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2026 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2027 && mg->mg_len >= 0) {
2028 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2029 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2030 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2032 if (prog->extflags & RXf_ANCH_GPOS) {
2033 if (s > reginfo.ganch)
2035 s = reginfo.ganch - prog->gofs;
2036 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2037 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2043 reginfo.ganch = strbeg + PTR2UV(data);
2044 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2045 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2047 } else { /* pos() not defined */
2048 reginfo.ganch = strbeg;
2049 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2050 "GPOS: reginfo.ganch = strbeg\n"));
2053 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2054 /* We have to be careful. If the previous successful match
2055 was from this regex we don't want a subsequent partially
2056 successful match to clobber the old results.
2057 So when we detect this possibility we add a swap buffer
2058 to the re, and switch the buffer each match. If we fail
2059 we switch it back, otherwise we leave it swapped.
2062 /* do we need a save destructor here for eval dies? */
2063 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2065 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2066 re_scream_pos_data d;
2068 d.scream_olds = &scream_olds;
2069 d.scream_pos = &scream_pos;
2070 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2072 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2073 goto phooey; /* not present */
2079 /* Simplest case: anchored match need be tried only once. */
2080 /* [unless only anchor is BOL and multiline is set] */
2081 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2082 if (s == startpos && regtry(®info, &startpos))
2084 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2085 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2090 dontbother = minlen - 1;
2091 end = HOP3c(strend, -dontbother, strbeg) - 1;
2092 /* for multiline we only have to try after newlines */
2093 if (prog->check_substr || prog->check_utf8) {
2094 /* because of the goto we can not easily reuse the macros for bifurcating the
2095 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2098 goto after_try_utf8;
2100 if (regtry(®info, &s)) {
2107 if (prog->extflags & RXf_USE_INTUIT) {
2108 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2117 } /* end search for check string in unicode */
2119 if (s == startpos) {
2120 goto after_try_latin;
2123 if (regtry(®info, &s)) {
2130 if (prog->extflags & RXf_USE_INTUIT) {
2131 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2140 } /* end search for check string in latin*/
2141 } /* end search for check string */
2142 else { /* search for newline */
2144 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2147 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2149 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2150 if (regtry(®info, &s))
2154 } /* end search for newline */
2155 } /* end anchored/multiline check string search */
2157 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2159 /* the warning about reginfo.ganch being used without intialization
2160 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2161 and we only enter this block when the same bit is set. */
2162 char *tmp_s = reginfo.ganch - prog->gofs;
2164 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2169 /* Messy cases: unanchored match. */
2170 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2171 /* we have /x+whatever/ */
2172 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2177 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2178 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2179 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2184 DEBUG_EXECUTE_r( did_match = 1 );
2185 if (regtry(®info, &s)) goto got_it;
2187 while (s < strend && *s == ch)
2195 DEBUG_EXECUTE_r( did_match = 1 );
2196 if (regtry(®info, &s)) goto got_it;
2198 while (s < strend && *s == ch)
2203 DEBUG_EXECUTE_r(if (!did_match)
2204 PerlIO_printf(Perl_debug_log,
2205 "Did not find anchored character...\n")
2208 else if (prog->anchored_substr != NULL
2209 || prog->anchored_utf8 != NULL
2210 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2211 && prog->float_max_offset < strend - s)) {
2216 char *last1; /* Last position checked before */
2220 if (prog->anchored_substr || prog->anchored_utf8) {
2221 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2222 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2223 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2224 back_max = back_min = prog->anchored_offset;
2226 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2227 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2228 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2229 back_max = prog->float_max_offset;
2230 back_min = prog->float_min_offset;
2234 if (must == &PL_sv_undef)
2235 /* could not downgrade utf8 check substring, so must fail */
2241 last = HOP3c(strend, /* Cannot start after this */
2242 -(I32)(CHR_SVLEN(must)
2243 - (SvTAIL(must) != 0) + back_min), strbeg);
2246 last1 = HOPc(s, -1);
2248 last1 = s - 1; /* bogus */
2250 /* XXXX check_substr already used to find "s", can optimize if
2251 check_substr==must. */
2253 dontbother = end_shift;
2254 strend = HOPc(strend, -dontbother);
2255 while ( (s <= last) &&
2256 ((flags & REXEC_SCREAM)
2257 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2258 end_shift, &scream_pos, 0))
2259 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2260 (unsigned char*)strend, must,
2261 multiline ? FBMrf_MULTILINE : 0))) ) {
