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
40 /* At least one required character in the target string is expressible only in
42 const char* const non_utf8_target_but_utf8_required
43 = "Can't match, because target string needs to be in UTF-8\n";
46 * pregcomp and pregexec -- regsub and regerror are not used in perl
48 * Copyright (c) 1986 by University of Toronto.
49 * Written by Henry Spencer. Not derived from licensed software.
51 * Permission is granted to anyone to use this software for any
52 * purpose on any computer system, and to redistribute it freely,
53 * subject to the following restrictions:
55 * 1. The author is not responsible for the consequences of use of
56 * this software, no matter how awful, even if they arise
59 * 2. The origin of this software must not be misrepresented, either
60 * by explicit claim or by omission.
62 * 3. Altered versions must be plainly marked as such, and must not
63 * be misrepresented as being the original software.
65 **** Alterations to Henry's code are...
67 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
68 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
69 **** by Larry Wall and others
71 **** You may distribute under the terms of either the GNU General Public
72 **** License or the Artistic License, as specified in the README file.
74 * Beware that some of this code is subtly aware of the way operator
75 * precedence is structured in regular expressions. Serious changes in
76 * regular-expression syntax might require a total rethink.
79 #define PERL_IN_REGEXEC_C
82 #ifdef PERL_IN_XSUB_RE
88 #include "inline_invlist.c"
89 #include "unicode_constants.h"
91 #define RF_tainted 1 /* tainted information used? e.g. locale */
92 #define RF_warned 2 /* warned about big count? */
94 #define RF_utf8 8 /* Pattern contains multibyte chars? */
96 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
98 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
101 #define STATIC static
104 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
105 * call if there are no complications: i.e., if everything matchable is
106 * straight forward in the bitmap */
107 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
108 : ANYOF_BITMAP_TEST(p,*(c)))
114 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
115 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
117 #define HOPc(pos,off) \
118 (char *)(PL_reg_match_utf8 \
119 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
121 #define HOPBACKc(pos, off) \
122 (char*)(PL_reg_match_utf8\
123 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
124 : (pos - off >= PL_bostr) \
128 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
129 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
132 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
133 #define NEXTCHR_IS_EOS (nextchr < 0)
135 #define SET_nextchr \
136 nextchr = ((locinput < PL_regeol) ? UCHARAT(locinput) : NEXTCHR_EOS)
138 #define SET_locinput(p) \
143 /* these are unrolled below in the CCC_TRY_XXX defined */
144 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
145 if (!CAT2(PL_utf8_,class)) { \
147 ENTER; save_re_context(); \
148 ok=CAT2(is_utf8_,class)((const U8*)str); \
149 PERL_UNUSED_VAR(ok); \
150 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
151 /* Doesn't do an assert to verify that is correct */
152 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
153 if (!CAT2(PL_utf8_,class)) { \
154 bool throw_away PERL_UNUSED_DECL; \
155 ENTER; save_re_context(); \
156 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
159 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
160 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
161 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
163 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
164 /* No asserts are done for some of these, in case called on a */ \
165 /* Unicode version in which they map to nothing */ \
166 LOAD_UTF8_CHARCLASS(X_regular_begin, HYPHEN_UTF8); \
167 LOAD_UTF8_CHARCLASS(X_extend, COMBINING_GRAVE_ACCENT_UTF8); \
169 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
171 /* The actual code for CCC_TRY, which uses several variables from the routine
172 * it's callable from. It is designed to be the bulk of a case statement.
173 * FUNC is the macro or function to call on non-utf8 targets that indicate if
174 * nextchr matches the class.
175 * UTF8_TEST is the whole test string to use for utf8 targets
176 * LOAD is what to use to test, and if not present to load in the swash for the
178 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
180 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
181 * utf8 and a variant, load the swash if necessary and test using the utf8
182 * test. Advance to the next character if test is ok, otherwise fail; If not
183 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
184 * fails, or advance to the next character */
186 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
187 if (NEXTCHR_IS_EOS) { \
190 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
191 LOAD_UTF8_CHARCLASS(CLASS, STR); \
192 if (POS_OR_NEG (UTF8_TEST)) { \
196 else if (POS_OR_NEG (FUNC(nextchr))) { \
199 goto increment_locinput;
201 /* Handle the non-locale cases for a character class and its complement. It
202 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
203 * This is because that code fails when the test succeeds, so we want to have
204 * the test fail so that the code succeeds. The swash is stored in a
205 * predictable PL_ place */
206 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
209 _CCC_TRY_CODE( !, FUNC, \
210 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
211 (U8*)locinput, TRUE)), \
214 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
215 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
216 (U8*)locinput, TRUE)), \
219 /* Generate the case statements for both locale and non-locale character
220 * classes in regmatch for classes that don't have special unicode semantics.
221 * Locales don't use an immediate swash, but an intermediary special locale
222 * function that is called on the pointer to the current place in the input
223 * string. That function will resolve to needing the same swash. One might
224 * think that because we don't know what the locale will match, we shouldn't
225 * check with the swash loading function that it loaded properly; ie, that we
226 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
227 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
229 #define CCC_TRY(NAME, NNAME, FUNC, \
230 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
231 NAMEA, NNAMEA, FUNCA, \
234 PL_reg_flags |= RF_tainted; \
235 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
237 PL_reg_flags |= RF_tainted; \
238 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
241 if (NEXTCHR_IS_EOS || ! FUNCA(nextchr)) { \
244 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
248 if (NEXTCHR_IS_EOS || FUNCA(nextchr)) { \
251 goto increment_locinput; \
252 /* Generate the non-locale cases */ \
253 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
255 /* This is like CCC_TRY, but has an extra set of parameters for generating case
256 * statements to handle separate Unicode semantics nodes */
257 #define CCC_TRY_U(NAME, NNAME, FUNC, \
258 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
259 NAMEU, NNAMEU, FUNCU, \
260 NAMEA, NNAMEA, FUNCA, \
262 CCC_TRY(NAME, NNAME, FUNC, \
263 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
264 NAMEA, NNAMEA, FUNCA, \
266 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
268 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
270 /* for use after a quantifier and before an EXACT-like node -- japhy */
271 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
273 * NOTE that *nothing* that affects backtracking should be in here, specifically
274 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
275 * node that is in between two EXACT like nodes when ascertaining what the required
276 * "follow" character is. This should probably be moved to regex compile time
277 * although it may be done at run time beause of the REF possibility - more
278 * investigation required. -- demerphq
280 #define JUMPABLE(rn) ( \
282 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
284 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
285 OP(rn) == PLUS || OP(rn) == MINMOD || \
287 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
289 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
291 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
294 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
295 we don't need this definition. */
296 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
297 #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 )
298 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
301 /* ... so we use this as its faster. */
302 #define IS_TEXT(rn) ( OP(rn)==EXACT )
303 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
304 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
305 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
310 Search for mandatory following text node; for lookahead, the text must
311 follow but for lookbehind (rn->flags != 0) we skip to the next step.
