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? e.g. locale */
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, non-ANYOFV nodes only: avoids the reginclass
98 * call if there are no complications: i.e., if everything matchable is
99 * straight forward in the bitmap */
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 */
126 /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just
127 * skip the check on EBCDIC platforms */
128 # define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class)
130 # define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
131 if (!CAT2(PL_utf8_,class)) { \
133 ENTER; save_re_context(); \
134 ok=CAT2(is_utf8_,class)((const U8*)str); \
135 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
138 /* Doesn't do an assert to verify that is correct */
139 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
140 if (!CAT2(PL_utf8_,class)) { \
141 bool throw_away PERL_UNUSED_DECL; \
142 ENTER; save_re_context(); \
143 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
146 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
147 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
148 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
150 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
151 LOAD_UTF8_CHARCLASS(X_begin, " "); \
152 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
153 /* These are utf8 constants, and not utf-ebcdic constants, so the \
154 * assert should likely and hopefully fail on an EBCDIC machine */ \
155 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
157 /* No asserts are done for these, in case called on an early \
158 * Unicode version in which they map to nothing */ \
159 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
160 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
161 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
162 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
163 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
164 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
165 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
167 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
169 /* The actual code for CCC_TRY, which uses several variables from the routine
170 * it's callable from. It is designed to be the bulk of a case statement.
171 * FUNC is the macro or function to call on non-utf8 targets that indicate if
172 * nextchr matches the class.
173 * UTF8_TEST is the whole test string to use for utf8 targets
174 * LOAD is what to use to test, and if not present to load in the swash for the
176 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
178 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
179 * utf8 and a variant, load the swash if necessary and test using the utf8
180 * test. Advance to the next character if test is ok, otherwise fail; If not
181 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
182 * fails, or advance to the next character */
184 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
185 if (locinput >= PL_regeol) { \
188 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
189 LOAD_UTF8_CHARCLASS(CLASS, STR); \
190 if (POS_OR_NEG (UTF8_TEST)) { \
193 locinput += PL_utf8skip[nextchr]; \
194 nextchr = UCHARAT(locinput); \
197 if (POS_OR_NEG (FUNC(nextchr))) { \
200 nextchr = UCHARAT(++locinput); \
203 /* Handle the non-locale cases for a character class and its complement. It
204 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
205 * This is because that code fails when the test succeeds, so we want to have
206 * the test fail so that the code succeeds. The swash is stored in a
207 * predictable PL_ place */
208 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
211 _CCC_TRY_CODE( !, FUNC, \
212 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
213 (U8*)locinput, TRUE)), \
216 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
217 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
218 (U8*)locinput, TRUE)), \
221 /* Generate the case statements for both locale and non-locale character
222 * classes in regmatch for classes that don't have special unicode semantics.
223 * Locales don't use an immediate swash, but an intermediary special locale
224 * function that is called on the pointer to the current place in the input
225 * string. That function will resolve to needing the same swash. One might
226 * think that because we don't know what the locale will match, we shouldn't
227 * check with the swash loading function that it loaded properly; ie, that we
228 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
229 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
231 #define CCC_TRY(NAME, NNAME, FUNC, \
232 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
233 NAMEA, NNAMEA, FUNCA, \
236 PL_reg_flags |= RF_tainted; \
237 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
239 PL_reg_flags |= RF_tainted; \
240 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
243 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
246 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
247 nextchr = UCHARAT(++locinput); \
250 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
254 locinput += PL_utf8skip[nextchr]; \
255 nextchr = UCHARAT(locinput); \
258 nextchr = UCHARAT(++locinput); \
261 /* Generate the non-locale cases */ \
262 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
264 /* This is like CCC_TRY, but has an extra set of parameters for generating case
265 * statements to handle separate Unicode semantics nodes */
266 #define CCC_TRY_U(NAME, NNAME, FUNC, \
267 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
268 NAMEU, NNAMEU, FUNCU, \
269 NAMEA, NNAMEA, FUNCA, \
271 CCC_TRY(NAME, NNAME, FUNC, \
272 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
273 NAMEA, NNAMEA, FUNCA, \
275 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
277 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
279 /* for use after a quantifier and before an EXACT-like node -- japhy */
280 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
282 * NOTE that *nothing* that affects backtracking should be in here, specifically
283 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
284 * node that is in between two EXACT like nodes when ascertaining what the required
285 * "follow" character is. This should probably be moved to regex compile time
286 * although it may be done at run time beause of the REF possibility - more
287 * investigation required. -- demerphq
289 #define JUMPABLE(rn) ( \
291 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
293 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
294 OP(rn) == PLUS || OP(rn) == MINMOD || \
296 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
298 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
300 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
303 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
304 we don't need this definition. */
305 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
306 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
307 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
310 /* ... so we use this as its faster. */
311 #define IS_TEXT(rn) ( OP(rn)==EXACT )
312 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
313 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
314 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
319 Search for mandatory following text node; for lookahead, the text must
320 follow but for lookbehind (rn->flags != 0) we skip to the next step.
322 #define FIND_NEXT_IMPT(rn) STMT_START { \
323 while (JUMPABLE(rn)) { \
324 const OPCODE type = OP(rn); \
325 if (type == SUSPEND || PL_regkind[type] == CURLY) \
326 rn = NEXTOPER(NEXTOPER(rn)); \
327 else if (type == PLUS) \
329 else if (type == IFMATCH) \
330 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
331 else rn += NEXT_OFF(rn); \
336 static void restore_pos(pTHX_ void *arg);
338 #define REGCP_PAREN_ELEMS 4
339 #define REGCP_OTHER_ELEMS 5
340 #define REGCP_FRAME_ELEMS 1
341 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
342 * are needed for the regexp context stack bookkeeping. */
345 S_regcppush(pTHX_ I32 parenfloor)
348 const int retval = PL_savestack_ix;
349 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
350 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
351 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
353 GET_RE_DEBUG_FLAGS_DECL;
355 if (paren_elems_to_push < 0)
356 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
357 paren_elems_to_push);
359 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
360 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
361 " out of range (%lu-%ld)",
362 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
364 SSGROW(total_elems + REGCP_FRAME_ELEMS);
366 for (p = PL_regsize; p > parenfloor; p--) {
367 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
368 SSPUSHINT(PL_regoffs[p].end);
369 SSPUSHINT(PL_regoffs[p].start);
370 SSPUSHPTR(PL_reg_start_tmp[p]);
372 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
373 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
374 (UV)p, (IV)PL_regoffs[p].start,
375 (IV)(PL_reg_start_tmp[p] - PL_bostr),
376 (IV)PL_regoffs[p].end
379 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
380 SSPUSHPTR(PL_regoffs);
381 SSPUSHINT(PL_regsize);
382 SSPUSHINT(*PL_reglastparen);
383 SSPUSHINT(*PL_reglastcloseparen);
384 SSPUSHPTR(PL_reginput);
385 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
390 /* These are needed since we do not localize EVAL nodes: */
391 #define REGCP_SET(cp) \
393 PerlIO_printf(Perl_debug_log, \
394 " Setting an EVAL scope, savestack=%"IVdf"\n", \
395 (IV)PL_savestack_ix)); \
398 #define REGCP_UNWIND(cp) \
400 if (cp != PL_savestack_ix) \
401 PerlIO_printf(Perl_debug_log, \
402 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
403 (IV)(cp), (IV)PL_savestack_ix)); \
407 S_regcppop(pTHX_ const regexp *rex)
412 GET_RE_DEBUG_FLAGS_DECL;
414 PERL_ARGS_ASSERT_REGCPPOP;
416 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
418 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
419 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
420 input = (char *) SSPOPPTR;
421 *PL_reglastcloseparen = SSPOPINT;
422 *PL_reglastparen = SSPOPINT;
423 PL_regsize = SSPOPINT;
424 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
426 i -= REGCP_OTHER_ELEMS;
427 /* Now restore the parentheses context. */
428 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
430 U32 paren = (U32)SSPOPINT;
431 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
432 PL_regoffs[paren].start = SSPOPINT;
434 if (paren <= *PL_reglastparen)
435 PL_regoffs[paren].end = tmps;
437 PerlIO_printf(Perl_debug_log,
438 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
439 (UV)paren, (IV)PL_regoffs[paren].start,
440 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
441 (IV)PL_regoffs[paren].end,
442 (paren > *PL_reglastparen ? "(no)" : ""));
446 if (*PL_reglastparen + 1 <= rex->nparens) {
447 PerlIO_printf(Perl_debug_log,
448 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
449 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
453 /* It would seem that the similar code in regtry()
454 * already takes care of this, and in fact it is in
455 * a better location to since this code can #if 0-ed out
456 * but the code in regtry() is needed or otherwise tests
457 * requiring null fields (pat.t#187 and split.t#{13,14}
458 * (as of patchlevel 7877) will fail. Then again,
459 * this code seems to be necessary or otherwise
460 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
461 * --jhi updated by dapm */
462 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
464 PL_regoffs[i].start = -1;
465 PL_regoffs[i].end = -1;
471 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
474 * pregexec and friends
477 #ifndef PERL_IN_XSUB_RE
479 - pregexec - match a regexp against a string
482 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
483 char *strbeg, I32 minend, SV *screamer, U32 nosave)
484 /* strend: pointer to null at end of string */
485 /* strbeg: real beginning of string */
486 /* minend: end of match must be >=minend after stringarg. */
487 /* nosave: For optimizations. */
489 PERL_ARGS_ASSERT_PREGEXEC;
492 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
493 nosave ? 0 : REXEC_COPY_STR);
498 * Need to implement the following flags for reg_anch:
500 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
502 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
503 * INTUIT_AUTORITATIVE_ML
504 * INTUIT_ONCE_NOML - Intuit can match in one location only.
507 * Another flag for this function: SECOND_TIME (so that float substrs
508 * with giant delta may be not rechecked).
511 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
513 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
514 Otherwise, only SvCUR(sv) is used to get strbeg. */
516 /* XXXX We assume that strpos is strbeg unless sv. */
518 /* XXXX Some places assume that there is a fixed substring.
519 An update may be needed if optimizer marks as "INTUITable"
520 RExen without fixed substrings. Similarly, it is assumed that
521 lengths of all the strings are no more than minlen, thus they
522 cannot come from lookahead.
523 (Or minlen should take into account lookahead.)