2262 /* we may be pointing at the wrong string */
2263 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2264 s = strbeg + (s - SvPVX_const(sv));
2265 DEBUG_EXECUTE_r( did_match = 1 );
2266 if (HOPc(s, -back_max) > last1) {
2267 last1 = HOPc(s, -back_min);
2268 s = HOPc(s, -back_max);
2271 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2273 last1 = HOPc(s, -back_min);
2277 while (s <= last1) {
2278 if (regtry(®info, &s))
2284 while (s <= last1) {
2285 if (regtry(®info, &s))
2291 DEBUG_EXECUTE_r(if (!did_match) {
2292 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2293 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2294 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2295 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2296 ? "anchored" : "floating"),
2297 quoted, RE_SV_TAIL(must));
2301 else if ( (c = progi->regstclass) ) {
2303 const OPCODE op = OP(progi->regstclass);
2304 /* don't bother with what can't match */
2305 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2306 strend = HOPc(strend, -(minlen - 1));
2309 SV * const prop = sv_newmortal();
2310 regprop(prog, prop, c);
2312 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2314 PerlIO_printf(Perl_debug_log,
2315 "Matching stclass %.*s against %s (%d bytes)\n",
2316 (int)SvCUR(prop), SvPVX_const(prop),
2317 quoted, (int)(strend - s));
2320 if (find_byclass(prog, c, s, strend, ®info))
2322 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2326 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2331 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2332 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2333 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2335 if (flags & REXEC_SCREAM) {
2336 last = screaminstr(sv, float_real, s - strbeg,
2337 end_shift, &scream_pos, 1); /* last one */
2339 last = scream_olds; /* Only one occurrence. */
2340 /* we may be pointing at the wrong string */
2341 else if (RXp_MATCH_COPIED(prog))
2342 s = strbeg + (s - SvPVX_const(sv));
2346 const char * const little = SvPV_const(float_real, len);
2348 if (SvTAIL(float_real)) {
2349 if (memEQ(strend - len + 1, little, len - 1))
2350 last = strend - len + 1;
2351 else if (!multiline)
2352 last = memEQ(strend - len, little, len)
2353 ? strend - len : NULL;
2359 last = rninstr(s, strend, little, little + len);
2361 last = strend; /* matching "$" */
2366 PerlIO_printf(Perl_debug_log,
2367 "%sCan't trim the tail, match fails (should not happen)%s\n",
2368 PL_colors[4], PL_colors[5]));
2369 goto phooey; /* Should not happen! */
2371 dontbother = strend - last + prog->float_min_offset;
2373 if (minlen && (dontbother < minlen))
2374 dontbother = minlen - 1;
2375 strend -= dontbother; /* this one's always in bytes! */
2376 /* We don't know much -- general case. */
2379 if (regtry(®info, &s))
2388 if (regtry(®info, &s))
2390 } while (s++ < strend);
2399 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2401 if (PL_reg_eval_set)
2402 restore_pos(aTHX_ prog);
2403 if (RXp_PAREN_NAMES(prog))
2404 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2406 /* make sure $`, $&, $', and $digit will work later */
2407 if ( !(flags & REXEC_NOT_FIRST) ) {
2408 RX_MATCH_COPY_FREE(rx);
2409 if (flags & REXEC_COPY_STR) {
2410 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2411 #ifdef PERL_OLD_COPY_ON_WRITE
2413 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2415 PerlIO_printf(Perl_debug_log,
2416 "Copy on write: regexp capture, type %d\n",
2419 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2420 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2421 assert (SvPOKp(prog->saved_copy));
2425 RX_MATCH_COPIED_on(rx);
2426 s = savepvn(strbeg, i);
2432 prog->subbeg = strbeg;
2433 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2440 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2441 PL_colors[4], PL_colors[5]));
2442 if (PL_reg_eval_set)
2443 restore_pos(aTHX_ prog);
2445 /* we failed :-( roll it back */
2446 Safefree(prog->offs);
2455 - regtry - try match at specific point
2457 STATIC I32 /* 0 failure, 1 success */
2458 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2462 REGEXP *const rx = reginfo->prog;
2463 regexp *const prog = (struct regexp *)SvANY(rx);
2464 RXi_GET_DECL(prog,progi);
2465 GET_RE_DEBUG_FLAGS_DECL;
2467 PERL_ARGS_ASSERT_REGTRY;
2469 reginfo->cutpoint=NULL;
2471 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2474 PL_reg_eval_set = RS_init;
2475 DEBUG_EXECUTE_r(DEBUG_s(
2476 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2477 (IV)(PL_stack_sp - PL_stack_base));
2480 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2481 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2483 /* Apparently this is not needed, judging by wantarray. */
2484 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2485 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2488 /* Make $_ available to executed code. */
2489 if (reginfo->sv != DEFSV) {
2491 DEFSV_set(reginfo->sv);
2494 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2495 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2496 /* prepare for quick setting of pos */
2497 #ifdef PERL_OLD_COPY_ON_WRITE
2498 if (SvIsCOW(reginfo->sv))
2499 sv_force_normal_flags(reginfo->sv, 0);
2501 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2502 &PL_vtbl_mglob, NULL, 0);
2506 PL_reg_oldpos = mg->mg_len;
2507 SAVEDESTRUCTOR_X(restore_pos, prog);
2509 if (!PL_reg_curpm) {
2510 Newxz(PL_reg_curpm, 1, PMOP);
2513 SV* const repointer = &PL_sv_undef;
2514 /* this regexp is also owned by the new PL_reg_curpm, which
2515 will try to free it. */
2516 av_push(PL_regex_padav, repointer);
2517 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2518 PL_regex_pad = AvARRAY(PL_regex_padav);
2523 /* It seems that non-ithreads works both with and without this code.