313 #define FIND_NEXT_IMPT(rn) STMT_START { \
314 while (JUMPABLE(rn)) { \
315 const OPCODE type = OP(rn); \
316 if (type == SUSPEND || PL_regkind[type] == CURLY) \
317 rn = NEXTOPER(NEXTOPER(rn)); \
318 else if (type == PLUS) \
320 else if (type == IFMATCH) \
321 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
322 else rn += NEXT_OFF(rn); \
327 static void restore_pos(pTHX_ void *arg);
329 #define REGCP_PAREN_ELEMS 3
330 #define REGCP_OTHER_ELEMS 3
331 #define REGCP_FRAME_ELEMS 1
332 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
333 * are needed for the regexp context stack bookkeeping. */
336 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor)
339 const int retval = PL_savestack_ix;
340 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
341 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
342 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
344 GET_RE_DEBUG_FLAGS_DECL;
346 PERL_ARGS_ASSERT_REGCPPUSH;
348 if (paren_elems_to_push < 0)
349 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
350 paren_elems_to_push);
352 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
353 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
354 " out of range (%lu-%ld)",
355 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
357 SSGROW(total_elems + REGCP_FRAME_ELEMS);
360 if ((int)PL_regsize > (int)parenfloor)
361 PerlIO_printf(Perl_debug_log,
362 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
367 for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
368 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
369 SSPUSHINT(rex->offs[p].end);
370 SSPUSHINT(rex->offs[p].start);
371 SSPUSHINT(rex->offs[p].start_tmp);
372 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
373 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
375 (IV)rex->offs[p].start,
376 (IV)rex->offs[p].start_tmp,
380 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
381 SSPUSHINT(PL_regsize);
382 SSPUSHINT(rex->lastparen);
383 SSPUSHINT(rex->lastcloseparen);
384 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
389 /* These are needed since we do not localize EVAL nodes: */
390 #define REGCP_SET(cp) \
392 PerlIO_printf(Perl_debug_log, \
393 " Setting an EVAL scope, savestack=%"IVdf"\n", \
394 (IV)PL_savestack_ix)); \
397 #define REGCP_UNWIND(cp) \
399 if (cp != PL_savestack_ix) \
400 PerlIO_printf(Perl_debug_log, \
401 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
402 (IV)(cp), (IV)PL_savestack_ix)); \
405 #define UNWIND_PAREN(lp, lcp) \
406 for (n = rex->lastparen; n > lp; n--) \
407 rex->offs[n].end = -1; \
408 rex->lastparen = n; \
409 rex->lastcloseparen = lcp;
413 S_regcppop(pTHX_ regexp *rex)
418 GET_RE_DEBUG_FLAGS_DECL;
420 PERL_ARGS_ASSERT_REGCPPOP;
422 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
424 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
425 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
426 rex->lastcloseparen = SSPOPINT;
427 rex->lastparen = SSPOPINT;
428 PL_regsize = SSPOPINT;
430 i -= REGCP_OTHER_ELEMS;
431 /* Now restore the parentheses context. */
433 if (i || rex->lastparen + 1 <= rex->nparens)
434 PerlIO_printf(Perl_debug_log,
435 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
441 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
443 rex->offs[paren].start_tmp = SSPOPINT;
444 rex->offs[paren].start = SSPOPINT;
446 if (paren <= rex->lastparen)
447 rex->offs[paren].end = tmps;
448 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
449 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
451 (IV)rex->offs[paren].start,
452 (IV)rex->offs[paren].start_tmp,
453 (IV)rex->offs[paren].end,
454 (paren > rex->lastparen ? "(skipped)" : ""));
459 /* It would seem that the similar code in regtry()
460 * already takes care of this, and in fact it is in
461 * a better location to since this code can #if 0-ed out
462 * but the code in regtry() is needed or otherwise tests
463 * requiring null fields (pat.t#187 and split.t#{13,14}
464 * (as of patchlevel 7877) will fail. Then again,
465 * this code seems to be necessary or otherwise
466 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
467 * --jhi updated by dapm */
468 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
470 rex->offs[i].start = -1;
471 rex->offs[i].end = -1;
472 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
473 " \\%"UVuf": %s ..-1 undeffing\n",
475 (i > PL_regsize) ? "-1" : " "
481 /* restore the parens and associated vars at savestack position ix,
482 * but without popping the stack */
485 S_regcp_restore(pTHX_ regexp *rex, I32 ix)
487 I32 tmpix = PL_savestack_ix;
488 PL_savestack_ix = ix;
490 PL_savestack_ix = tmpix;
493 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
496 * pregexec and friends
499 #ifndef PERL_IN_XSUB_RE
501 - pregexec - match a regexp against a string
504 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
505 char *strbeg, I32 minend, SV *screamer, U32 nosave)
506 /* stringarg: the point in the string at which to begin matching */
507 /* strend: pointer to null at end of string */
508 /* strbeg: real beginning of string */
509 /* minend: end of match must be >= minend bytes after stringarg. */
510 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
511 * itself is accessed via the pointers above */
512 /* nosave: For optimizations. */
514 PERL_ARGS_ASSERT_PREGEXEC;
517 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
518 nosave ? 0 : REXEC_COPY_STR);
523 * Need to implement the following flags for reg_anch:
525 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
527 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
528 * INTUIT_AUTORITATIVE_ML
529 * INTUIT_ONCE_NOML - Intuit can match in one location only.
532 * Another flag for this function: SECOND_TIME (so that float substrs
533 * with giant delta may be not rechecked).
536 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
538 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
539 Otherwise, only SvCUR(sv) is used to get strbeg. */
541 /* XXXX We assume that strpos is strbeg unless sv. */
543 /* XXXX Some places assume that there is a fixed substring.
544 An update may be needed if optimizer marks as "INTUITable"
545 RExen without fixed substrings. Similarly, it is assumed that
546 lengths of all the strings are no more than minlen, thus they
547 cannot come from lookahead.
548 (Or minlen should take into account lookahead.)