524 NOTE: Some of this comment is not correct. minlen does now take account
525 of lookahead/behind. Further research is required. -- demerphq
529 /* A failure to find a constant substring means that there is no need to make
530 an expensive call to REx engine, thus we celebrate a failure. Similarly,
531 finding a substring too deep into the string means that less calls to
532 regtry() should be needed.
534 REx compiler's optimizer found 4 possible hints:
535 a) Anchored substring;
537 c) Whether we are anchored (beginning-of-line or \G);
538 d) First node (of those at offset 0) which may distinguish positions;
539 We use a)b)d) and multiline-part of c), and try to find a position in the
540 string which does not contradict any of them.
543 /* Most of decisions we do here should have been done at compile time.
544 The nodes of the REx which we used for the search should have been
545 deleted from the finite automaton. */
548 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
549 char *strend, const U32 flags, re_scream_pos_data *data)
552 struct regexp *const prog = (struct regexp *)SvANY(rx);
553 register I32 start_shift = 0;
554 /* Should be nonnegative! */
555 register I32 end_shift = 0;
560 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
562 register char *other_last = NULL; /* other substr checked before this */
563 char *check_at = NULL; /* check substr found at this pos */
564 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
565 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
566 RXi_GET_DECL(prog,progi);
568 const char * const i_strpos = strpos;
570 GET_RE_DEBUG_FLAGS_DECL;
572 PERL_ARGS_ASSERT_RE_INTUIT_START;
573 PERL_UNUSED_ARG(flags);
574 PERL_UNUSED_ARG(data);
576 RX_MATCH_UTF8_set(rx,utf8_target);
579 PL_reg_flags |= RF_utf8;
582 debug_start_match(rx, utf8_target, strpos, strend,
583 sv ? "Guessing start of match in sv for"
584 : "Guessing start of match in string for");
587 /* CHR_DIST() would be more correct here but it makes things slow. */
588 if (prog->minlen > strend - strpos) {
589 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
590 "String too short... [re_intuit_start]\n"));
594 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
597 if (!prog->check_utf8 && prog->check_substr)
598 to_utf8_substr(prog);
599 check = prog->check_utf8;
601 if (!prog->check_substr && prog->check_utf8)
602 to_byte_substr(prog);
603 check = prog->check_substr;
605 if (check == &PL_sv_undef) {
606 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
607 "Non-utf8 string cannot match utf8 check string\n"));
610 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
611 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
612 || ( (prog->extflags & RXf_ANCH_BOL)
613 && !multiline ) ); /* Check after \n? */
616 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
617 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
618 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
620 && (strpos != strbeg)) {
621 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
624 if (prog->check_offset_min == prog->check_offset_max &&
625 !(prog->extflags & RXf_CANY_SEEN)) {
626 /* Substring at constant offset from beg-of-str... */
629 s = HOP3c(strpos, prog->check_offset_min, strend);
632 slen = SvCUR(check); /* >= 1 */
634 if ( strend - s > slen || strend - s < slen - 1
635 || (strend - s == slen && strend[-1] != '\n')) {
636 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
639 /* Now should match s[0..slen-2] */
641 if (slen && (*SvPVX_const(check) != *s
643 && memNE(SvPVX_const(check), s, slen)))) {
645 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
649 else if (*SvPVX_const(check) != *s
650 || ((slen = SvCUR(check)) > 1
651 && memNE(SvPVX_const(check), s, slen)))
654 goto success_at_start;
657 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
659 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
660 end_shift = prog->check_end_shift;
663 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
664 - (SvTAIL(check) != 0);
665 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
667 if (end_shift < eshift)
671 else { /* Can match at random position */
674 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
675 end_shift = prog->check_end_shift;
677 /* end shift should be non negative here */
680 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
682 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
683 (IV)end_shift, RX_PRECOMP(prog));
687 /* Find a possible match in the region s..strend by looking for
688 the "check" substring in the region corrected by start/end_shift. */
691 I32 srch_start_shift = start_shift;
692 I32 srch_end_shift = end_shift;
695 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
696 srch_end_shift -= ((strbeg - s) - srch_start_shift);
697 srch_start_shift = strbeg - s;
699 DEBUG_OPTIMISE_MORE_r({
700 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
701 (IV)prog->check_offset_min,
702 (IV)srch_start_shift,
704 (IV)prog->check_end_shift);
707 if (prog->extflags & RXf_CANY_SEEN) {
708 start_point= (U8*)(s + srch_start_shift);
709 end_point= (U8*)(strend - srch_end_shift);
711 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
712 end_point= HOP3(strend, -srch_end_shift, strbeg);
714 DEBUG_OPTIMISE_MORE_r({
715 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
716 (int)(end_point - start_point),
717 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
721 s = fbm_instr( start_point, end_point,
722 check, multiline ? FBMrf_MULTILINE : 0);
724 /* Update the count-of-usability, remove useless subpatterns,
728 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
729 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
730 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
731 (s ? "Found" : "Did not find"),
732 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
733 ? "anchored" : "floating"),
736 (s ? " at offset " : "...\n") );
741 /* Finish the diagnostic message */
742 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
744 /* XXX dmq: first branch is for positive lookbehind...
745 Our check string is offset from the beginning of the pattern.
746 So we need to do any stclass tests offset forward from that
755 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
756 Start with the other substr.
757 XXXX no SCREAM optimization yet - and a very coarse implementation
758 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
759 *always* match. Probably should be marked during compile...
760 Probably it is right to do no SCREAM here...
763 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
764 : (prog->float_substr && prog->anchored_substr))
766 /* Take into account the "other" substring. */
767 /* XXXX May be hopelessly wrong for UTF... */
770 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
773 char * const last = HOP3c(s, -start_shift, strbeg);
775 char * const saved_s = s;
778 t = s - prog->check_offset_max;
779 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
781 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
786 t = HOP3c(t, prog->anchored_offset, strend);
787 if (t < other_last) /* These positions already checked */
789 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
792 /* XXXX It is not documented what units *_offsets are in.
793 We assume bytes, but this is clearly wrong.
794 Meaning this code needs to be carefully reviewed for errors.
798 /* On end-of-str: see comment below. */
799 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
800 if (must == &PL_sv_undef) {
802 DEBUG_r(must = prog->anchored_utf8); /* for debug */
807 HOP3(HOP3(last1, prog->anchored_offset, strend)
808 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
810 multiline ? FBMrf_MULTILINE : 0
813 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
814 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
815 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
816 (s ? "Found" : "Contradicts"),
817 quoted, RE_SV_TAIL(must));
822 if (last1 >= last2) {
823 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
824 ", giving up...\n"));
827 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
828 ", trying floating at offset %ld...\n",
829 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
830 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
831 s = HOP3c(last, 1, strend);
835 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
836 (long)(s - i_strpos)));
837 t = HOP3c(s, -prog->anchored_offset, strbeg);
838 other_last = HOP3c(s, 1, strend);
846 else { /* Take into account the floating substring. */
848 char * const saved_s = s;
851 t = HOP3c(s, -start_shift, strbeg);
853 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
854 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
855 last = HOP3c(t, prog->float_max_offset, strend);
856 s = HOP3c(t, prog->float_min_offset, strend);
859 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
860 must = utf8_target ? prog->float_utf8 : prog->float_substr;
861 /* fbm_instr() takes into account exact value of end-of-str
862 if the check is SvTAIL(ed). Since false positives are OK,
863 and end-of-str is not later than strend we are OK. */
864 if (must == &PL_sv_undef) {
866 DEBUG_r(must = prog->float_utf8); /* for debug message */
869 s = fbm_instr((unsigned char*)s,
870 (unsigned char*)last + SvCUR(must)
872 must, multiline ? FBMrf_MULTILINE : 0);
874 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
875 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
876 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
877 (s ? "Found" : "Contradicts"),
878 quoted, RE_SV_TAIL(must));
882 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
883 ", giving up...\n"));
886 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
887 ", trying anchored starting at offset %ld...\n",
888 (long)(saved_s + 1 - i_strpos)));
890 s = HOP3c(t, 1, strend);
894 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
895 (long)(s - i_strpos)));
896 other_last = s; /* Fix this later. --Hugo */
906 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
908 DEBUG_OPTIMISE_MORE_r(
909 PerlIO_printf(Perl_debug_log,
910 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
911 (IV)prog->check_offset_min,
912 (IV)prog->check_offset_max,
920 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
922 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
925 /* Fixed substring is found far enough so that the match
926 cannot start at strpos. */
928 if (ml_anch && t[-1] != '\n') {
929 /* Eventually fbm_*() should handle this, but often
930 anchored_offset is not 0, so this check will not be wasted. */
931 /* XXXX In the code below we prefer to look for "^" even in
932 presence of anchored substrings. And we search even
933 beyond the found float position. These pessimizations
934 are historical artefacts only. */
936 while (t < strend - prog->minlen) {
938 if (t < check_at - prog->check_offset_min) {
939 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
940 /* Since we moved from the found position,
941 we definitely contradict the found anchored
942 substr. Due to the above check we do not
943 contradict "check" substr.
944 Thus we can arrive here only if check substr
945 is float. Redo checking for "other"=="fixed".
948 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
949 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
950 goto do_other_anchored;
952 /* We don't contradict the found floating substring. */
953 /* XXXX Why not check for STCLASS? */
955 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
956 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
959 /* Position contradicts check-string */
960 /* XXXX probably better to look for check-string
961 than for "\n", so one should lower the limit for t? */
962 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
963 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
964 other_last = strpos = s = t + 1;
969 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
970 PL_colors[0], PL_colors[1]));
974 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
975 PL_colors[0], PL_colors[1]));
979 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
982 /* The found string does not prohibit matching at strpos,
983 - no optimization of calling REx engine can be performed,
984 unless it was an MBOL and we are not after MBOL,
985 or a future STCLASS check will fail this. */
987 /* Even in this situation we may use MBOL flag if strpos is offset
988 wrt the start of the string. */
989 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
990 && (strpos != strbeg) && strpos[-1] != '\n'
991 /* May be due to an implicit anchor of m{.*foo} */
992 && !(prog->intflags & PREGf_IMPLICIT))
997 DEBUG_EXECUTE_r( if (ml_anch)
998 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
999 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1002 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1004 prog->check_utf8 /* Could be deleted already */
1005 && --BmUSEFUL(prog->check_utf8) < 0
1006 && (prog->check_utf8 == prog->float_utf8)
1008 prog->check_substr /* Could be deleted already */
1009 && --BmUSEFUL(prog->check_substr) < 0
1010 && (prog->check_substr == prog->float_substr)
1013 /* If flags & SOMETHING - do not do it many times on the same match */
1014 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1015 /* XXX Does the destruction order has to change with utf8_target? */
1016 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1017 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1018 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1019 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1020 check = NULL; /* abort */
1022 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1023 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1024 if (prog->intflags & PREGf_IMPLICIT)
1025 prog->extflags &= ~RXf_ANCH_MBOL;
1026 /* XXXX This is a remnant of the old implementation. It
1027 looks wasteful, since now INTUIT can use many
1028 other heuristics. */
1029 prog->extflags &= ~RXf_USE_INTUIT;
1030 /* XXXX What other flags might need to be cleared in this branch? */
1036 /* Last resort... */
1037 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1038 /* trie stclasses are too expensive to use here, we are better off to
1039 leave it to regmatch itself */
1040 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1041 /* minlen == 0 is possible if regstclass is \b or \B,
1042 and the fixed substr is ''$.