2524 So for efficiency reasons it seems best not to have the code
2525 compiled when it is not needed. */
2526 /* This is safe against NULLs: */
2527 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2528 /* PM_reg_curpm owns a reference to this regexp. */
2531 PM_SETRE(PL_reg_curpm, rx);
2532 PL_reg_oldcurpm = PL_curpm;
2533 PL_curpm = PL_reg_curpm;
2534 if (RXp_MATCH_COPIED(prog)) {
2535 /* Here is a serious problem: we cannot rewrite subbeg,
2536 since it may be needed if this match fails. Thus
2537 $` inside (?{}) could fail... */
2538 PL_reg_oldsaved = prog->subbeg;
2539 PL_reg_oldsavedlen = prog->sublen;
2540 #ifdef PERL_OLD_COPY_ON_WRITE
2541 PL_nrs = prog->saved_copy;
2543 RXp_MATCH_COPIED_off(prog);
2546 PL_reg_oldsaved = NULL;
2547 prog->subbeg = PL_bostr;
2548 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2550 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2551 prog->offs[0].start = *startpos - PL_bostr;
2552 PL_reginput = *startpos;
2553 PL_reglastparen = &prog->lastparen;
2554 PL_reglastcloseparen = &prog->lastcloseparen;
2555 prog->lastparen = 0;
2556 prog->lastcloseparen = 0;
2558 PL_regoffs = prog->offs;
2559 if (PL_reg_start_tmpl <= prog->nparens) {
2560 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2561 if(PL_reg_start_tmp)
2562 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2564 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2567 /* XXXX What this code is doing here?!!! There should be no need
2568 to do this again and again, PL_reglastparen should take care of
2571 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2572 * Actually, the code in regcppop() (which Ilya may be meaning by
2573 * PL_reglastparen), is not needed at all by the test suite
2574 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2575 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2576 * Meanwhile, this code *is* needed for the
2577 * above-mentioned test suite tests to succeed. The common theme
2578 * on those tests seems to be returning null fields from matches.
2579 * --jhi updated by dapm */
2581 if (prog->nparens) {
2582 regexp_paren_pair *pp = PL_regoffs;
2584 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2592 if (regmatch(reginfo, progi->program + 1)) {
2593 PL_regoffs[0].end = PL_reginput - PL_bostr;
2596 if (reginfo->cutpoint)
2597 *startpos= reginfo->cutpoint;
2598 REGCP_UNWIND(lastcp);
2603 #define sayYES goto yes
2604 #define sayNO goto no
2605 #define sayNO_SILENT goto no_silent
2607 /* we dont use STMT_START/END here because it leads to
2608 "unreachable code" warnings, which are bogus, but distracting. */
2609 #define CACHEsayNO \
2610 if (ST.cache_mask) \
2611 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2614 /* this is used to determine how far from the left messages like
2615 'failed...' are printed. It should be set such that messages
2616 are inline with the regop output that created them.
2618 #define REPORT_CODE_OFF 32
2621 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2622 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2624 #define SLAB_FIRST(s) (&(s)->states[0])
2625 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2627 /* grab a new slab and return the first slot in it */
2629 STATIC regmatch_state *
2632 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2635 regmatch_slab *s = PL_regmatch_slab->next;
2637 Newx(s, 1, regmatch_slab);
2638 s->prev = PL_regmatch_slab;
2640 PL_regmatch_slab->next = s;
2642 PL_regmatch_slab = s;
2643 return SLAB_FIRST(s);
2647 /* push a new state then goto it */
2649 #define PUSH_STATE_GOTO(state, node) \
2651 st->resume_state = state; \
2654 /* push a new state with success backtracking, then goto it */
2656 #define PUSH_YES_STATE_GOTO(state, node) \
2658 st->resume_state = state; \
2659 goto push_yes_state;
2665 regmatch() - main matching routine
2667 This is basically one big switch statement in a loop. We execute an op,
2668 set 'next' to point the next op, and continue. If we come to a point which
2669 we may need to backtrack to on failure such as (A|B|C), we push a
2670 backtrack state onto the backtrack stack. On failure, we pop the top
2671 state, and re-enter the loop at the state indicated. If there are no more
2672 states to pop, we return failure.
2674 Sometimes we also need to backtrack on success; for example /A+/, where
2675 after successfully matching one A, we need to go back and try to
2676 match another one; similarly for lookahead assertions: if the assertion
2677 completes successfully, we backtrack to the state just before the assertion
2678 and then carry on. In these cases, the pushed state is marked as
2679 'backtrack on success too'. This marking is in fact done by a chain of
2680 pointers, each pointing to the previous 'yes' state. On success, we pop to
2681 the nearest yes state, discarding any intermediate failure-only states.
2682 Sometimes a yes state is pushed just to force some cleanup code to be
2683 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2684 it to free the inner regex.
2686 Note that failure backtracking rewinds the cursor position, while
2687 success backtracking leaves it alone.
2689 A pattern is complete when the END op is executed, while a subpattern
2690 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2691 ops trigger the "pop to last yes state if any, otherwise return true"
2694 A common convention in this function is to use A and B to refer to the two
2695 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2696 the subpattern to be matched possibly multiple times, while B is the entire
2697 rest of the pattern. Variable and state names reflect this convention.
2699 The states in the main switch are the union of ops and failure/success of
2700 substates associated with with that op. For example, IFMATCH is the op
2701 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2702 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2703 successfully matched A and IFMATCH_A_fail is a state saying that we have
2704 just failed to match A. Resume states always come in pairs. The backtrack
2705 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2706 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2707 on success or failure.
2709 The struct that holds a backtracking state is actually a big union, with
2710 one variant for each major type of op. The variable st points to the
2711 top-most backtrack struct. To make the code clearer, within each
2712 block of code we #define ST to alias the relevant union.
2714 Here's a concrete example of a (vastly oversimplified) IFMATCH
2720 #define ST st->u.ifmatch
2722 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2723 ST.foo = ...; // some state we wish to save
2725 // push a yes backtrack state with a resume value of
2726 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2728 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2731 case IFMATCH_A: // we have successfully executed A; now continue with B
2733 bar = ST.foo; // do something with the preserved value
2736 case IFMATCH_A_fail: // A failed, so the assertion failed
2737 ...; // do some housekeeping, then ...