549 NOTE: Some of this comment is not correct. minlen does now take account
550 of lookahead/behind. Further research is required. -- demerphq
554 /* A failure to find a constant substring means that there is no need to make
555 an expensive call to REx engine, thus we celebrate a failure. Similarly,
556 finding a substring too deep into the string means that less calls to
557 regtry() should be needed.
559 REx compiler's optimizer found 4 possible hints:
560 a) Anchored substring;
562 c) Whether we are anchored (beginning-of-line or \G);
563 d) First node (of those at offset 0) which may distinguish positions;
564 We use a)b)d) and multiline-part of c), and try to find a position in the
565 string which does not contradict any of them.
568 /* Most of decisions we do here should have been done at compile time.
569 The nodes of the REx which we used for the search should have been
570 deleted from the finite automaton. */
573 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
574 char *strend, const U32 flags, re_scream_pos_data *data)
577 struct regexp *const prog = (struct regexp *)SvANY(rx);
579 /* Should be nonnegative! */
585 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
587 char *other_last = NULL; /* other substr checked before this */
588 char *check_at = NULL; /* check substr found at this pos */
589 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
590 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
591 RXi_GET_DECL(prog,progi);
593 const char * const i_strpos = strpos;
595 GET_RE_DEBUG_FLAGS_DECL;
597 PERL_ARGS_ASSERT_RE_INTUIT_START;
598 PERL_UNUSED_ARG(flags);
599 PERL_UNUSED_ARG(data);
601 RX_MATCH_UTF8_set(rx,utf8_target);
604 PL_reg_flags |= RF_utf8;
607 debug_start_match(rx, utf8_target, strpos, strend,
608 sv ? "Guessing start of match in sv for"
609 : "Guessing start of match in string for");
612 /* CHR_DIST() would be more correct here but it makes things slow. */
613 if (prog->minlen > strend - strpos) {
614 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
615 "String too short... [re_intuit_start]\n"));
619 /* XXX we need to pass strbeg as a separate arg: the following is
620 * guesswork and can be wrong... */
621 if (sv && SvPOK(sv)) {
622 char * p = SvPVX(sv);
623 STRLEN cur = SvCUR(sv);
624 if (p <= strpos && strpos < p + cur) {
626 assert(p <= strend && strend <= p + cur);
629 strbeg = strend - cur;
636 if (!prog->check_utf8 && prog->check_substr)
637 to_utf8_substr(prog);
638 check = prog->check_utf8;
640 if (!prog->check_substr && prog->check_utf8) {
641 if (! to_byte_substr(prog)) {
642 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
643 non_utf8_target_but_utf8_required));
647 check = prog->check_substr;
649 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
650 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
651 || ( (prog->extflags & RXf_ANCH_BOL)
652 && !multiline ) ); /* Check after \n? */
655 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
656 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
657 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
659 && (strpos != strbeg)) {
660 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
663 if (prog->check_offset_min == prog->check_offset_max &&
664 !(prog->extflags & RXf_CANY_SEEN)) {
665 /* Substring at constant offset from beg-of-str... */
668 s = HOP3c(strpos, prog->check_offset_min, strend);
671 slen = SvCUR(check); /* >= 1 */
673 if ( strend - s > slen || strend - s < slen - 1
674 || (strend - s == slen && strend[-1] != '\n')) {
675 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
678 /* Now should match s[0..slen-2] */
680 if (slen && (*SvPVX_const(check) != *s
682 && memNE(SvPVX_const(check), s, slen)))) {
684 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
688 else if (*SvPVX_const(check) != *s
689 || ((slen = SvCUR(check)) > 1
690 && memNE(SvPVX_const(check), s, slen)))
693 goto success_at_start;
696 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
698 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
699 end_shift = prog->check_end_shift;
702 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
703 - (SvTAIL(check) != 0);
704 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
706 if (end_shift < eshift)
710 else { /* Can match at random position */
713 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
714 end_shift = prog->check_end_shift;
716 /* end shift should be non negative here */
719 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
721 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
722 (IV)end_shift, RX_PRECOMP(prog));
726 /* Find a possible match in the region s..strend by looking for
727 the "check" substring in the region corrected by start/end_shift. */
730 I32 srch_start_shift = start_shift;
731 I32 srch_end_shift = end_shift;
734 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
735 srch_end_shift -= ((strbeg - s) - srch_start_shift);
736 srch_start_shift = strbeg - s;
738 DEBUG_OPTIMISE_MORE_r({
739 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
740 (IV)prog->check_offset_min,
741 (IV)srch_start_shift,
743 (IV)prog->check_end_shift);
746 if (prog->extflags & RXf_CANY_SEEN) {
747 start_point= (U8*)(s + srch_start_shift);
748 end_point= (U8*)(strend - srch_end_shift);
750 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
751 end_point= HOP3(strend, -srch_end_shift, strbeg);
753 DEBUG_OPTIMISE_MORE_r({
754 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
755 (int)(end_point - start_point),
756 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
760 s = fbm_instr( start_point, end_point,
761 check, multiline ? FBMrf_MULTILINE : 0);
763 /* Update the count-of-usability, remove useless subpatterns,
767 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
768 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
769 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
770 (s ? "Found" : "Did not find"),
771 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
772 ? "anchored" : "floating"),
775 (s ? " at offset " : "...\n") );
780 /* Finish the diagnostic message */
781 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
783 /* XXX dmq: first branch is for positive lookbehind...
784 Our check string is offset from the beginning of the pattern.
785 So we need to do any stclass tests offset forward from that
794 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
795 Start with the other substr.
796 XXXX no SCREAM optimization yet - and a very coarse implementation
797 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
798 *always* match. Probably should be marked during compile...
799 Probably it is right to do no SCREAM here...
802 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
803 : (prog->float_substr && prog->anchored_substr))
805 /* Take into account the "other" substring. */
806 /* XXXX May be hopelessly wrong for UTF... */
809 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
812 char * const last = HOP3c(s, -start_shift, strbeg);
814 char * const saved_s = s;
817 t = s - prog->check_offset_max;
818 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
820 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
825 t = HOP3c(t, prog->anchored_offset, strend);
826 if (t < other_last) /* These positions already checked */
828 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
831 /* XXXX It is not documented what units *_offsets are in.