1043 Since minlen is already taken into account, s+1 is before strend;
1044 accidentally, minlen >= 1 guaranties no false positives at s + 1
1045 even for \b or \B. But (minlen? 1 : 0) below assumes that
1046 regstclass does not come from lookahead... */
1047 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1048 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1049 const U8* const str = (U8*)STRING(progi->regstclass);
1050 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1051 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1054 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1055 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1056 else if (prog->float_substr || prog->float_utf8)
1057 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1061 if (checked_upto < s)
1063 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1064 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1067 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1072 const char *what = NULL;
1074 if (endpos == strend) {
1075 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1076 "Could not match STCLASS...\n") );
1079 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1080 "This position contradicts STCLASS...\n") );
1081 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1083 checked_upto = HOPBACKc(endpos, start_shift);
1084 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1085 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1086 /* Contradict one of substrings */
1087 if (prog->anchored_substr || prog->anchored_utf8) {
1088 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1089 DEBUG_EXECUTE_r( what = "anchored" );
1091 s = HOP3c(t, 1, strend);
1092 if (s + start_shift + end_shift > strend) {
1093 /* XXXX Should be taken into account earlier? */
1094 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1095 "Could not match STCLASS...\n") );
1100 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1101 "Looking for %s substr starting at offset %ld...\n",
1102 what, (long)(s + start_shift - i_strpos)) );
1105 /* Have both, check_string is floating */
1106 if (t + start_shift >= check_at) /* Contradicts floating=check */
1107 goto retry_floating_check;
1108 /* Recheck anchored substring, but not floating... */
1112 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1113 "Looking for anchored substr starting at offset %ld...\n",
1114 (long)(other_last - i_strpos)) );
1115 goto do_other_anchored;
1117 /* Another way we could have checked stclass at the
1118 current position only: */
1123 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1124 "Looking for /%s^%s/m starting at offset %ld...\n",
1125 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1128 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1130 /* Check is floating substring. */
1131 retry_floating_check:
1132 t = check_at - start_shift;
1133 DEBUG_EXECUTE_r( what = "floating" );
1134 goto hop_and_restart;
1137 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1138 "By STCLASS: moving %ld --> %ld\n",
1139 (long)(t - i_strpos), (long)(s - i_strpos))
1143 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1144 "Does not contradict STCLASS...\n");
1149 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1150 PL_colors[4], (check ? "Guessed" : "Giving up"),
1151 PL_colors[5], (long)(s - i_strpos)) );
1154 fail_finish: /* Substring not found */
1155 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1156 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1158 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1159 PL_colors[4], PL_colors[5]));
1163 #define DECL_TRIE_TYPE(scan) \
1164 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1165 trie_type = ((scan->flags == EXACT) \
1166 ? (utf8_target ? trie_utf8 : trie_plain) \
1167 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1169 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1170 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1172 switch (trie_type) { \
1173 case trie_utf8_fold: \
1174 if ( foldlen>0 ) { \
1175 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1180 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1181 len = UTF8SKIP(uc); \
1182 skiplen = UNISKIP( uvc ); \
1183 foldlen -= skiplen; \
1184 uscan = foldbuf + skiplen; \
1187 case trie_latin_utf8_fold: \
1188 if ( foldlen>0 ) { \
1189 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1195 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1196 skiplen = UNISKIP( uvc ); \
1197 foldlen -= skiplen; \
1198 uscan = foldbuf + skiplen; \
1202 uvc = utf8n_to_uvuni( (const 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_SCAN(CoNd) \
1226 && (ln == 1 || folder(s, pat_string, ln)) \
1227 && (!reginfo || regtry(reginfo, &s)) ) \
1233 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1235 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1241 #define REXEC_FBC_SCAN(CoDe) \
1243 while (s < strend) { \
1249 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1250 REXEC_FBC_UTF8_SCAN( \
1252 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1261 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1264 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1273 #define REXEC_FBC_TRYIT \
1274 if ((!reginfo || regtry(reginfo, &s))) \
1277 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1278 if (utf8_target) { \
1279 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1282 REXEC_FBC_CLASS_SCAN(CoNd); \
1285 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1286 if (utf8_target) { \
1288 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1291 REXEC_FBC_CLASS_SCAN(CoNd); \
1294 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1295 PL_reg_flags |= RF_tainted; \
1296 if (utf8_target) { \
1297 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1300 REXEC_FBC_CLASS_SCAN(CoNd); \
1303 #define DUMP_EXEC_POS(li,s,doutf8) \
1304 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1307 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1308 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1309 tmp = TEST_NON_UTF8(tmp); \
1310 REXEC_FBC_UTF8_SCAN( \
1311 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1320 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1321 if (s == PL_bostr) { \
1325 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1326 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1329 LOAD_UTF8_CHARCLASS_ALNUM(); \
1330 REXEC_FBC_UTF8_SCAN( \
1331 if (tmp == ! (TeSt2_UtF8)) { \
1340 /* The only difference between the BOUND and NBOUND cases is that
1341 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1342 * NBOUND. This is accomplished by passing it in either the if or else clause,
1343 * with the other one being empty */
1344 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1345 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1347 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1348 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1350 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1351 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1353 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1354 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1357 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1358 * be passed in completely with the variable name being tested, which isn't
1359 * such a clean interface, but this is easier to read than it was before. We
1360 * are looking for the boundary (or non-boundary between a word and non-word
1361 * character. The utf8 and non-utf8 cases have the same logic, but the details
1362 * must be different. Find the "wordness" of the character just prior to this
1363 * one, and compare it with the wordness of this one. If they differ, we have
1364 * a boundary. At the beginning of the string, pretend that the previous
1365 * character was a new-line */
1366 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1367 if (utf8_target) { \
1370 else { /* Not utf8 */ \
1371 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1372 tmp = TEST_NON_UTF8(tmp); \
1374 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1383 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1386 /* We know what class REx starts with. Try to find this position... */
1387 /* if reginfo is NULL, its a dryrun */
1388 /* annoyingly all the vars in this routine have different names from their counterparts
1389 in regmatch. /grrr */
1392 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1393 const char *strend, regmatch_info *reginfo)
1396 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1397 char *pat_string; /* The pattern's exactish string */
1398 char *pat_end; /* ptr to end char of pat_string */
1399 re_fold_t folder; /* Function for computing non-utf8 folds */
1400 const U8 *fold_array; /* array for folding ords < 256 */
1403 register STRLEN uskip;
1407 register I32 tmp = 1; /* Scratch variable? */
1408 register const bool utf8_target = PL_reg_match_utf8;
1409 UV utf8_fold_flags = 0;
1410 RXi_GET_DECL(prog,progi);
1412 PERL_ARGS_ASSERT_FIND_BYCLASS;
1414 /* We know what class it must start with. */
1418 if (utf8_target || OP(c) == ANYOFV) {
1419 STRLEN inclasslen = strend - s;
1420 REXEC_FBC_UTF8_CLASS_SCAN(
1421 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1424 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1429 if (tmp && (!reginfo || regtry(reginfo, &s)))
1437 if (UTF_PATTERN || utf8_target) {
1438 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1439 goto do_exactf_utf8;
1441 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1442 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1443 goto do_exactf_non_utf8; /* isn't dealt with by these */
1448 /* regcomp.c already folded this if pattern is in UTF-8 */
1449 utf8_fold_flags = 0;
1450 goto do_exactf_utf8;
1452 fold_array = PL_fold;
1454 goto do_exactf_non_utf8;
1457 if (UTF_PATTERN || utf8_target) {
1458 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1459 goto do_exactf_utf8;
1461 fold_array = PL_fold_locale;
1462 folder = foldEQ_locale;
1463 goto do_exactf_non_utf8;
1467 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1469 goto do_exactf_utf8;
1471 case EXACTFU_TRICKYFOLD:
1473 if (UTF_PATTERN || utf8_target) {
1474 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1475 goto do_exactf_utf8;
1478 /* Any 'ss' in the pattern should have been replaced by regcomp,
1479 * so we don't have to worry here about this single special case
1480 * in the Latin1 range */
1481 fold_array = PL_fold_latin1;
1482 folder = foldEQ_latin1;
1486 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1487 are no glitches with fold-length differences
1488 between the target string and pattern */
1490 /* The idea in the non-utf8 EXACTF* cases is to first find the
1491 * first character of the EXACTF* node and then, if necessary,
1492 * case-insensitively compare the full text of the node. c1 is the
1493 * first character. c2 is its fold. This logic will not work for
1494 * Unicode semantics and the german sharp ss, which hence should
1495 * not be compiled into a node that gets here. */
1496 pat_string = STRING(c);
1497 ln = STR_LEN(c); /* length to match in octets/bytes */
1499 /* We know that we have to match at least 'ln' bytes (which is the
1500 * same as characters, since not utf8). If we have to match 3
1501 * characters, and there are only 2 availabe, we know without
1502 * trying that it will fail; so don't start a match past the
1503 * required minimum number from the far end */
1504 e = HOP3c(strend, -((I32)ln), s);
1506 if (!reginfo && e < s) {
1507 e = s; /* Due to minlen logic of intuit() */
1511 c2 = fold_array[c1];
1512 if (c1 == c2) { /* If char and fold are the same */
1513 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1516 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1525 /* If one of the operands is in utf8, we can't use the simpler
1526 * folding above, due to the fact that many different characters
1527 * can have the same fold, or portion of a fold, or different-
1529 pat_string = STRING(c);
1530 ln = STR_LEN(c); /* length to match in octets/bytes */
1531 pat_end = pat_string + ln;
1532 lnc = (UTF_PATTERN) /* length to match in characters */
1533 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1536 /* We have 'lnc' characters to match in the pattern, but because of
1537 * multi-character folding, each character in the target can match
1538 * up to 3 characters (Unicode guarantees it will never exceed
1539 * this) if it is utf8-encoded; and up to 2 if not (based on the
1540 * fact that the Latin 1 folds are already determined, and the
1541 * only multi-char fold in that range is the sharp-s folding to
1542 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1543 * string character. Adjust lnc accordingly, rounding up, so that
1544 * if we need to match at least 4+1/3 chars, that really is 5. */
1545 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1546 lnc = (lnc + expansion - 1) / expansion;
1548 /* As in the non-UTF8 case, if we have to match 3 characters, and
1549 * only 2 are left, it's guaranteed to fail, so don't start a
1550 * match that would require us to go beyond the end of the string
1552 e = HOP3c(strend, -((I32)lnc), s);
1554 if (!reginfo && e < s) {
1555 e = s; /* Due to minlen logic of intuit() */
1558 /* XXX Note that we could recalculate e to stop the loop earlier,
1559 * as the worst case expansion above will rarely be met, and as we
1560 * go along we would usually find that e moves further to the left.