2738 sayNO; // propagate the failure
2745 For any old-timers reading this who are familiar with the old recursive
2746 approach, the code above is equivalent to:
2748 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2757 ...; // do some housekeeping, then ...
2758 sayNO; // propagate the failure
2761 The topmost backtrack state, pointed to by st, is usually free. If you
2762 want to claim it, populate any ST.foo fields in it with values you wish to
2763 save, then do one of
2765 PUSH_STATE_GOTO(resume_state, node);
2766 PUSH_YES_STATE_GOTO(resume_state, node);
2768 which sets that backtrack state's resume value to 'resume_state', pushes a
2769 new free entry to the top of the backtrack stack, then goes to 'node'.
2770 On backtracking, the free slot is popped, and the saved state becomes the
2771 new free state. An ST.foo field in this new top state can be temporarily
2772 accessed to retrieve values, but once the main loop is re-entered, it
2773 becomes available for reuse.
2775 Note that the depth of the backtrack stack constantly increases during the
2776 left-to-right execution of the pattern, rather than going up and down with
2777 the pattern nesting. For example the stack is at its maximum at Z at the
2778 end of the pattern, rather than at X in the following:
2780 /(((X)+)+)+....(Y)+....Z/
2782 The only exceptions to this are lookahead/behind assertions and the cut,
2783 (?>A), which pop all the backtrack states associated with A before
2786 Bascktrack state structs are allocated in slabs of about 4K in size.
2787 PL_regmatch_state and st always point to the currently active state,
2788 and PL_regmatch_slab points to the slab currently containing
2789 PL_regmatch_state. The first time regmatch() is called, the first slab is
2790 allocated, and is never freed until interpreter destruction. When the slab
2791 is full, a new one is allocated and chained to the end. At exit from
2792 regmatch(), slabs allocated since entry are freed.
2797 #define DEBUG_STATE_pp(pp) \
2799 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2800 PerlIO_printf(Perl_debug_log, \
2801 " %*s"pp" %s%s%s%s%s\n", \
2803 PL_reg_name[st->resume_state], \
2804 ((st==yes_state||st==mark_state) ? "[" : ""), \
2805 ((st==yes_state) ? "Y" : ""), \
2806 ((st==mark_state) ? "M" : ""), \
2807 ((st==yes_state||st==mark_state) ? "]" : "") \
2812 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2817 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2818 const char *start, const char *end, const char *blurb)
2820 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2822 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2827 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2828 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2830 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2831 start, end - start, 60);
2833 PerlIO_printf(Perl_debug_log,
2834 "%s%s REx%s %s against %s\n",
2835 PL_colors[4], blurb, PL_colors[5], s0, s1);
2837 if (utf8_target||utf8_pat)
2838 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2839 utf8_pat ? "pattern" : "",
2840 utf8_pat && utf8_target ? " and " : "",
2841 utf8_target ? "string" : ""
2847 S_dump_exec_pos(pTHX_ const char *locinput,
2848 const regnode *scan,
2849 const char *loc_regeol,
2850 const char *loc_bostr,
2851 const char *loc_reg_starttry,
2852 const bool utf8_target)
2854 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2855 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2856 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2857 /* The part of the string before starttry has one color
2858 (pref0_len chars), between starttry and current
2859 position another one (pref_len - pref0_len chars),
2860 after the current position the third one.
2861 We assume that pref0_len <= pref_len, otherwise we
2862 decrease pref0_len. */
2863 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2864 ? (5 + taill) - l : locinput - loc_bostr;
2867 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2869 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2871 pref0_len = pref_len - (locinput - loc_reg_starttry);
2872 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2873 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2874 ? (5 + taill) - pref_len : loc_regeol - locinput);
2875 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2879 if (pref0_len > pref_len)
2880 pref0_len = pref_len;
2882 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2884 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2885 (locinput - pref_len),pref0_len, 60, 4, 5);
2887 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2888 (locinput - pref_len + pref0_len),
2889 pref_len - pref0_len, 60, 2, 3);
2891 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2892 locinput, loc_regeol - locinput, 10, 0, 1);
2894 const STRLEN tlen=len0+len1+len2;
2895 PerlIO_printf(Perl_debug_log,
2896 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2897 (IV)(locinput - loc_bostr),
2900 (docolor ? "" : "> <"),
2902 (int)(tlen > 19 ? 0 : 19 - tlen),
2909 /* reg_check_named_buff_matched()
2910 * Checks to see if a named buffer has matched. The data array of
2911 * buffer numbers corresponding to the buffer is expected to reside
2912 * in the regexp->data->data array in the slot stored in the ARG() of
2913 * node involved. Note that this routine doesn't actually care about the
2914 * name, that information is not preserved from compilation to execution.
2915 * Returns the index of the leftmost defined buffer with the given name
2916 * or 0 if non of the buffers matched.