832 We assume bytes, but this is clearly wrong.
833 Meaning this code needs to be carefully reviewed for errors.
837 /* On end-of-str: see comment below. */
838 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
839 if (must == &PL_sv_undef) {
841 DEBUG_r(must = prog->anchored_utf8); /* for debug */
846 HOP3(HOP3(last1, prog->anchored_offset, strend)
847 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
849 multiline ? FBMrf_MULTILINE : 0
852 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
853 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
854 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
855 (s ? "Found" : "Contradicts"),
856 quoted, RE_SV_TAIL(must));
861 if (last1 >= last2) {
862 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
863 ", giving up...\n"));
866 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
867 ", trying floating at offset %ld...\n",
868 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
869 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
870 s = HOP3c(last, 1, strend);
874 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
875 (long)(s - i_strpos)));
876 t = HOP3c(s, -prog->anchored_offset, strbeg);
877 other_last = HOP3c(s, 1, strend);
885 else { /* Take into account the floating substring. */
887 char * const saved_s = s;
890 t = HOP3c(s, -start_shift, strbeg);
892 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
893 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
894 last = HOP3c(t, prog->float_max_offset, strend);
895 s = HOP3c(t, prog->float_min_offset, strend);
898 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
899 must = utf8_target ? prog->float_utf8 : prog->float_substr;
900 /* fbm_instr() takes into account exact value of end-of-str
901 if the check is SvTAIL(ed). Since false positives are OK,
902 and end-of-str is not later than strend we are OK. */
903 if (must == &PL_sv_undef) {
905 DEBUG_r(must = prog->float_utf8); /* for debug message */
908 s = fbm_instr((unsigned char*)s,
909 (unsigned char*)last + SvCUR(must)
911 must, multiline ? FBMrf_MULTILINE : 0);
913 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
914 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
915 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
916 (s ? "Found" : "Contradicts"),
917 quoted, RE_SV_TAIL(must));
921 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
922 ", giving up...\n"));
925 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
926 ", trying anchored starting at offset %ld...\n",
927 (long)(saved_s + 1 - i_strpos)));
929 s = HOP3c(t, 1, strend);
933 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
934 (long)(s - i_strpos)));
935 other_last = s; /* Fix this later. --Hugo */
945 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
947 DEBUG_OPTIMISE_MORE_r(
948 PerlIO_printf(Perl_debug_log,
949 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
950 (IV)prog->check_offset_min,
951 (IV)prog->check_offset_max,
959 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
961 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
964 /* Fixed substring is found far enough so that the match
965 cannot start at strpos. */
967 if (ml_anch && t[-1] != '\n') {
968 /* Eventually fbm_*() should handle this, but often
969 anchored_offset is not 0, so this check will not be wasted. */
970 /* XXXX In the code below we prefer to look for "^" even in
971 presence of anchored substrings. And we search even
972 beyond the found float position. These pessimizations
973 are historical artefacts only. */
975 while (t < strend - prog->minlen) {
977 if (t < check_at - prog->check_offset_min) {
978 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
979 /* Since we moved from the found position,
980 we definitely contradict the found anchored
981 substr. Due to the above check we do not
982 contradict "check" substr.
983 Thus we can arrive here only if check substr
984 is float. Redo checking for "other"=="fixed".
987 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
988 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
989 goto do_other_anchored;
991 /* We don't contradict the found floating substring. */
992 /* XXXX Why not check for STCLASS? */
994 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
995 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
998 /* Position contradicts check-string */
999 /* XXXX probably better to look for check-string
1000 than for "\n", so one should lower the limit for t? */
1001 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1002 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1003 other_last = strpos = s = t + 1;
1008 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1009 PL_colors[0], PL_colors[1]));
1013 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1014 PL_colors[0], PL_colors[1]));
1018 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1021 /* The found string does not prohibit matching at strpos,
1022 - no optimization of calling REx engine can be performed,
1023 unless it was an MBOL and we are not after MBOL,
1024 or a future STCLASS check will fail this. */
1026 /* Even in this situation we may use MBOL flag if strpos is offset
1027 wrt the start of the string. */
1028 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1029 && (strpos != strbeg) && strpos[-1] != '\n'
1030 /* May be due to an implicit anchor of m{.*foo} */
1031 && !(prog->intflags & PREGf_IMPLICIT))
1036 DEBUG_EXECUTE_r( if (ml_anch)
1037 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1038 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1041 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1043 prog->check_utf8 /* Could be deleted already */
1044 && --BmUSEFUL(prog->check_utf8) < 0
1045 && (prog->check_utf8 == prog->float_utf8)
1047 prog->check_substr /* Could be deleted already */
1048 && --BmUSEFUL(prog->check_substr) < 0
1049 && (prog->check_substr == prog->float_substr)
1052 /* If flags & SOMETHING - do not do it many times on the same match */
1053 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1054 /* XXX Does the destruction order has to change with utf8_target? */
1055 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1056 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1057 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1058 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1059 check = NULL; /* abort */
1061 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1062 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1063 if (prog->intflags & PREGf_IMPLICIT)
1064 prog->extflags &= ~RXf_ANCH_MBOL;
1065 /* XXXX This is a remnant of the old implementation. It
1066 looks wasteful, since now INTUIT can use many
1067 other heuristics. */
1068 prog->extflags &= ~RXf_USE_INTUIT;
1069 /* XXXX What other flags might need to be cleared in this branch? */
1075 /* Last resort... */
1076 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1077 /* trie stclasses are too expensive to use here, we are better off to
1078 leave it to regmatch itself */
1079 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1080 /* minlen == 0 is possible if regstclass is \b or \B,
1081 and the fixed substr is ''$.