1561 * This would happen only after we reached the point in the loop
1562 * where if there were no expansion we should fail. Unclear if
1563 * worth the expense */
1566 char *my_strend= (char *)strend;
1567 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1568 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1569 && (!reginfo || regtry(reginfo, &s)) )
1573 s += (utf8_target) ? UTF8SKIP(s) : 1;
1578 PL_reg_flags |= RF_tainted;
1579 FBC_BOUND(isALNUM_LC,
1580 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1581 isALNUM_LC_utf8((U8*)s));
1584 PL_reg_flags |= RF_tainted;
1585 FBC_NBOUND(isALNUM_LC,
1586 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1587 isALNUM_LC_utf8((U8*)s));
1590 FBC_BOUND(isWORDCHAR,
1592 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1595 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1597 isWORDCHAR_A((U8*)s));
1600 FBC_NBOUND(isWORDCHAR,
1602 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1605 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1607 isWORDCHAR_A((U8*)s));
1610 FBC_BOUND(isWORDCHAR_L1,
1612 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1615 FBC_NBOUND(isWORDCHAR_L1,
1617 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1620 REXEC_FBC_CSCAN_TAINT(
1621 isALNUM_LC_utf8((U8*)s),
1626 REXEC_FBC_CSCAN_PRELOAD(
1627 LOAD_UTF8_CHARCLASS_ALNUM(),
1628 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1629 isWORDCHAR_L1((U8) *s)
1633 REXEC_FBC_CSCAN_PRELOAD(
1634 LOAD_UTF8_CHARCLASS_ALNUM(),
1635 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1640 /* Don't need to worry about utf8, as it can match only a single
1641 * byte invariant character */
1642 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1645 REXEC_FBC_CSCAN_PRELOAD(
1646 LOAD_UTF8_CHARCLASS_ALNUM(),
1647 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1648 ! isWORDCHAR_L1((U8) *s)
1652 REXEC_FBC_CSCAN_PRELOAD(
1653 LOAD_UTF8_CHARCLASS_ALNUM(),
1654 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1665 REXEC_FBC_CSCAN_TAINT(
1666 !isALNUM_LC_utf8((U8*)s),
1671 REXEC_FBC_CSCAN_PRELOAD(
1672 LOAD_UTF8_CHARCLASS_SPACE(),
1673 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1678 REXEC_FBC_CSCAN_PRELOAD(
1679 LOAD_UTF8_CHARCLASS_SPACE(),
1680 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1685 /* Don't need to worry about utf8, as it can match only a single
1686 * byte invariant character */
1687 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1690 REXEC_FBC_CSCAN_TAINT(
1691 isSPACE_LC_utf8((U8*)s),
1696 REXEC_FBC_CSCAN_PRELOAD(
1697 LOAD_UTF8_CHARCLASS_SPACE(),
1698 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1699 ! isSPACE_L1((U8) *s)
1703 REXEC_FBC_CSCAN_PRELOAD(
1704 LOAD_UTF8_CHARCLASS_SPACE(),
1705 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1716 REXEC_FBC_CSCAN_TAINT(
1717 !isSPACE_LC_utf8((U8*)s),
1722 REXEC_FBC_CSCAN_PRELOAD(
1723 LOAD_UTF8_CHARCLASS_DIGIT(),
1724 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1729 /* Don't need to worry about utf8, as it can match only a single
1730 * byte invariant character */
1731 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1734 REXEC_FBC_CSCAN_TAINT(
1735 isDIGIT_LC_utf8((U8*)s),
1740 REXEC_FBC_CSCAN_PRELOAD(
1741 LOAD_UTF8_CHARCLASS_DIGIT(),
1742 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1753 REXEC_FBC_CSCAN_TAINT(
1754 !isDIGIT_LC_utf8((U8*)s),
1761 is_LNBREAK_latin1(s)
1773 !is_VERTWS_latin1(s)
1779 is_HORIZWS_latin1(s)
1784 !is_HORIZWS_utf8(s),
1785 !is_HORIZWS_latin1(s)
1792 /* what trie are we using right now */
1794 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1796 = (reg_trie_data*)progi->data->data[ aho->trie ];
1797 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1799 const char *last_start = strend - trie->minlen;
1801 const char *real_start = s;
1803 STRLEN maxlen = trie->maxlen;
1805 U8 **points; /* map of where we were in the input string
1806 when reading a given char. For ASCII this
1807 is unnecessary overhead as the relationship
1808 is always 1:1, but for Unicode, especially
1809 case folded Unicode this is not true. */
1810 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1814 GET_RE_DEBUG_FLAGS_DECL;
1816 /* We can't just allocate points here. We need to wrap it in
1817 * an SV so it gets freed properly if there is a croak while
1818 * running the match */
1821 sv_points=newSV(maxlen * sizeof(U8 *));
1822 SvCUR_set(sv_points,
1823 maxlen * sizeof(U8 *));
1824 SvPOK_on(sv_points);
1825 sv_2mortal(sv_points);
1826 points=(U8**)SvPV_nolen(sv_points );
1827 if ( trie_type != trie_utf8_fold
1828 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1831 bitmap=(U8*)trie->bitmap;
1833 bitmap=(U8*)ANYOF_BITMAP(c);
1835 /* this is the Aho-Corasick algorithm modified a touch
1836 to include special handling for long "unknown char"
1837 sequences. The basic idea being that we use AC as long
1838 as we are dealing with a possible matching char, when
1839 we encounter an unknown char (and we have not encountered
1840 an accepting state) we scan forward until we find a legal
1842 AC matching is basically that of trie matching, except
1843 that when we encounter a failing transition, we fall back
1844 to the current states "fail state", and try the current char
1845 again, a process we repeat until we reach the root state,
1846 state 1, or a legal transition. If we fail on the root state
1847 then we can either terminate if we have reached an accepting
1848 state previously, or restart the entire process from the beginning
1852 while (s <= last_start) {
1853 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1861 U8 *uscan = (U8*)NULL;
1862 U8 *leftmost = NULL;
1864 U32 accepted_word= 0;
1868 while ( state && uc <= (U8*)strend ) {
1870 U32 word = aho->states[ state ].wordnum;
1874 DEBUG_TRIE_EXECUTE_r(
1875 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1876 dump_exec_pos( (char *)uc, c, strend, real_start,
1877 (char *)uc, utf8_target );
1878 PerlIO_printf( Perl_debug_log,
1879 " Scanning for legal start char...\n");
1883 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1887 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1893 if (uc >(U8*)last_start) break;
1897 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1898 if (!leftmost || lpos < leftmost) {
1899 DEBUG_r(accepted_word=word);
1905 points[pointpos++ % maxlen]= uc;
1906 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1907 uscan, len, uvc, charid, foldlen,
1909 DEBUG_TRIE_EXECUTE_r({
1910 dump_exec_pos( (char *)uc, c, strend, real_start,
1912 PerlIO_printf(Perl_debug_log,
1913 " Charid:%3u CP:%4"UVxf" ",
1919 word = aho->states[ state ].wordnum;
1921 base = aho->states[ state ].trans.base;
1923 DEBUG_TRIE_EXECUTE_r({
1925 dump_exec_pos( (char *)uc, c, strend, real_start,
1927 PerlIO_printf( Perl_debug_log,
1928 "%sState: %4"UVxf", word=%"UVxf,
1929 failed ? " Fail transition to " : "",
1930 (UV)state, (UV)word);
1936 ( ((offset = base + charid
1937 - 1 - trie->uniquecharcount)) >= 0)
1938 && ((U32)offset < trie->lasttrans)
1939 && trie->trans[offset].check == state
1940 && (tmp=trie->trans[offset].next))
1942 DEBUG_TRIE_EXECUTE_r(
1943 PerlIO_printf( Perl_debug_log," - legal\n"));
1948 DEBUG_TRIE_EXECUTE_r(
1949 PerlIO_printf( Perl_debug_log," - fail\n"));
1951 state = aho->fail[state];
1955 /* we must be accepting here */
1956 DEBUG_TRIE_EXECUTE_r(
1957 PerlIO_printf( Perl_debug_log," - accepting\n"));
1966 if (!state) state = 1;
1969 if ( aho->states[ state ].wordnum ) {
1970 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1971 if (!leftmost || lpos < leftmost) {
1972 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1977 s = (char*)leftmost;
1978 DEBUG_TRIE_EXECUTE_r({
1980 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1981 (UV)accepted_word, (IV)(s - real_start)
1984 if (!reginfo || regtry(reginfo, &s)) {
1990 DEBUG_TRIE_EXECUTE_r({
1991 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1994 DEBUG_TRIE_EXECUTE_r(
1995 PerlIO_printf( Perl_debug_log,"No match.\n"));
2004 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2014 - regexec_flags - match a regexp against a string
2017 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2018 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2019 /* strend: pointer to null at end of string */
2020 /* strbeg: real beginning of string */
2021 /* minend: end of match must be >=minend after stringarg. */
2022 /* data: May be used for some additional optimizations.