2919 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2922 RXi_GET_DECL(rex,rexi);
2923 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2924 I32 *nums=(I32*)SvPVX(sv_dat);
2926 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2928 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2929 if ((I32)*PL_reglastparen >= nums[n] &&
2930 PL_regoffs[nums[n]].end != -1)
2939 /* free all slabs above current one - called during LEAVE_SCOPE */
2942 S_clear_backtrack_stack(pTHX_ void *p)
2944 regmatch_slab *s = PL_regmatch_slab->next;
2949 PL_regmatch_slab->next = NULL;
2951 regmatch_slab * const osl = s;
2958 #define SETREX(Re1,Re2) \
2959 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2962 STATIC I32 /* 0 failure, 1 success */
2963 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2965 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2969 register const bool utf8_target = PL_reg_match_utf8;
2970 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2971 REGEXP *rex_sv = reginfo->prog;
2972 regexp *rex = (struct regexp *)SvANY(rex_sv);
2973 RXi_GET_DECL(rex,rexi);
2975 /* the current state. This is a cached copy of PL_regmatch_state */
2976 register regmatch_state *st;
2977 /* cache heavy used fields of st in registers */
2978 register regnode *scan;
2979 register regnode *next;
2980 register U32 n = 0; /* general value; init to avoid compiler warning */
2981 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2982 register char *locinput = PL_reginput;
2983 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2985 bool result = 0; /* return value of S_regmatch */
2986 int depth = 0; /* depth of backtrack stack */
2987 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2988 const U32 max_nochange_depth =
2989 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2990 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2991 regmatch_state *yes_state = NULL; /* state to pop to on success of
2993 /* mark_state piggy backs on the yes_state logic so that when we unwind
2994 the stack on success we can update the mark_state as we go */
2995 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2996 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2997 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2999 bool no_final = 0; /* prevent failure from backtracking? */
3000 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3001 char *startpoint = PL_reginput;
3002 SV *popmark = NULL; /* are we looking for a mark? */
3003 SV *sv_commit = NULL; /* last mark name seen in failure */
3004 SV *sv_yes_mark = NULL; /* last mark name we have seen
3005 during a successfull match */
3006 U32 lastopen = 0; /* last open we saw */
3007 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3008 SV* const oreplsv = GvSV(PL_replgv);
3009 /* these three flags are set by various ops to signal information to
3010 * the very next op. They have a useful lifetime of exactly one loop
3011 * iteration, and are not preserved or restored by state pushes/pops
3013 bool sw = 0; /* the condition value in (?(cond)a|b) */
3014 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3015 int logical = 0; /* the following EVAL is:
3019 or the following IFMATCH/UNLESSM is:
3020 false: plain (?=foo)
3021 true: used as a condition: (?(?=foo))
3024 GET_RE_DEBUG_FLAGS_DECL;
3027 PERL_ARGS_ASSERT_REGMATCH;
3029 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3030 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3032 /* on first ever call to regmatch, allocate first slab */
3033 if (!PL_regmatch_slab) {
3034 Newx(PL_regmatch_slab, 1, regmatch_slab);
3035 PL_regmatch_slab->prev = NULL;
3036 PL_regmatch_slab->next = NULL;
3037 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3040 oldsave = PL_savestack_ix;
3041 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3042 SAVEVPTR(PL_regmatch_slab);
3043 SAVEVPTR(PL_regmatch_state);
3045 /* grab next free state slot */
3046 st = ++PL_regmatch_state;
3047 if (st > SLAB_LAST(PL_regmatch_slab))
3048 st = PL_regmatch_state = S_push_slab(aTHX);
3050 /* Note that nextchr is a byte even in UTF */
3051 nextchr = UCHARAT(locinput);
3053 while (scan != NULL) {
3056 SV * const prop = sv_newmortal();
3057 regnode *rnext=regnext(scan);
3058 DUMP_EXEC_POS( locinput, scan, utf8_target );
3059 regprop(rex, prop, scan);
3061 PerlIO_printf(Perl_debug_log,
3062 "%3"IVdf":%*s%s(%"IVdf")\n",
3063 (IV)(scan - rexi->program), depth*2, "",
3065 (PL_regkind[OP(scan)] == END || !rnext) ?
3066 0 : (IV)(rnext - rexi->program));
3069 next = scan + NEXT_OFF(scan);
3072 state_num = OP(scan);
3076 assert(PL_reglastparen == &rex->lastparen);
3077 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3078 assert(PL_regoffs == rex->offs);
3080 switch (state_num) {
3082 if (locinput == PL_bostr)
3084 /* reginfo->till = reginfo->bol; */
3089 if (locinput == PL_bostr ||
3090 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3096 if (locinput == PL_bostr)
3100 if (locinput == reginfo->ganch)
3105 /* update the startpoint */
3106 st->u.keeper.val = PL_regoffs[0].start;
3107 PL_reginput = locinput;
3108 PL_regoffs[0].start = locinput - PL_bostr;
3109 PUSH_STATE_GOTO(KEEPS_next, next);
3111 case KEEPS_next_fail:
3112 /* rollback the start point change */
3113 PL_regoffs[0].start = st->u.keeper.val;
3119 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3124 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3126 if (PL_regeol - locinput > 1)
3130 if (PL_regeol != locinput)
3134 if (!nextchr && locinput >= PL_regeol)
3137 locinput += PL_utf8skip[nextchr];
3138 if (locinput > PL_regeol)
3140 nextchr = UCHARAT(locinput);
3143 nextchr = UCHARAT(++locinput);
3146 if (!nextchr && locinput >= PL_regeol)
3148 nextchr = UCHARAT(++locinput);
3151 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3154 locinput += PL_utf8skip[nextchr];
3155 if (locinput > PL_regeol)
3157 nextchr = UCHARAT(locinput);
3160 nextchr = UCHARAT(++locinput);
3164 #define ST st->u.trie
3166 /* In this case the charclass data is available inline so
3167 we can fail fast without a lot of extra overhead.