1082 Since minlen is already taken into account, s+1 is before strend;
1083 accidentally, minlen >= 1 guaranties no false positives at s + 1
1084 even for \b or \B. But (minlen? 1 : 0) below assumes that
1085 regstclass does not come from lookahead... */
1086 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1087 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1088 const U8* const str = (U8*)STRING(progi->regstclass);
1089 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1090 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1093 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1094 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1095 else if (prog->float_substr || prog->float_utf8)
1096 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1100 if (checked_upto < s)
1102 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1103 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1106 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1111 const char *what = NULL;
1113 if (endpos == strend) {
1114 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1115 "Could not match STCLASS...\n") );
1118 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1119 "This position contradicts STCLASS...\n") );
1120 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1122 checked_upto = HOPBACKc(endpos, start_shift);
1123 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1124 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1125 /* Contradict one of substrings */
1126 if (prog->anchored_substr || prog->anchored_utf8) {
1127 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1128 DEBUG_EXECUTE_r( what = "anchored" );
1130 s = HOP3c(t, 1, strend);
1131 if (s + start_shift + end_shift > strend) {
1132 /* XXXX Should be taken into account earlier? */
1133 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1134 "Could not match STCLASS...\n") );
1139 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1140 "Looking for %s substr starting at offset %ld...\n",
1141 what, (long)(s + start_shift - i_strpos)) );
1144 /* Have both, check_string is floating */
1145 if (t + start_shift >= check_at) /* Contradicts floating=check */
1146 goto retry_floating_check;
1147 /* Recheck anchored substring, but not floating... */
1151 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1152 "Looking for anchored substr starting at offset %ld...\n",
1153 (long)(other_last - i_strpos)) );
1154 goto do_other_anchored;
1156 /* Another way we could have checked stclass at the
1157 current position only: */
1162 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1163 "Looking for /%s^%s/m starting at offset %ld...\n",
1164 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1167 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1169 /* Check is floating substring. */
1170 retry_floating_check:
1171 t = check_at - start_shift;
1172 DEBUG_EXECUTE_r( what = "floating" );
1173 goto hop_and_restart;
1176 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1177 "By STCLASS: moving %ld --> %ld\n",
1178 (long)(t - i_strpos), (long)(s - i_strpos))
1182 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1183 "Does not contradict STCLASS...\n");
1188 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1189 PL_colors[4], (check ? "Guessed" : "Giving up"),
1190 PL_colors[5], (long)(s - i_strpos)) );
1193 fail_finish: /* Substring not found */
1194 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1195 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1197 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1198 PL_colors[4], PL_colors[5]));
1202 #define DECL_TRIE_TYPE(scan) \
1203 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1204 trie_type = ((scan->flags == EXACT) \
1205 ? (utf8_target ? trie_utf8 : trie_plain) \
1206 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1208 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1209 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1211 switch (trie_type) { \
1212 case trie_utf8_fold: \
1213 if ( foldlen>0 ) { \
1214 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1219 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1220 len = UTF8SKIP(uc); \
1221 skiplen = UNISKIP( uvc ); \
1222 foldlen -= skiplen; \
1223 uscan = foldbuf + skiplen; \
1226 case trie_latin_utf8_fold: \
1227 if ( foldlen>0 ) { \
1228 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1234 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1235 skiplen = UNISKIP( uvc ); \
1236 foldlen -= skiplen; \
1237 uscan = foldbuf + skiplen; \
1241 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1248 charid = trie->charmap[ uvc ]; \
1252 if (widecharmap) { \
1253 SV** const svpp = hv_fetch(widecharmap, \
1254 (char*)&uvc, sizeof(UV), 0); \
1256 charid = (U16)SvIV(*svpp); \
1261 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1265 && (ln == 1 || folder(s, pat_string, ln)) \
1266 && (!reginfo || regtry(reginfo, &s)) ) \
1272 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1274 while (s < strend && s + (uskip = UTF8SKIP(s)) <= strend) { \
1280 #define REXEC_FBC_SCAN(CoDe) \
1282 while (s < strend) { \
1288 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1289 REXEC_FBC_UTF8_SCAN( \
1291 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1300 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1303 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1312 #define REXEC_FBC_TRYIT \
1313 if ((!reginfo || regtry(reginfo, &s))) \
1316 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1317 if (utf8_target) { \
1318 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1321 REXEC_FBC_CLASS_SCAN(CoNd); \
1324 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1325 if (utf8_target) { \
1327 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1330 REXEC_FBC_CLASS_SCAN(CoNd); \
1333 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1334 PL_reg_flags |= RF_tainted; \
1335 if (utf8_target) { \
1336 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1339 REXEC_FBC_CLASS_SCAN(CoNd); \
1342 #define DUMP_EXEC_POS(li,s,doutf8) \
1343 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1346 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1347 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1348 tmp = TEST_NON_UTF8(tmp); \
1349 REXEC_FBC_UTF8_SCAN( \
1350 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1359 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1360 if (s == PL_bostr) { \
1364 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1365 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1368 LOAD_UTF8_CHARCLASS_ALNUM(); \
1369 REXEC_FBC_UTF8_SCAN( \
1370 if (tmp == ! (TeSt2_UtF8)) { \
1379 /* The only difference between the BOUND and NBOUND cases is that
1380 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1381 * NBOUND. This is accomplished by passing it in either the if or else clause,
1382 * with the other one being empty */
1383 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1384 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1386 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1387 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1389 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1390 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1392 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1393 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1396 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1397 * be passed in completely with the variable name being tested, which isn't
1398 * such a clean interface, but this is easier to read than it was before. We
1399 * are looking for the boundary (or non-boundary between a word and non-word