2023 Currently its only used, with a U32 cast, for transmitting
2024 the ganch offset when doing a /g match. This will change */
2025 /* nosave: For optimizations. */
2028 struct regexp *const prog = (struct regexp *)SvANY(rx);
2029 /*register*/ char *s;
2030 register regnode *c;
2031 /*register*/ char *startpos = stringarg;
2032 I32 minlen; /* must match at least this many chars */
2033 I32 dontbother = 0; /* how many characters not to try at end */
2034 I32 end_shift = 0; /* Same for the end. */ /* CC */
2035 I32 scream_pos = -1; /* Internal iterator of scream. */
2036 char *scream_olds = NULL;
2037 const bool utf8_target = cBOOL(DO_UTF8(sv));
2039 RXi_GET_DECL(prog,progi);
2040 regmatch_info reginfo; /* create some info to pass to regtry etc */
2041 regexp_paren_pair *swap = NULL;
2042 GET_RE_DEBUG_FLAGS_DECL;
2044 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2045 PERL_UNUSED_ARG(data);
2047 /* Be paranoid... */
2048 if (prog == NULL || startpos == NULL) {
2049 Perl_croak(aTHX_ "NULL regexp parameter");
2053 multiline = prog->extflags & RXf_PMf_MULTILINE;
2054 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2056 RX_MATCH_UTF8_set(rx, utf8_target);
2058 debug_start_match(rx, utf8_target, startpos, strend,
2062 minlen = prog->minlen;
2064 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2065 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2066 "String too short [regexec_flags]...\n"));
2071 /* Check validity of program. */
2072 if (UCHARAT(progi->program) != REG_MAGIC) {
2073 Perl_croak(aTHX_ "corrupted regexp program");
2077 PL_reg_eval_set = 0;
2081 PL_reg_flags |= RF_utf8;
2083 /* Mark beginning of line for ^ and lookbehind. */
2084 reginfo.bol = startpos; /* XXX not used ??? */
2088 /* Mark end of line for $ (and such) */
2091 /* see how far we have to get to not match where we matched before */
2092 reginfo.till = startpos+minend;
2094 /* If there is a "must appear" string, look for it. */
2097 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2099 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2100 reginfo.ganch = startpos + prog->gofs;
2101 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2102 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2103 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2105 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2106 && mg->mg_len >= 0) {
2107 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2108 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2109 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2111 if (prog->extflags & RXf_ANCH_GPOS) {
2112 if (s > reginfo.ganch)
2114 s = reginfo.ganch - prog->gofs;
2115 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2116 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2122 reginfo.ganch = strbeg + PTR2UV(data);
2123 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2124 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2126 } else { /* pos() not defined */
2127 reginfo.ganch = strbeg;
2128 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2129 "GPOS: reginfo.ganch = strbeg\n"));
2132 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2133 /* We have to be careful. If the previous successful match
2134 was from this regex we don't want a subsequent partially
2135 successful match to clobber the old results.
2136 So when we detect this possibility we add a swap buffer
2137 to the re, and switch the buffer each match. If we fail
2138 we switch it back, otherwise we leave it swapped.
2141 /* do we need a save destructor here for eval dies? */
2142 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2144 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2145 re_scream_pos_data d;
2147 d.scream_olds = &scream_olds;
2148 d.scream_pos = &scream_pos;
2149 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2151 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2152 goto phooey; /* not present */
2158 /* Simplest case: anchored match need be tried only once. */
2159 /* [unless only anchor is BOL and multiline is set] */
2160 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2161 if (s == startpos && regtry(®info, &startpos))
2163 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2164 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2169 dontbother = minlen - 1;
2170 end = HOP3c(strend, -dontbother, strbeg) - 1;
2171 /* for multiline we only have to try after newlines */
2172 if (prog->check_substr || prog->check_utf8) {
2173 /* because of the goto we can not easily reuse the macros for bifurcating the
2174 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2177 goto after_try_utf8;
2179 if (regtry(®info, &s)) {
2186 if (prog->extflags & RXf_USE_INTUIT) {
2187 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2196 } /* end search for check string in unicode */
2198 if (s == startpos) {
2199 goto after_try_latin;
2202 if (regtry(®info, &s)) {
2209 if (prog->extflags & RXf_USE_INTUIT) {
2210 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2219 } /* end search for check string in latin*/
2220 } /* end search for check string */
2221 else { /* search for newline */
2223 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2226 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2227 while (s <= end) { /* note it could be possible to match at the end of the string */
2228 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2229 if (regtry(®info, &s))
2233 } /* end search for newline */
2234 } /* end anchored/multiline check string search */
2236 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2238 /* the warning about reginfo.ganch being used without initialization
2239 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2240 and we only enter this block when the same bit is set. */
2241 char *tmp_s = reginfo.ganch - prog->gofs;
2243 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2248 /* Messy cases: unanchored match. */
2249 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2250 /* we have /x+whatever/ */
2251 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2256 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2257 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2258 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2263 DEBUG_EXECUTE_r( did_match = 1 );
2264 if (regtry(®info, &s)) goto got_it;
2266 while (s < strend && *s == ch)
2274 DEBUG_EXECUTE_r( did_match = 1 );
2275 if (regtry(®info, &s)) goto got_it;
2277 while (s < strend && *s == ch)
2282 DEBUG_EXECUTE_r(if (!did_match)
2283 PerlIO_printf(Perl_debug_log,
2284 "Did not find anchored character...\n")
2287 else if (prog->anchored_substr != NULL
2288 || prog->anchored_utf8 != NULL
2289 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2290 && prog->float_max_offset < strend - s)) {
2295 char *last1; /* Last position checked before */
2299 if (prog->anchored_substr || prog->anchored_utf8) {
2300 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2301 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2302 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2303 back_max = back_min = prog->anchored_offset;
2305 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2306 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2307 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2308 back_max = prog->float_max_offset;
2309 back_min = prog->float_min_offset;
2313 if (must == &PL_sv_undef)
2314 /* could not downgrade utf8 check substring, so must fail */
2320 last = HOP3c(strend, /* Cannot start after this */
2321 -(I32)(CHR_SVLEN(must)
2322 - (SvTAIL(must) != 0) + back_min), strbeg);
2325 last1 = HOPc(s, -1);
2327 last1 = s - 1; /* bogus */
2329 /* XXXX check_substr already used to find "s", can optimize if
2330 check_substr==must. */
2332 dontbother = end_shift;
2333 strend = HOPc(strend, -dontbother);
2334 while ( (s <= last) &&
2335 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2336 (unsigned char*)strend, must,
2337 multiline ? FBMrf_MULTILINE : 0)) ) {
2338 DEBUG_EXECUTE_r( did_match = 1 );
2339 if (HOPc(s, -back_max) > last1) {
2340 last1 = HOPc(s, -back_min);
2341 s = HOPc(s, -back_max);
2344 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2346 last1 = HOPc(s, -back_min);
2350 while (s <= last1) {
2351 if (regtry(®info, &s))
2357 while (s <= last1) {
2358 if (regtry(®info, &s))
2364 DEBUG_EXECUTE_r(if (!did_match) {
2365 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2366 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2367 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2368 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2369 ? "anchored" : "floating"),
2370 quoted, RE_SV_TAIL(must));
2374 else if ( (c = progi->regstclass) ) {
2376 const OPCODE op = OP(progi->regstclass);
2377 /* don't bother with what can't match */
2378 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2379 strend = HOPc(strend, -(minlen - 1));
2382 SV * const prop = sv_newmortal();
2383 regprop(prog, prop, c);
2385 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2387 PerlIO_printf(Perl_debug_log,
2388 "Matching stclass %.*s against %s (%d bytes)\n",
2389 (int)SvCUR(prop), SvPVX_const(prop),
2390 quoted, (int)(strend - s));
2393 if (find_byclass(prog, c, s, strend, ®info))
2395 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2399 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2406 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2407 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2408 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2410 little = SvPV_const(float_real, len);
2411 if (SvTAIL(float_real)) {
2412 /* This means that float_real contains an artificial \n on the end
2413 * due to the presence of something like this: /foo$/
2414 * where we can match both "foo" and "foo\n" at the end of the string.
2415 * So we have to compare the end of the string first against the float_real
2416 * without the \n and then against the full float_real with the string.
2417 * We have to watch out for cases where the string might be smaller
2418 * than the float_real or the float_real without the \n.
2420 char *checkpos= strend - len;
2422 PerlIO_printf(Perl_debug_log,
2423 "%sChecking for float_real.%s\n",
2424 PL_colors[4], PL_colors[5]));
2425 if (checkpos + 1 < strbeg) {
2426 /* can't match, even if we remove the trailing \n string is too short to match */
2428 PerlIO_printf(Perl_debug_log,
2429 "%sString shorter than required trailing substring, cannot match.%s\n",
2430 PL_colors[4], PL_colors[5]));
2432 } else if (memEQ(checkpos + 1, little, len - 1)) {
2433 /* can match, the end of the string matches without the "\n" */
2434 last = checkpos + 1;
2435 } else if (checkpos < strbeg) {
2436 /* cant match, string is too short when the "\n" is included */
2438 PerlIO_printf(Perl_debug_log,
2439 "%sString does not contain required trailing substring, cannot match.%s\n",
2440 PL_colors[4], PL_colors[5]));
2442 } else if (!multiline) {
2443 /* non multiline match, so compare with the "\n" at the end of the string */
2444 if (memEQ(checkpos, little, len)) {
2448 PerlIO_printf(Perl_debug_log,
2449 "%sString does not contain required trailing substring, cannot match.%s\n",
2450 PL_colors[4], PL_colors[5]));
2454 /* multiline match, so we have to search for a place where the full string is located */
2460 last = rninstr(s, strend, little, little + len);
2462 last = strend; /* matching "$" */
2465 /* at one point this block contained a comment which was probably
2466 * incorrect, which said that this was a "should not happen" case.