3169 if (scan->flags == EXACT || !utf8_target) {
3170 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3172 PerlIO_printf(Perl_debug_log,
3173 "%*s %sfailed to match trie start class...%s\n",
3174 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3182 /* the basic plan of execution of the trie is:
3183 * At the beginning, run though all the states, and
3184 * find the longest-matching word. Also remember the position
3185 * of the shortest matching word. For example, this pattern:
3188 * when matched against the string "abcde", will generate
3189 * accept states for all words except 3, with the longest
3190 * matching word being 4, and the shortest being 1 (with
3191 * the position being after char 1 of the string).
3193 * Then for each matching word, in word order (i.e. 1,2,4,5),
3194 * we run the remainder of the pattern; on each try setting
3195 * the current position to the character following the word,
3196 * returning to try the next word on failure.
3198 * We avoid having to build a list of words at runtime by
3199 * using a compile-time structure, wordinfo[].prev, which
3200 * gives, for each word, the previous accepting word (if any).
3201 * In the case above it would contain the mappings 1->2, 2->0,
3202 * 3->0, 4->5, 5->1. We can use this table to generate, from
3203 * the longest word (4 above), a list of all words, by
3204 * following the list of prev pointers; this gives us the
3205 * unordered list 4,5,1,2. Then given the current word we have
3206 * just tried, we can go through the list and find the
3207 * next-biggest word to try (so if we just failed on word 2,
3208 * the next in the list is 4).
3210 * Since at runtime we don't record the matching position in
3211 * the string for each word, we have to work that out for
3212 * each word we're about to process. The wordinfo table holds
3213 * the character length of each word; given that we recorded
3214 * at the start: the position of the shortest word and its
3215 * length in chars, we just need to move the pointer the
3216 * difference between the two char lengths. Depending on
3217 * Unicode status and folding, that's cheap or expensive.
3219 * This algorithm is optimised for the case where are only a
3220 * small number of accept states, i.e. 0,1, or maybe 2.
3221 * With lots of accepts states, and having to try all of them,
3222 * it becomes quadratic on number of accept states to find all
3227 /* what type of TRIE am I? (utf8 makes this contextual) */
3228 DECL_TRIE_TYPE(scan);
3230 /* what trie are we using right now */
3231 reg_trie_data * const trie
3232 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3233 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3234 U32 state = trie->startstate;
3236 if (trie->bitmap && trie_type != trie_utf8_fold &&
3237 !TRIE_BITMAP_TEST(trie,*locinput)
3239 if (trie->states[ state ].wordnum) {
3241 PerlIO_printf(Perl_debug_log,
3242 "%*s %smatched empty string...%s\n",
3243 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3249 PerlIO_printf(Perl_debug_log,
3250 "%*s %sfailed to match trie start class...%s\n",
3251 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3258 U8 *uc = ( U8* )locinput;
3262 U8 *uscan = (U8*)NULL;
3263 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3264 U32 charcount = 0; /* how many input chars we have matched */
3265 U32 accepted = 0; /* have we seen any accepting states? */
3268 ST.jump = trie->jump;
3271 ST.longfold = FALSE; /* char longer if folded => it's harder */
3274 /* fully traverse the TRIE; note the position of the
3275 shortest accept state and the wordnum of the longest
3278 while ( state && uc <= (U8*)PL_regeol ) {
3279 U32 base = trie->states[ state ].trans.base;
3283 wordnum = trie->states[ state ].wordnum;
3285 if (wordnum) { /* it's an accept state */
3288 /* record first match position */
3290 ST.firstpos = (U8*)locinput;
3295 ST.firstchars = charcount;
3298 if (!ST.nextword || wordnum < ST.nextword)
3299 ST.nextword = wordnum;
3300 ST.topword = wordnum;
3303 DEBUG_TRIE_EXECUTE_r({
3304 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3305 PerlIO_printf( Perl_debug_log,
3306 "%*s %sState: %4"UVxf" Accepted: %c ",
3307 2+depth * 2, "", PL_colors[4],
3308 (UV)state, (accepted ? 'Y' : 'N'));
3311 /* read a char and goto next state */
3314 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3315 uscan, len, uvc, charid, foldlen,
3322 base + charid - 1 - trie->uniquecharcount)) >= 0)
3324 && ((U32)offset < trie->lasttrans)
3325 && trie->trans[offset].check == state)
3327 state = trie->trans[offset].next;
3338 DEBUG_TRIE_EXECUTE_r(
3339 PerlIO_printf( Perl_debug_log,
3340 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3341 charid, uvc, (UV)state, PL_colors[5] );
3347 /* calculate total number of accept states */
3352 w = trie->wordinfo[w].prev;
3355 ST.accepted = accepted;
3359 PerlIO_printf( Perl_debug_log,
3360 "%*s %sgot %"IVdf" possible matches%s\n",