1400 * character. The utf8 and non-utf8 cases have the same logic, but the details
1401 * must be different. Find the "wordness" of the character just prior to this
1402 * one, and compare it with the wordness of this one. If they differ, we have
1403 * a boundary. At the beginning of the string, pretend that the previous
1404 * character was a new-line */
1405 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1406 if (utf8_target) { \
1409 else { /* Not utf8 */ \
1410 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1411 tmp = TEST_NON_UTF8(tmp); \
1413 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1422 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1425 /* We know what class REx starts with. Try to find this position... */
1426 /* if reginfo is NULL, its a dryrun */
1427 /* annoyingly all the vars in this routine have different names from their counterparts
1428 in regmatch. /grrr */
1431 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1432 const char *strend, regmatch_info *reginfo)
1435 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1436 char *pat_string; /* The pattern's exactish string */
1437 char *pat_end; /* ptr to end char of pat_string */
1438 re_fold_t folder; /* Function for computing non-utf8 folds */
1439 const U8 *fold_array; /* array for folding ords < 256 */
1446 I32 tmp = 1; /* Scratch variable? */
1447 const bool utf8_target = PL_reg_match_utf8;
1448 UV utf8_fold_flags = 0;
1449 RXi_GET_DECL(prog,progi);
1451 PERL_ARGS_ASSERT_FIND_BYCLASS;
1453 /* We know what class it must start with. */
1457 if (utf8_target || OP(c) == ANYOFV) {
1458 STRLEN inclasslen = strend - s;
1459 REXEC_FBC_UTF8_CLASS_SCAN(
1460 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1463 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1468 if (tmp && (!reginfo || regtry(reginfo, &s)))
1476 if (UTF_PATTERN || utf8_target) {
1477 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1478 goto do_exactf_utf8;
1480 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1481 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1482 goto do_exactf_non_utf8; /* isn't dealt with by these */
1487 /* regcomp.c already folded this if pattern is in UTF-8 */
1488 utf8_fold_flags = 0;
1489 goto do_exactf_utf8;
1491 fold_array = PL_fold;
1493 goto do_exactf_non_utf8;
1496 if (UTF_PATTERN || utf8_target) {
1497 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1498 goto do_exactf_utf8;
1500 fold_array = PL_fold_locale;
1501 folder = foldEQ_locale;
1502 goto do_exactf_non_utf8;
1506 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1508 goto do_exactf_utf8;
1510 case EXACTFU_TRICKYFOLD:
1512 if (UTF_PATTERN || utf8_target) {
1513 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1514 goto do_exactf_utf8;
1517 /* Any 'ss' in the pattern should have been replaced by regcomp,
1518 * so we don't have to worry here about this single special case
1519 * in the Latin1 range */
1520 fold_array = PL_fold_latin1;
1521 folder = foldEQ_latin1;
1525 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1526 are no glitches with fold-length differences
1527 between the target string and pattern */
1529 /* The idea in the non-utf8 EXACTF* cases is to first find the
1530 * first character of the EXACTF* node and then, if necessary,
1531 * case-insensitively compare the full text of the node. c1 is the
1532 * first character. c2 is its fold. This logic will not work for
1533 * Unicode semantics and the german sharp ss, which hence should
1534 * not be compiled into a node that gets here. */
1535 pat_string = STRING(c);
1536 ln = STR_LEN(c); /* length to match in octets/bytes */
1538 /* We know that we have to match at least 'ln' bytes (which is the
1539 * same as characters, since not utf8). If we have to match 3
1540 * characters, and there are only 2 availabe, we know without
1541 * trying that it will fail; so don't start a match past the
1542 * required minimum number from the far end */
1543 e = HOP3c(strend, -((I32)ln), s);
1545 if (!reginfo && e < s) {
1546 e = s; /* Due to minlen logic of intuit() */
1550 c2 = fold_array[c1];
1551 if (c1 == c2) { /* If char and fold are the same */
1552 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1555 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1564 /* If one of the operands is in utf8, we can't use the simpler
1565 * folding above, due to the fact that many different characters
1566 * can have the same fold, or portion of a fold, or different-
1568 pat_string = STRING(c);
1569 ln = STR_LEN(c); /* length to match in octets/bytes */
1570 pat_end = pat_string + ln;
1571 lnc = (UTF_PATTERN) /* length to match in characters */
1572 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1575 /* We have 'lnc' characters to match in the pattern, but because of
1576 * multi-character folding, each character in the target can match
1577 * up to 3 characters (Unicode guarantees it will never exceed
1578 * this) if it is utf8-encoded; and up to 2 if not (based on the
1579 * fact that the Latin 1 folds are already determined, and the
1580 * only multi-char fold in that range is the sharp-s folding to
1581 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1582 * string character. Adjust lnc accordingly, rounding up, so that
1583 * if we need to match at least 4+1/3 chars, that really is 5. */
1584 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1585 lnc = (lnc + expansion - 1) / expansion;
1587 /* As in the non-UTF8 case, if we have to match 3 characters, and
1588 * only 2 are left, it's guaranteed to fail, so don't start a
1589 * match that would require us to go beyond the end of the string
1591 e = HOP3c(strend, -((I32)lnc), s);
1593 if (!reginfo && e < s) {
1594 e = s; /* Due to minlen logic of intuit() */
1597 /* XXX Note that we could recalculate e to stop the loop earlier,
1598 * as the worst case expansion above will rarely be met, and as we
1599 * go along we would usually find that e moves further to the left.
1600 * This would happen only after we reached the point in the loop
1601 * where if there were no expansion we should fail. Unclear if
1602 * worth the expense */
1605 char *my_strend= (char *)strend;
1606 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1607 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1608 && (!reginfo || regtry(reginfo, &s)) )
1612 s += (utf8_target) ? UTF8SKIP(s) : 1;
1617 PL_reg_flags |= RF_tainted;
1618 FBC_BOUND(isALNUM_LC,
1619 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1620 isALNUM_LC_utf8((U8*)s));
1623 PL_reg_flags |= RF_tainted;
1624 FBC_NBOUND(isALNUM_LC,
1625 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1626 isALNUM_LC_utf8((U8*)s));
1629 FBC_BOUND(isWORDCHAR,
1631 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1634 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1636 isWORDCHAR_A((U8*)s));
1639 FBC_NBOUND(isWORDCHAR,
1641 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1644 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1646 isWORDCHAR_A((U8*)s));
1649 FBC_BOUND(isWORDCHAR_L1,
1651 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1654 FBC_NBOUND(isWORDCHAR_L1,
1656 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1659 REXEC_FBC_CSCAN_TAINT(
1660 isALNUM_LC_utf8((U8*)s),
1665 REXEC_FBC_CSCAN_PRELOAD(
1666 LOAD_UTF8_CHARCLASS_ALNUM(),
1667 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1668 isWORDCHAR_L1((U8) *s)
1672 REXEC_FBC_CSCAN_PRELOAD(
1673 LOAD_UTF8_CHARCLASS_ALNUM(),
1674 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1679 /* Don't need to worry about utf8, as it can match only a single
1680 * byte invariant character */