2467 * Even if it was true when it was written I am pretty sure it is
2468 * not anymore, so I have removed the comment and replaced it with
2471 PerlIO_printf(Perl_debug_log,
2472 "String does not contain required substring, cannot match.\n"
2476 dontbother = strend - last + prog->float_min_offset;
2478 if (minlen && (dontbother < minlen))
2479 dontbother = minlen - 1;
2480 strend -= dontbother; /* this one's always in bytes! */
2481 /* We don't know much -- general case. */
2484 if (regtry(®info, &s))
2493 if (regtry(®info, &s))
2495 } while (s++ < strend);
2504 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2506 if (PL_reg_eval_set)
2507 restore_pos(aTHX_ prog);
2508 if (RXp_PAREN_NAMES(prog))
2509 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2511 /* make sure $`, $&, $', and $digit will work later */
2512 if ( !(flags & REXEC_NOT_FIRST) ) {
2513 RX_MATCH_COPY_FREE(rx);
2514 if (flags & REXEC_COPY_STR) {
2515 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2516 #ifdef PERL_OLD_COPY_ON_WRITE
2518 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2520 PerlIO_printf(Perl_debug_log,
2521 "Copy on write: regexp capture, type %d\n",
2524 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2525 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2526 assert (SvPOKp(prog->saved_copy));
2530 RX_MATCH_COPIED_on(rx);
2531 s = savepvn(strbeg, i);
2537 prog->subbeg = strbeg;
2538 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2545 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2546 PL_colors[4], PL_colors[5]));
2547 if (PL_reg_eval_set)
2548 restore_pos(aTHX_ prog);
2550 /* we failed :-( roll it back */
2551 Safefree(prog->offs);
2560 - regtry - try match at specific point
2562 STATIC I32 /* 0 failure, 1 success */
2563 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2567 REGEXP *const rx = reginfo->prog;
2568 regexp *const prog = (struct regexp *)SvANY(rx);
2569 RXi_GET_DECL(prog,progi);
2570 GET_RE_DEBUG_FLAGS_DECL;
2572 PERL_ARGS_ASSERT_REGTRY;
2574 reginfo->cutpoint=NULL;
2576 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2579 PL_reg_eval_set = RS_init;
2580 DEBUG_EXECUTE_r(DEBUG_s(
2581 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2582 (IV)(PL_stack_sp - PL_stack_base));
2585 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2586 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2588 /* Apparently this is not needed, judging by wantarray. */
2589 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2590 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2593 /* Make $_ available to executed code. */
2594 if (reginfo->sv != DEFSV) {
2596 DEFSV_set(reginfo->sv);
2599 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2600 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2601 /* prepare for quick setting of pos */
2602 #ifdef PERL_OLD_COPY_ON_WRITE
2603 if (SvIsCOW(reginfo->sv))
2604 sv_force_normal_flags(reginfo->sv, 0);
2606 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2607 &PL_vtbl_mglob, NULL, 0);
2611 PL_reg_oldpos = mg->mg_len;
2612 SAVEDESTRUCTOR_X(restore_pos, prog);
2614 if (!PL_reg_curpm) {
2615 Newxz(PL_reg_curpm, 1, PMOP);
2618 SV* const repointer = &PL_sv_undef;
2619 /* this regexp is also owned by the new PL_reg_curpm, which
2620 will try to free it. */
2621 av_push(PL_regex_padav, repointer);
2622 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2623 PL_regex_pad = AvARRAY(PL_regex_padav);
2628 /* It seems that non-ithreads works both with and without this code.
2629 So for efficiency reasons it seems best not to have the code
2630 compiled when it is not needed. */
2631 /* This is safe against NULLs: */
2632 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2633 /* PM_reg_curpm owns a reference to this regexp. */
2634 (void)ReREFCNT_inc(rx);
2636 PM_SETRE(PL_reg_curpm, rx);
2637 PL_reg_oldcurpm = PL_curpm;
2638 PL_curpm = PL_reg_curpm;
2639 if (RXp_MATCH_COPIED(prog)) {
2640 /* Here is a serious problem: we cannot rewrite subbeg,
2641 since it may be needed if this match fails. Thus
2642 $` inside (?{}) could fail... */
2643 PL_reg_oldsaved = prog->subbeg;
2644 PL_reg_oldsavedlen = prog->sublen;
2645 #ifdef PERL_OLD_COPY_ON_WRITE
2646 PL_nrs = prog->saved_copy;
2648 RXp_MATCH_COPIED_off(prog);
2651 PL_reg_oldsaved = NULL;
2652 prog->subbeg = PL_bostr;
2653 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2655 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2656 prog->offs[0].start = *startpos - PL_bostr;
2657 PL_reginput = *startpos;
2658 PL_reglastparen = &prog->lastparen;
2659 PL_reglastcloseparen = &prog->lastcloseparen;
2660 prog->lastparen = 0;
2661 prog->lastcloseparen = 0;
2663 PL_regoffs = prog->offs;
2664 if (PL_reg_start_tmpl <= prog->nparens) {
2665 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2666 if(PL_reg_start_tmp)
2667 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2669 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2672 /* XXXX What this code is doing here?!!! There should be no need
2673 to do this again and again, PL_reglastparen should take care of
2676 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2677 * Actually, the code in regcppop() (which Ilya may be meaning by
2678 * PL_reglastparen), is not needed at all by the test suite
2679 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2680 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2681 * Meanwhile, this code *is* needed for the
2682 * above-mentioned test suite tests to succeed. The common theme
2683 * on those tests seems to be returning null fields from matches.
2684 * --jhi updated by dapm */
2686 if (prog->nparens) {
2687 regexp_paren_pair *pp = PL_regoffs;
2689 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2697 if (regmatch(reginfo, progi->program + 1)) {
2698 PL_regoffs[0].end = PL_reginput - PL_bostr;
2701 if (reginfo->cutpoint)
2702 *startpos= reginfo->cutpoint;
2703 REGCP_UNWIND(lastcp);
2708 #define sayYES goto yes
2709 #define sayNO goto no
2710 #define sayNO_SILENT goto no_silent
2712 /* we dont use STMT_START/END here because it leads to
2713 "unreachable code" warnings, which are bogus, but distracting. */
2714 #define CACHEsayNO \
2715 if (ST.cache_mask) \
2716 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2719 /* this is used to determine how far from the left messages like
2720 'failed...' are printed. It should be set such that messages
2721 are inline with the regop output that created them.
2723 #define REPORT_CODE_OFF 32
2726 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2727 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2729 #define SLAB_FIRST(s) (&(s)->states[0])
2730 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2732 /* grab a new slab and return the first slot in it */
2734 STATIC regmatch_state *
2737 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2740 regmatch_slab *s = PL_regmatch_slab->next;
2742 Newx(s, 1, regmatch_slab);
2743 s->prev = PL_regmatch_slab;
2745 PL_regmatch_slab->next = s;
2747 PL_regmatch_slab = s;
2748 return SLAB_FIRST(s);
2752 /* push a new state then goto it */
2754 #define PUSH_STATE_GOTO(state, node) \
2756 st->resume_state = state; \
2759 /* push a new state with success backtracking, then goto it */
2761 #define PUSH_YES_STATE_GOTO(state, node) \
2763 st->resume_state = state; \
2764 goto push_yes_state;
2770 regmatch() - main matching routine
2772 This is basically one big switch statement in a loop. We execute an op,
2773 set 'next' to point the next op, and continue. If we come to a point which
2774 we may need to backtrack to on failure such as (A|B|C), we push a
2775 backtrack state onto the backtrack stack. On failure, we pop the top
2776 state, and re-enter the loop at the state indicated. If there are no more
2777 states to pop, we return failure.
2779 Sometimes we also need to backtrack on success; for example /A+/, where
2780 after successfully matching one A, we need to go back and try to
2781 match another one; similarly for lookahead assertions: if the assertion
2782 completes successfully, we backtrack to the state just before the assertion
2783 and then carry on. In these cases, the pushed state is marked as
2784 'backtrack on success too'. This marking is in fact done by a chain of
2785 pointers, each pointing to the previous 'yes' state. On success, we pop to
2786 the nearest yes state, discarding any intermediate failure-only states.
2787 Sometimes a yes state is pushed just to force some cleanup code to be
2788 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2789 it to free the inner regex.
2791 Note that failure backtracking rewinds the cursor position, while
2792 success backtracking leaves it alone.
2794 A pattern is complete when the END op is executed, while a subpattern
2795 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2796 ops trigger the "pop to last yes state if any, otherwise return true"
2799 A common convention in this function is to use A and B to refer to the two
2800 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2801 the subpattern to be matched possibly multiple times, while B is the entire
2802 rest of the pattern. Variable and state names reflect this convention.
2804 The states in the main switch are the union of ops and failure/success of
2805 substates associated with with that op. For example, IFMATCH is the op
2806 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2807 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2808 successfully matched A and IFMATCH_A_fail is a state saying that we have
2809 just failed to match A. Resume states always come in pairs. The backtrack
2810 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2811 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2812 on success or failure.
2814 The struct that holds a backtracking state is actually a big union, with
2815 one variant for each major type of op. The variable st points to the
2816 top-most backtrack struct. To make the code clearer, within each
2817 block of code we #define ST to alias the relevant union.
2819 Here's a concrete example of a (vastly oversimplified) IFMATCH
2825 #define ST st->u.ifmatch
2827 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2828 ST.foo = ...; // some state we wish to save
2830 // push a yes backtrack state with a resume value of
2831 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2833 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2836 case IFMATCH_A: // we have successfully executed A; now continue with B
2838 bar = ST.foo; // do something with the preserved value
2841 case IFMATCH_A_fail: // A failed, so the assertion failed
2842 ...; // do some housekeeping, then ...
2843 sayNO; // propagate the failure
2850 For any old-timers reading this who are familiar with the old recursive
2851 approach, the code above is equivalent to:
2853 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2862 ...; // do some housekeeping, then ...
2863 sayNO; // propagate the failure
2866 The topmost backtrack state, pointed to by st, is usually free. If you
2867 want to claim it, populate any ST.foo fields in it with values you wish to
2868 save, then do one of
2870 PUSH_STATE_GOTO(resume_state, node);
2871 PUSH_YES_STATE_GOTO(resume_state, node);
2873 which sets that backtrack state's resume value to 'resume_state', pushes a
2874 new free entry to the top of the backtrack stack, then goes to 'node'.
2875 On backtracking, the free slot is popped, and the saved state becomes the
2876 new free state. An ST.foo field in this new top state can be temporarily
2877 accessed to retrieve values, but once the main loop is re-entered, it
2878 becomes available for reuse.