3361 REPORT_CODE_OFF + depth * 2, "",
3362 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3364 goto trie_first_try; /* jump into the fail handler */
3368 case TRIE_next_fail: /* we failed - try next alternative */
3370 REGCP_UNWIND(ST.cp);
3371 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3372 PL_regoffs[n].end = -1;
3373 *PL_reglastparen = n;
3375 if (!--ST.accepted) {
3377 PerlIO_printf( Perl_debug_log,
3378 "%*s %sTRIE failed...%s\n",
3379 REPORT_CODE_OFF+depth*2, "",
3386 /* Find next-highest word to process. Note that this code
3387 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3388 register U16 min = 0;
3390 register U16 const nextword = ST.nextword;
3391 register reg_trie_wordinfo * const wordinfo
3392 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3393 for (word=ST.topword; word; word=wordinfo[word].prev) {
3394 if (word > nextword && (!min || word < min))
3407 ST.lastparen = *PL_reglastparen;
3411 /* find start char of end of current word */
3413 U32 chars; /* how many chars to skip */
3414 U8 *uc = ST.firstpos;
3415 reg_trie_data * const trie
3416 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3418 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3420 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3424 /* the hard option - fold each char in turn and find
3425 * its folded length (which may be different */
3426 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3434 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3442 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3447 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3461 PL_reginput = (char *)uc;
3464 scan = (ST.jump && ST.jump[ST.nextword])
3465 ? ST.me + ST.jump[ST.nextword]
3469 PerlIO_printf( Perl_debug_log,
3470 "%*s %sTRIE matched word #%d, continuing%s\n",
3471 REPORT_CODE_OFF+depth*2, "",
3478 if (ST.accepted > 1 || has_cutgroup) {
3479 PUSH_STATE_GOTO(TRIE_next, scan);
3482 /* only one choice left - just continue */
3484 AV *const trie_words
3485 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3486 SV ** const tmp = av_fetch( trie_words,
3488 SV *sv= tmp ? sv_newmortal() : NULL;
3490 PerlIO_printf( Perl_debug_log,
3491 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3492 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3494 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3495 PL_colors[0], PL_colors[1],
3496 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3498 : "not compiled under -Dr",
3502 locinput = PL_reginput;
3503 nextchr = UCHARAT(locinput);
3504 continue; /* execute rest of RE */
3509 char *s = STRING(scan);
3511 if (utf8_target != UTF_PATTERN) {
3512 /* The target and the pattern have differing utf8ness. */
3514 const char * const e = s + ln;
3517 /* The target is utf8, the pattern is not utf8. */
3522 if (NATIVE_TO_UNI(*(U8*)s) !=
3523 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3531 /* The target is not utf8, the pattern is utf8. */
3536 if (NATIVE_TO_UNI(*((U8*)l)) !=
3537 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3545 nextchr = UCHARAT(locinput);
3548 /* The target and the pattern have the same utf8ness. */
3549 /* Inline the first character, for speed. */
3550 if (UCHARAT(s) != nextchr)
3552 if (PL_regeol - locinput < ln)
3554 if (ln > 1 && memNE(s, locinput, ln))
3557 nextchr = UCHARAT(locinput);
3562 const U8 * fold_array;
3565 PL_reg_flags |= RF_tainted;
3566 folder = foldEQ_locale;
3567 fold_array = PL_fold_locale;
3571 folder = foldEQ_latin1;
3572 fold_array = PL_fold_latin1;
3577 fold_array = PL_fold;
3583 if (utf8_target || UTF_PATTERN) {
3584 /* Either target or the pattern are utf8. */
3585 const char * const l = locinput;
3586 char *e = PL_regeol;
3588 if (! foldEQ_utf8(s, 0, ln, cBOOL(UTF_PATTERN),
3589 l, &e, 0, utf8_target)) {
3590 /* One more case for the sharp s:
3591 * pack("U0U*", 0xDF) =~ /ss/i,
3592 * the 0xC3 0x9F are the UTF-8
3593 * byte sequence for the U+00DF. */
3595 if (!(utf8_target &&
3596 toLOWER(s[0]) == 's' &&
3598 toLOWER(s[1]) == 's' &&
3605 nextchr = UCHARAT(locinput);
3609 /* Neither the target and the pattern are utf8. */
3611 /* Inline the first character, for speed. */
3612 if (UCHARAT(s) != nextchr &&
3613 UCHARAT(s) != fold_array[nextchr])
3617 if (PL_regeol - locinput < ln)
3619 if (ln > 1 && ! folder(s, locinput, ln))
3622 nextchr = UCHARAT(locinput);
3627 PL_reg_flags |= RF_tainted;
3631 /* was last char in word? */
3633 if (locinput == PL_bostr)
3636 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3638 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3640 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3641 ln = isALNUM_uni(ln);
3642 LOAD_UTF8_CHARCLASS_ALNUM();
3643 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3646 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3647 n = isALNUM_LC_utf8((U8*)locinput);
3651 ln = (locinput != PL_bostr) ?