1681 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1684 REXEC_FBC_CSCAN_PRELOAD(
1685 LOAD_UTF8_CHARCLASS_ALNUM(),
1686 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1687 ! isWORDCHAR_L1((U8) *s)
1691 REXEC_FBC_CSCAN_PRELOAD(
1692 LOAD_UTF8_CHARCLASS_ALNUM(),
1693 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1704 REXEC_FBC_CSCAN_TAINT(
1705 !isALNUM_LC_utf8((U8*)s),
1710 REXEC_FBC_CSCAN_PRELOAD(
1711 LOAD_UTF8_CHARCLASS_SPACE(),
1712 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1717 REXEC_FBC_CSCAN_PRELOAD(
1718 LOAD_UTF8_CHARCLASS_SPACE(),
1719 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1724 /* Don't need to worry about utf8, as it can match only a single
1725 * byte invariant character */
1726 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1729 REXEC_FBC_CSCAN_TAINT(
1730 isSPACE_LC_utf8((U8*)s),
1735 REXEC_FBC_CSCAN_PRELOAD(
1736 LOAD_UTF8_CHARCLASS_SPACE(),
1737 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1738 ! isSPACE_L1((U8) *s)
1742 REXEC_FBC_CSCAN_PRELOAD(
1743 LOAD_UTF8_CHARCLASS_SPACE(),
1744 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1755 REXEC_FBC_CSCAN_TAINT(
1756 !isSPACE_LC_utf8((U8*)s),
1761 REXEC_FBC_CSCAN_PRELOAD(
1762 LOAD_UTF8_CHARCLASS_DIGIT(),
1763 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1768 /* Don't need to worry about utf8, as it can match only a single
1769 * byte invariant character */
1770 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1773 REXEC_FBC_CSCAN_TAINT(
1774 isDIGIT_LC_utf8((U8*)s),
1779 REXEC_FBC_CSCAN_PRELOAD(
1780 LOAD_UTF8_CHARCLASS_DIGIT(),
1781 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1792 REXEC_FBC_CSCAN_TAINT(
1793 !isDIGIT_LC_utf8((U8*)s),
1799 is_LNBREAK_utf8_safe(s, strend),
1800 is_LNBREAK_latin1_safe(s, strend)
1805 is_VERTWS_utf8_safe(s, strend),
1806 is_VERTWS_latin1_safe(s, strend)
1811 !is_VERTWS_utf8_safe(s, strend),
1812 !is_VERTWS_latin1_safe(s, strend)
1817 is_HORIZWS_utf8_safe(s, strend),
1818 is_HORIZWS_latin1_safe(s, strend)
1823 !is_HORIZWS_utf8_safe(s, strend),
1824 !is_HORIZWS_latin1_safe(s, strend)
1828 /* Don't need to worry about utf8, as it can match only a single
1829 * byte invariant character. The flag in this node type is the
1830 * class number to pass to _generic_isCC() to build a mask for
1831 * searching in PL_charclass[] */
1832 REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
1836 !_generic_isCC_A(*s, FLAGS(c)),
1837 !_generic_isCC_A(*s, FLAGS(c))
1845 /* what trie are we using right now */
1847 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1849 = (reg_trie_data*)progi->data->data[ aho->trie ];
1850 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1852 const char *last_start = strend - trie->minlen;
1854 const char *real_start = s;
1856 STRLEN maxlen = trie->maxlen;
1858 U8 **points; /* map of where we were in the input string
1859 when reading a given char. For ASCII this
1860 is unnecessary overhead as the relationship
1861 is always 1:1, but for Unicode, especially
1862 case folded Unicode this is not true. */
1863 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1867 GET_RE_DEBUG_FLAGS_DECL;
1869 /* We can't just allocate points here. We need to wrap it in
1870 * an SV so it gets freed properly if there is a croak while
1871 * running the match */
1874 sv_points=newSV(maxlen * sizeof(U8 *));
1875 SvCUR_set(sv_points,
1876 maxlen * sizeof(U8 *));
1877 SvPOK_on(sv_points);
1878 sv_2mortal(sv_points);
1879 points=(U8**)SvPV_nolen(sv_points );
1880 if ( trie_type != trie_utf8_fold
1881 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1884 bitmap=(U8*)trie->bitmap;
1886 bitmap=(U8*)ANYOF_BITMAP(c);
1888 /* this is the Aho-Corasick algorithm modified a touch
1889 to include special handling for long "unknown char"
1890 sequences. The basic idea being that we use AC as long
1891 as we are dealing with a possible matching char, when
1892 we encounter an unknown char (and we have not encountered
1893 an accepting state) we scan forward until we find a legal
1895 AC matching is basically that of trie matching, except
1896 that when we encounter a failing transition, we fall back
1897 to the current states "fail state", and try the current char
1898 again, a process we repeat until we reach the root state,
1899 state 1, or a legal transition. If we fail on the root state
1900 then we can either terminate if we have reached an accepting
1901 state previously, or restart the entire process from the beginning
1905 while (s <= last_start) {
1906 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1914 U8 *uscan = (U8*)NULL;
1915 U8 *leftmost = NULL;
1917 U32 accepted_word= 0;
1921 while ( state && uc <= (U8*)strend ) {
1923 U32 word = aho->states[ state ].wordnum;
1927 DEBUG_TRIE_EXECUTE_r(
1928 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1929 dump_exec_pos( (char *)uc, c, strend, real_start,
1930 (char *)uc, utf8_target );
1931 PerlIO_printf( Perl_debug_log,
1932 " Scanning for legal start char...\n");
1936 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1940 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1946 if (uc >(U8*)last_start) break;
1950 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1951 if (!leftmost || lpos < leftmost) {
1952 DEBUG_r(accepted_word=word);
1958 points[pointpos++ % maxlen]= uc;
1959 if (foldlen || uc < (U8*)strend) {
1960 REXEC_TRIE_READ_CHAR(trie_type, trie,
1962 uscan, len, uvc, charid, foldlen,
1964 DEBUG_TRIE_EXECUTE_r({
1965 dump_exec_pos( (char *)uc, c, strend,
1966 real_start, s, utf8_target);
1967 PerlIO_printf(Perl_debug_log,
1968 " Charid:%3u CP:%4"UVxf" ",
1980 word = aho->states[ state ].wordnum;
1982 base = aho->states[ state ].trans.base;
1984 DEBUG_TRIE_EXECUTE_r({
1986 dump_exec_pos( (char *)uc, c, strend, real_start,
1988 PerlIO_printf( Perl_debug_log,
1989 "%sState: %4"UVxf", word=%"UVxf,
1990 failed ? " Fail transition to " : "",
1991 (UV)state, (UV)word);
1997 ( ((offset = base + charid
1998 - 1 - trie->uniquecharcount)) >= 0)
1999 && ((U32)offset < trie->lasttrans)
2000 && trie->trans[offset].check == state
2001 && (tmp=trie->trans[offset].next))
2003 DEBUG_TRIE_EXECUTE_r(
2004 PerlIO_printf( Perl_debug_log," - legal\n"));
2009 DEBUG_TRIE_EXECUTE_r(
2010 PerlIO_printf( Perl_debug_log," - fail\n"));
2012 state = aho->fail[state];
2016 /* we must be accepting here */
2017 DEBUG_TRIE_EXECUTE_r(
2018 PerlIO_printf( Perl_debug_log," - accepting\n"));
2027 if (!state) state = 1;
2030 if ( aho->states[ state ].wordnum ) {
2031 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2032 if (!leftmost || lpos < leftmost) {
2033 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2038 s = (char*)leftmost;
2039 DEBUG_TRIE_EXECUTE_r({
2041 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2042 (UV)accepted_word, (IV)(s - real_start)
2045 if (!reginfo || regtry(reginfo, &s)) {
2051 DEBUG_TRIE_EXECUTE_r({
2052 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2055 DEBUG_TRIE_EXECUTE_r(
2056 PerlIO_printf( Perl_debug_log,"No match.\n"));
2065 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2075 - regexec_flags - match a regexp against a string
2078 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2079 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2080 /* stringarg: the point in the string at which to begin matching */
2081 /* strend: pointer to null at end of string */
2082 /* strbeg: real beginning of string */
2083 /* minend: end of match must be >= minend bytes after stringarg. */
2084 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2085 * itself is accessed via the pointers above */
2086 /* data: May be used for some additional optimizations.