2880 Note that the depth of the backtrack stack constantly increases during the
2881 left-to-right execution of the pattern, rather than going up and down with
2882 the pattern nesting. For example the stack is at its maximum at Z at the
2883 end of the pattern, rather than at X in the following:
2885 /(((X)+)+)+....(Y)+....Z/
2887 The only exceptions to this are lookahead/behind assertions and the cut,
2888 (?>A), which pop all the backtrack states associated with A before
2891 Backtrack state structs are allocated in slabs of about 4K in size.
2892 PL_regmatch_state and st always point to the currently active state,
2893 and PL_regmatch_slab points to the slab currently containing
2894 PL_regmatch_state. The first time regmatch() is called, the first slab is
2895 allocated, and is never freed until interpreter destruction. When the slab
2896 is full, a new one is allocated and chained to the end. At exit from
2897 regmatch(), slabs allocated since entry are freed.
2902 #define DEBUG_STATE_pp(pp) \
2904 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2905 PerlIO_printf(Perl_debug_log, \
2906 " %*s"pp" %s%s%s%s%s\n", \
2908 PL_reg_name[st->resume_state], \
2909 ((st==yes_state||st==mark_state) ? "[" : ""), \
2910 ((st==yes_state) ? "Y" : ""), \
2911 ((st==mark_state) ? "M" : ""), \
2912 ((st==yes_state||st==mark_state) ? "]" : "") \
2917 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2922 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2923 const char *start, const char *end, const char *blurb)
2925 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2927 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2932 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2933 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2935 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2936 start, end - start, 60);
2938 PerlIO_printf(Perl_debug_log,
2939 "%s%s REx%s %s against %s\n",
2940 PL_colors[4], blurb, PL_colors[5], s0, s1);
2942 if (utf8_target||utf8_pat)
2943 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2944 utf8_pat ? "pattern" : "",
2945 utf8_pat && utf8_target ? " and " : "",
2946 utf8_target ? "string" : ""
2952 S_dump_exec_pos(pTHX_ const char *locinput,
2953 const regnode *scan,
2954 const char *loc_regeol,
2955 const char *loc_bostr,
2956 const char *loc_reg_starttry,
2957 const bool utf8_target)
2959 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2960 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2961 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2962 /* The part of the string before starttry has one color
2963 (pref0_len chars), between starttry and current
2964 position another one (pref_len - pref0_len chars),
2965 after the current position the third one.
2966 We assume that pref0_len <= pref_len, otherwise we
2967 decrease pref0_len. */
2968 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2969 ? (5 + taill) - l : locinput - loc_bostr;
2972 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2974 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2976 pref0_len = pref_len - (locinput - loc_reg_starttry);
2977 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2978 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2979 ? (5 + taill) - pref_len : loc_regeol - locinput);
2980 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2984 if (pref0_len > pref_len)
2985 pref0_len = pref_len;
2987 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2989 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2990 (locinput - pref_len),pref0_len, 60, 4, 5);
2992 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2993 (locinput - pref_len + pref0_len),
2994 pref_len - pref0_len, 60, 2, 3);
2996 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2997 locinput, loc_regeol - locinput, 10, 0, 1);
2999 const STRLEN tlen=len0+len1+len2;
3000 PerlIO_printf(Perl_debug_log,
3001 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3002 (IV)(locinput - loc_bostr),
3005 (docolor ? "" : "> <"),
3007 (int)(tlen > 19 ? 0 : 19 - tlen),
3014 /* reg_check_named_buff_matched()
3015 * Checks to see if a named buffer has matched. The data array of
3016 * buffer numbers corresponding to the buffer is expected to reside
3017 * in the regexp->data->data array in the slot stored in the ARG() of
3018 * node involved. Note that this routine doesn't actually care about the
3019 * name, that information is not preserved from compilation to execution.
3020 * Returns the index of the leftmost defined buffer with the given name
3021 * or 0 if non of the buffers matched.
3024 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3027 RXi_GET_DECL(rex,rexi);
3028 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3029 I32 *nums=(I32*)SvPVX(sv_dat);
3031 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3033 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3034 if ((I32)*PL_reglastparen >= nums[n] &&
3035 PL_regoffs[nums[n]].end != -1)
3044 /* free all slabs above current one - called during LEAVE_SCOPE */
3047 S_clear_backtrack_stack(pTHX_ void *p)
3049 regmatch_slab *s = PL_regmatch_slab->next;
3054 PL_regmatch_slab->next = NULL;
3056 regmatch_slab * const osl = s;
3063 #define SETREX(Re1,Re2) \
3064 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3067 STATIC I32 /* 0 failure, 1 success */
3068 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3070 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3074 register const bool utf8_target = PL_reg_match_utf8;
3075 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3076 REGEXP *rex_sv = reginfo->prog;
3077 regexp *rex = (struct regexp *)SvANY(rex_sv);
3078 RXi_GET_DECL(rex,rexi);
3080 /* the current state. This is a cached copy of PL_regmatch_state */
3081 register regmatch_state *st;
3082 /* cache heavy used fields of st in registers */
3083 register regnode *scan;
3084 register regnode *next;
3085 register U32 n = 0; /* general value; init to avoid compiler warning */
3086 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3087 register char *locinput = PL_reginput;
3088 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3090 bool result = 0; /* return value of S_regmatch */
3091 int depth = 0; /* depth of backtrack stack */
3092 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3093 const U32 max_nochange_depth =
3094 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3095 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3096 regmatch_state *yes_state = NULL; /* state to pop to on success of
3098 /* mark_state piggy backs on the yes_state logic so that when we unwind
3099 the stack on success we can update the mark_state as we go */
3100 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3101 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3102 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3104 bool no_final = 0; /* prevent failure from backtracking? */
3105 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3106 char *startpoint = PL_reginput;
3107 SV *popmark = NULL; /* are we looking for a mark? */
3108 SV *sv_commit = NULL; /* last mark name seen in failure */
3109 SV *sv_yes_mark = NULL; /* last mark name we have seen
3110 during a successful match */
3111 U32 lastopen = 0; /* last open we saw */
3112 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3113 SV* const oreplsv = GvSV(PL_replgv);
3114 /* these three flags are set by various ops to signal information to
3115 * the very next op. They have a useful lifetime of exactly one loop
3116 * iteration, and are not preserved or restored by state pushes/pops
3118 bool sw = 0; /* the condition value in (?(cond)a|b) */
3119 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3120 int logical = 0; /* the following EVAL is:
3124 or the following IFMATCH/UNLESSM is:
3125 false: plain (?=foo)
3126 true: used as a condition: (?(?=foo))
3128 PAD* const initial_pad = PL_comppad;
3130 GET_RE_DEBUG_FLAGS_DECL;
3133 PERL_ARGS_ASSERT_REGMATCH;
3135 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3136 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3138 /* on first ever call to regmatch, allocate first slab */
3139 if (!PL_regmatch_slab) {
3140 Newx(PL_regmatch_slab, 1, regmatch_slab);
3141 PL_regmatch_slab->prev = NULL;
3142 PL_regmatch_slab->next = NULL;
3143 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3146 oldsave = PL_savestack_ix;
3147 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3148 SAVEVPTR(PL_regmatch_slab);
3149 SAVEVPTR(PL_regmatch_state);
3151 /* grab next free state slot */
3152 st = ++PL_regmatch_state;
3153 if (st > SLAB_LAST(PL_regmatch_slab))
3154 st = PL_regmatch_state = S_push_slab(aTHX);
3156 /* Note that nextchr is a byte even in UTF */
3157 nextchr = UCHARAT(locinput);
3159 while (scan != NULL) {
3162 SV * const prop = sv_newmortal();
3163 regnode *rnext=regnext(scan);
3164 DUMP_EXEC_POS( locinput, scan, utf8_target );
3165 regprop(rex, prop, scan);
3167 PerlIO_printf(Perl_debug_log,
3168 "%3"IVdf":%*s%s(%"IVdf")\n",
3169 (IV)(scan - rexi->program), depth*2, "",
3171 (PL_regkind[OP(scan)] == END || !rnext) ?
3172 0 : (IV)(rnext - rexi->program));
3175 next = scan + NEXT_OFF(scan);
3178 state_num = OP(scan);
3182 assert(PL_reglastparen == &rex->lastparen);
3183 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3184 assert(PL_regoffs == rex->offs);
3186 switch (state_num) {
3188 if (locinput == PL_bostr)
3190 /* reginfo->till = reginfo->bol; */
3195 if (locinput == PL_bostr ||
3196 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3202 if (locinput == PL_bostr)
3206 if (locinput == reginfo->ganch)
3211 /* update the startpoint */
3212 st->u.keeper.val = PL_regoffs[0].start;
3213 PL_reginput = locinput;
3214 PL_regoffs[0].start = locinput - PL_bostr;
3215 PUSH_STATE_GOTO(KEEPS_next, next);
3217 case KEEPS_next_fail:
3218 /* rollback the start point change */
3219 PL_regoffs[0].start = st->u.keeper.val;
3225 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3230 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3232 if (PL_regeol - locinput > 1)
3236 if (PL_regeol != locinput)
3240 if (!nextchr && locinput >= PL_regeol)
3243 locinput += PL_utf8skip[nextchr];
3244 if (locinput > PL_regeol)
3246 nextchr = UCHARAT(locinput);
3249 nextchr = UCHARAT(++locinput);
3252 if (!nextchr && locinput >= PL_regeol)
3254 nextchr = UCHARAT(++locinput);
3257 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3260 locinput += PL_utf8skip[nextchr];
3261 if (locinput > PL_regeol)
3263 nextchr = UCHARAT(locinput);
3266 nextchr = UCHARAT(++locinput);
3270 #define ST st->u.trie
3272 /* In this case the charclass data is available inline so
3273 we can fail fast without a lot of extra overhead.
3275 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3277 PerlIO_printf(Perl_debug_log,
3278 "%*s %sfailed to match trie start class...%s\n",
3279 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3286 /* the basic plan of execution of the trie is:
3287 * At the beginning, run though all the states, and
3288 * find the longest-matching word. Also remember the position
3289 * of the shortest matching word. For example, this pattern:
3292 * when matched against the string "abcde", will generate
3293 * accept states for all words except 3, with the longest
3294 * matching word being 4, and the shortest being 1 (with
3295 * the position being after char 1 of the string).
3297 * Then for each matching word, in word order (i.e. 1,2,4,5),
3298 * we run the remainder of the pattern; on each try setting
3299 * the current position to the character following the word,
3300 * returning to try the next word on failure.
3302 * We avoid having to build a list of words at runtime by
3303 * using a compile-time structure, wordinfo[].prev, which
3304 * gives, for each word, the previous accepting word (if any).