3652 UCHARAT(locinput - 1) : '\n';
3653 if (FLAGS(scan) & USE_UNI) {
3655 /* Here, can't be BOUNDL or NBOUNDL because they never set
3656 * the flags to USE_UNI */
3657 ln = isWORDCHAR_L1(ln);
3658 n = isWORDCHAR_L1(nextchr);
3660 else if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3662 n = isALNUM(nextchr);
3665 ln = isALNUM_LC(ln);
3666 n = isALNUM_LC(nextchr);
3669 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3670 OP(scan) == BOUNDL))
3675 STRLEN inclasslen = PL_regeol - locinput;
3676 if (locinput >= PL_regeol)
3679 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3681 locinput += inclasslen;
3682 nextchr = UCHARAT(locinput);
3687 nextchr = UCHARAT(locinput);
3688 if (!nextchr && locinput >= PL_regeol)
3690 if (!REGINCLASS(rex, scan, (U8*)locinput))
3692 nextchr = UCHARAT(++locinput);
3696 /* If we might have the case of the German sharp s
3697 * in a casefolding Unicode character class. */
3699 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3700 locinput += SHARP_S_SKIP;
3701 nextchr = UCHARAT(locinput);
3706 /* Special char classes - The defines start on line 129 or so */
3707 CCC_TRY_AFF_U( ALNUM, ALNUML, perl_word, "a", isALNUM_LC_utf8, isWORDCHAR_L1, isALNUM_LC);
3708 CCC_TRY_NEG_U(NALNUM, NALNUML, perl_word, "a", isALNUM_LC_utf8, isWORDCHAR_L1, isALNUM_LC);
3710 CCC_TRY_AFF_U( SPACE, SPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE_L1, isSPACE_LC);
3711 CCC_TRY_NEG_U(NSPACE, NSPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE_L1, isSPACE_LC);
3713 CCC_TRY_AFF( DIGIT, DIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3714 CCC_TRY_NEG(NDIGIT, NDIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3716 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3717 a Unicode extended Grapheme Cluster */
3718 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3719 extended Grapheme Cluster is:
3722 | Prepend* Begin Extend*
3725 Begin is (Hangul-syllable | ! Control)
3726 Extend is (Grapheme_Extend | Spacing_Mark)
3727 Control is [ GCB_Control CR LF ]
3729 The discussion below shows how the code for CLUMP is derived
3730 from this regex. Note that most of these concepts are from
3731 property values of the Grapheme Cluster Boundary (GCB) property.
3732 No code point can have multiple property values for a given
3733 property. Thus a code point in Prepend can't be in Control, but
3734 it must be in !Control. This is why Control above includes
3735 GCB_Control plus CR plus LF. The latter two are used in the GCB
3736 property separately, and so can't be in GCB_Control, even though
3737 they logically are controls. Control is not the same as gc=cc,
3738 but includes format and other characters as well.
3740 The Unicode definition of Hangul-syllable is:
3742 | (L* ( ( V | LV ) V* | LVT ) T*)
3745 Each of these is a value for the GCB property, and hence must be
3746 disjoint, so the order they are tested is immaterial, so the
3747 above can safely be changed to
3750 | (L* ( LVT | ( V | LV ) V*) T*)
3752 The last two terms can be combined like this:
3754 | (( LVT | ( V | LV ) V*) T*))
3756 And refactored into this:
3757 L* (L | LVT T* | V V* T* | LV V* T*)
3759 That means that if we have seen any L's at all we can quit
3760 there, but if the next character is a LVT, a V or and LV we
3763 There is a subtlety with Prepend* which showed up in testing.
3764 Note that the Begin, and only the Begin is required in:
3765 | Prepend* Begin Extend*
3766 Also, Begin contains '! Control'. A Prepend must be a '!
3767 Control', which means it must be a Begin. What it comes down to
3768 is that if we match Prepend* and then find no suitable Begin
3769 afterwards, that if we backtrack the last Prepend, that one will
3770 be a suitable Begin.
3773 if (locinput >= PL_regeol)
3775 if (! utf8_target) {
3777 /* Match either CR LF or '.', as all the other possibilities
3779 locinput++; /* Match the . or CR */
3781 && locinput < PL_regeol
3782 && UCHARAT(locinput) == '\n') locinput++;
3786 /* Utf8: See if is ( CR LF ); already know that locinput <
3787 * PL_regeol, so locinput+1 is in bounds */
3788 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3792 /* In case have to backtrack to beginning, then match '.' */
3793 char *starting = locinput;
3795 /* In case have to backtrack the last prepend */
3796 char *previous_prepend = 0;
3798 LOAD_UTF8_CHARCLASS_GCB();
3800 /* Match (prepend)* */
3801 while (locinput < PL_regeol
3802 && swash_fetch(PL_utf8_X_prepend,
3803 (U8*)locinput, utf8_target))
3805 previous_prepend = locinput;
3806 locinput += UTF8SKIP(locinput);
3809 /* As noted above, if we matched a prepend character, but
3810 * the next thing won't match, back off the last prepend we
3811 * matched, as it is guaranteed to match the begin */
3812 if (previous_prepend
3813 && (locinput >= PL_regeol
3814 || ! swash_fetch(PL_utf8_X_begin,
3815 (U8*)locinput, utf8_target)))
3817 locinput = previous_prepend;
3820 /* Note that here we know PL_regeol > locinput, as we
3821 * tested that upon input to this switch case, and if we
3822 * moved locinput forward, we tested the result just above
3823 * and it either passed, or we backed off so that it will
3825 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3827 /* Here did not match the required 'Begin' in the
3828 * second term. So just match the very first
3829 * character, the '.' of the final term of the regex */
3830 locinput = starting + UTF8SKIP(starting);
3833 /* Here is the beginning of a character that can have
3834 * an extender. It is either a hangul syllable, or a
3836 if (swash_fetch(PL_utf8_X_non_hangul,
3837 (U8*)locinput, utf8_target))
3840 /* Here not a Hangul syllable, must be a
3841 * ('! * Control') */
3842 locinput += UTF8SKIP(locinput);
3845 /* Here is a Hangul syllable. It can be composed
3846 * of several individual characters. One
3847 * possibility is T+ */
3848 if (swash_fetch(PL_utf8_X_T,
3849 (U8*)locinput, utf8_target))
3851 while (locinput < PL_regeol