3305 * In the case above it would contain the mappings 1->2, 2->0,
3306 * 3->0, 4->5, 5->1. We can use this table to generate, from
3307 * the longest word (4 above), a list of all words, by
3308 * following the list of prev pointers; this gives us the
3309 * unordered list 4,5,1,2. Then given the current word we have
3310 * just tried, we can go through the list and find the
3311 * next-biggest word to try (so if we just failed on word 2,
3312 * the next in the list is 4).
3314 * Since at runtime we don't record the matching position in
3315 * the string for each word, we have to work that out for
3316 * each word we're about to process. The wordinfo table holds
3317 * the character length of each word; given that we recorded
3318 * at the start: the position of the shortest word and its
3319 * length in chars, we just need to move the pointer the
3320 * difference between the two char lengths. Depending on
3321 * Unicode status and folding, that's cheap or expensive.
3323 * This algorithm is optimised for the case where are only a
3324 * small number of accept states, i.e. 0,1, or maybe 2.
3325 * With lots of accepts states, and having to try all of them,
3326 * it becomes quadratic on number of accept states to find all
3331 /* what type of TRIE am I? (utf8 makes this contextual) */
3332 DECL_TRIE_TYPE(scan);
3334 /* what trie are we using right now */
3335 reg_trie_data * const trie
3336 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3337 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3338 U32 state = trie->startstate;
3340 if (trie->bitmap && !TRIE_BITMAP_TEST(trie,*locinput) ) {
3341 if (trie->states[ state ].wordnum) {
3343 PerlIO_printf(Perl_debug_log,
3344 "%*s %smatched empty string...%s\n",
3345 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3351 PerlIO_printf(Perl_debug_log,
3352 "%*s %sfailed to match trie start class...%s\n",
3353 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3360 U8 *uc = ( U8* )locinput;
3364 U8 *uscan = (U8*)NULL;
3365 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3366 U32 charcount = 0; /* how many input chars we have matched */
3367 U32 accepted = 0; /* have we seen any accepting states? */
3370 ST.jump = trie->jump;
3373 ST.longfold = FALSE; /* char longer if folded => it's harder */
3376 /* fully traverse the TRIE; note the position of the
3377 shortest accept state and the wordnum of the longest
3380 while ( state && uc <= (U8*)PL_regeol ) {
3381 U32 base = trie->states[ state ].trans.base;
3385 wordnum = trie->states[ state ].wordnum;
3387 if (wordnum) { /* it's an accept state */
3390 /* record first match position */
3392 ST.firstpos = (U8*)locinput;
3397 ST.firstchars = charcount;
3400 if (!ST.nextword || wordnum < ST.nextword)
3401 ST.nextword = wordnum;
3402 ST.topword = wordnum;
3405 DEBUG_TRIE_EXECUTE_r({
3406 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3407 PerlIO_printf( Perl_debug_log,
3408 "%*s %sState: %4"UVxf" Accepted: %c ",
3409 2+depth * 2, "", PL_colors[4],
3410 (UV)state, (accepted ? 'Y' : 'N'));
3413 /* read a char and goto next state */
3416 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3417 uscan, len, uvc, charid, foldlen,
3424 base + charid - 1 - trie->uniquecharcount)) >= 0)
3426 && ((U32)offset < trie->lasttrans)
3427 && trie->trans[offset].check == state)
3429 state = trie->trans[offset].next;
3440 DEBUG_TRIE_EXECUTE_r(
3441 PerlIO_printf( Perl_debug_log,
3442 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3443 charid, uvc, (UV)state, PL_colors[5] );
3449 /* calculate total number of accept states */
3454 w = trie->wordinfo[w].prev;
3457 ST.accepted = accepted;
3461 PerlIO_printf( Perl_debug_log,
3462 "%*s %sgot %"IVdf" possible matches%s\n",
3463 REPORT_CODE_OFF + depth * 2, "",
3464 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3466 goto trie_first_try; /* jump into the fail handler */
3470 case TRIE_next_fail: /* we failed - try next alternative */
3472 REGCP_UNWIND(ST.cp);
3473 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3474 PL_regoffs[n].end = -1;
3475 *PL_reglastparen = n;
3477 if (!--ST.accepted) {
3479 PerlIO_printf( Perl_debug_log,
3480 "%*s %sTRIE failed...%s\n",
3481 REPORT_CODE_OFF+depth*2, "",
3488 /* Find next-highest word to process. Note that this code
3489 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3490 register U16 min = 0;
3492 register U16 const nextword = ST.nextword;
3493 register reg_trie_wordinfo * const wordinfo
3494 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3495 for (word=ST.topword; word; word=wordinfo[word].prev) {
3496 if (word > nextword && (!min || word < min))
3509 ST.lastparen = *PL_reglastparen;
3513 /* find start char of end of current word */
3515 U32 chars; /* how many chars to skip */
3516 U8 *uc = ST.firstpos;
3517 reg_trie_data * const trie
3518 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3520 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3522 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3526 /* the hard option - fold each char in turn and find
3527 * its folded length (which may be different */
3528 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3536 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3544 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3549 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3563 PL_reginput = (char *)uc;
3566 scan = (ST.jump && ST.jump[ST.nextword])
3567 ? ST.me + ST.jump[ST.nextword]
3571 PerlIO_printf( Perl_debug_log,
3572 "%*s %sTRIE matched word #%d, continuing%s\n",
3573 REPORT_CODE_OFF+depth*2, "",
3580 if (ST.accepted > 1 || has_cutgroup) {
3581 PUSH_STATE_GOTO(TRIE_next, scan);
3584 /* only one choice left - just continue */
3586 AV *const trie_words
3587 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3588 SV ** const tmp = av_fetch( trie_words,
3590 SV *sv= tmp ? sv_newmortal() : NULL;
3592 PerlIO_printf( Perl_debug_log,
3593 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3594 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3596 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3597 PL_colors[0], PL_colors[1],
3598 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3600 : "not compiled under -Dr",
3604 locinput = PL_reginput;
3605 nextchr = UCHARAT(locinput);
3606 continue; /* execute rest of RE */
3611 char *s = STRING(scan);
3613 if (utf8_target != UTF_PATTERN) {
3614 /* The target and the pattern have differing utf8ness. */
3616 const char * const e = s + ln;
3619 /* The target is utf8, the pattern is not utf8. */
3624 if (NATIVE_TO_UNI(*(U8*)s) !=
3625 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3633 /* The target is not utf8, the pattern is utf8. */
3638 if (NATIVE_TO_UNI(*((U8*)l)) !=
3639 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3647 nextchr = UCHARAT(locinput);
3650 /* The target and the pattern have the same utf8ness. */
3651 /* Inline the first character, for speed. */
3652 if (UCHARAT(s) != nextchr)
3654 if (PL_regeol - locinput < ln)
3656 if (ln > 1 && memNE(s, locinput, ln))
3659 nextchr = UCHARAT(locinput);
3664 const U8 * fold_array;
3666 U32 fold_utf8_flags;
3668 PL_reg_flags |= RF_tainted;
3669 folder = foldEQ_locale;
3670 fold_array = PL_fold_locale;
3671 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3675 case EXACTFU_TRICKYFOLD:
3677 folder = foldEQ_latin1;
3678 fold_array = PL_fold_latin1;
3679 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3683 folder = foldEQ_latin1;
3684 fold_array = PL_fold_latin1;
3685 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3690 fold_array = PL_fold;
3691 fold_utf8_flags = 0;
3697 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
3698 /* Either target or the pattern are utf8, or has the issue where
3699 * the fold lengths may differ. */
3700 const char * const l = locinput;
3701 char *e = PL_regeol;
3703 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3704 l, &e, 0, utf8_target, fold_utf8_flags))
3709 nextchr = UCHARAT(locinput);
3713 /* Neither the target nor the pattern are utf8 */
3714 if (UCHARAT(s) != nextchr &&
3715 UCHARAT(s) != fold_array[nextchr])
3719 if (PL_regeol - locinput < ln)
3721 if (ln > 1 && ! folder(s, locinput, ln))
3724 nextchr = UCHARAT(locinput);
3728 /* XXX Could improve efficiency by separating these all out using a
3729 * macro or in-line function. At that point regcomp.c would no longer
3730 * have to set the FLAGS fields of these */
3733 PL_reg_flags |= RF_tainted;
3741 /* was last char in word? */
3743 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3744 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3746 if (locinput == PL_bostr)
3749 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3751 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3753 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3754 ln = isALNUM_uni(ln);
3755 LOAD_UTF8_CHARCLASS_ALNUM();
3756 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3759 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3760 n = isALNUM_LC_utf8((U8*)locinput);
3765 /* Here the string isn't utf8, or is utf8 and only ascii
3766 * characters are to match \w. In the latter case looking at
3767 * the byte just prior to the current one may be just the final
3768 * byte of a multi-byte character. This is ok. There are two
3770 * 1) it is a single byte character, and then the test is doing
3771 * just what it's supposed to.
3772 * 2) it is a multi-byte character, in which case the final
3773 * byte is never mistakable for ASCII, and so the test
3774 * will say it is not a word character, which is the
3775 * correct answer. */
3776 ln = (locinput != PL_bostr) ?
3777 UCHARAT(locinput - 1) : '\n';
3778 switch (FLAGS(scan)) {
3779 case REGEX_UNICODE_CHARSET:
3780 ln = isWORDCHAR_L1(ln);
3781 n = isWORDCHAR_L1(nextchr);
3783 case REGEX_LOCALE_CHARSET:
3784 ln = isALNUM_LC(ln);
3785 n = isALNUM_LC(nextchr);
3787 case REGEX_DEPENDS_CHARSET:
3789 n = isALNUM(nextchr);
3791 case REGEX_ASCII_RESTRICTED_CHARSET:
3792 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3793 ln = isWORDCHAR_A(ln);
3794 n = isWORDCHAR_A(nextchr);
3797 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3801 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3803 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3808 if (utf8_target || state_num == ANYOFV) {
3809 STRLEN inclasslen = PL_regeol - locinput;
3810 if (locinput >= PL_regeol)
3813 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3815 locinput += inclasslen;
3816 nextchr = UCHARAT(locinput);
3821 nextchr = UCHARAT(locinput);
3822 if (!nextchr && locinput >= PL_regeol)
3824 if (!REGINCLASS(rex, scan, (U8*)locinput))
3826 nextchr = UCHARAT(++locinput);
3830 /* Special char classes - The defines start on line 129 or so */
3831 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3832 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3833 ALNUMU, NALNUMU, isWORDCHAR_L1,