2087 Currently its only used, with a U32 cast, for transmitting
2088 the ganch offset when doing a /g match. This will change */
2089 /* nosave: For optimizations. */
2093 struct regexp *const prog = (struct regexp *)SvANY(rx);
2094 /*register*/ char *s;
2096 /*register*/ char *startpos = stringarg;
2097 I32 minlen; /* must match at least this many chars */
2098 I32 dontbother = 0; /* how many characters not to try at end */
2099 I32 end_shift = 0; /* Same for the end. */ /* CC */
2100 I32 scream_pos = -1; /* Internal iterator of scream. */
2101 char *scream_olds = NULL;
2102 const bool utf8_target = cBOOL(DO_UTF8(sv));
2104 RXi_GET_DECL(prog,progi);
2105 regmatch_info reginfo; /* create some info to pass to regtry etc */
2106 regexp_paren_pair *swap = NULL;
2107 GET_RE_DEBUG_FLAGS_DECL;
2109 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2110 PERL_UNUSED_ARG(data);
2112 /* Be paranoid... */
2113 if (prog == NULL || startpos == NULL) {
2114 Perl_croak(aTHX_ "NULL regexp parameter");
2118 multiline = prog->extflags & RXf_PMf_MULTILINE;
2119 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2121 RX_MATCH_UTF8_set(rx, utf8_target);
2123 debug_start_match(rx, utf8_target, startpos, strend,
2127 minlen = prog->minlen;
2129 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2130 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2131 "String too short [regexec_flags]...\n"));
2136 /* Check validity of program. */
2137 if (UCHARAT(progi->program) != REG_MAGIC) {
2138 Perl_croak(aTHX_ "corrupted regexp program");
2142 PL_reg_state.re_state_eval_setup_done = FALSE;
2146 PL_reg_flags |= RF_utf8;
2148 /* Mark beginning of line for ^ and lookbehind. */
2149 reginfo.bol = startpos; /* XXX not used ??? */
2153 /* Mark end of line for $ (and such) */
2156 /* see how far we have to get to not match where we matched before */
2157 reginfo.till = startpos+minend;
2159 /* If there is a "must appear" string, look for it. */
2162 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2164 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2165 reginfo.ganch = startpos + prog->gofs;
2166 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2167 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2168 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2170 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2171 && mg->mg_len >= 0) {
2172 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2173 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2174 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2176 if (prog->extflags & RXf_ANCH_GPOS) {
2177 if (s > reginfo.ganch)
2179 s = reginfo.ganch - prog->gofs;
2180 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2181 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2187 reginfo.ganch = strbeg + PTR2UV(data);
2188 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2189 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2191 } else { /* pos() not defined */
2192 reginfo.ganch = strbeg;
2193 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2194 "GPOS: reginfo.ganch = strbeg\n"));
2197 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2198 /* We have to be careful. If the previous successful match
2199 was from this regex we don't want a subsequent partially
2200 successful match to clobber the old results.
2201 So when we detect this possibility we add a swap buffer
2202 to the re, and switch the buffer each match. If we fail
2203 we switch it back, otherwise we leave it swapped.
2206 /* do we need a save destructor here for eval dies? */
2207 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2208 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2209 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2215 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2216 re_scream_pos_data d;
2218 d.scream_olds = &scream_olds;
2219 d.scream_pos = &scream_pos;
2220 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2222 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2223 goto phooey; /* not present */
2229 /* Simplest case: anchored match need be tried only once. */
2230 /* [unless only anchor is BOL and multiline is set] */
2231 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2232 if (s == startpos && regtry(®info, &startpos))
2234 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2235 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2240 dontbother = minlen - 1;
2241 end = HOP3c(strend, -dontbother, strbeg) - 1;
2242 /* for multiline we only have to try after newlines */
2243 if (prog->check_substr || prog->check_utf8) {
2244 /* because of the goto we can not easily reuse the macros for bifurcating the
2245 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2248 goto after_try_utf8;
2250 if (regtry(®info, &s)) {
2257 if (prog->extflags & RXf_USE_INTUIT) {
2258 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2267 } /* end search for check string in unicode */
2269 if (s == startpos) {
2270 goto after_try_latin;
2273 if (regtry(®info, &s)) {
2280 if (prog->extflags & RXf_USE_INTUIT) {
2281 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2290 } /* end search for check string in latin*/
2291 } /* end search for check string */
2292 else { /* search for newline */
2294 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2297 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2298 while (s <= end) { /* note it could be possible to match at the end of the string */
2299 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2300 if (regtry(®info, &s))
2304 } /* end search for newline */
2305 } /* end anchored/multiline check string search */
2307 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2309 /* the warning about reginfo.ganch being used without initialization
2310 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2311 and we only enter this block when the same bit is set. */
2312 char *tmp_s = reginfo.ganch - prog->gofs;
2314 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2319 /* Messy cases: unanchored match. */
2320 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2321 /* we have /x+whatever/ */
2322 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2328 if (! prog->anchored_utf8) {
2329 to_utf8_substr(prog);
2331 ch = SvPVX_const(prog->anchored_utf8)[0];
2334 DEBUG_EXECUTE_r( did_match = 1 );
2335 if (regtry(®info, &s)) goto got_it;
2337 while (s < strend && *s == ch)
2344 if (! prog->anchored_substr) {
2345 if (! to_byte_substr(prog)) {
2346 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2347 non_utf8_target_but_utf8_required));
2351 ch = SvPVX_const(prog->anchored_substr)[0];
2354 DEBUG_EXECUTE_r( did_match = 1 );
2355 if (regtry(®info, &s)) goto got_it;
2357 while (s < strend && *s == ch)
2362 DEBUG_EXECUTE_r(if (!did_match)
2363 PerlIO_printf(Perl_debug_log,
2364 "Did not find anchored character...\n")
2367 else if (prog->anchored_substr != NULL
2368 || prog->anchored_utf8 != NULL
2369 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2370 && prog->float_max_offset < strend - s)) {
2375 char *last1; /* Last position checked before */
2379 if (prog->anchored_substr || prog->anchored_utf8) {
2381 if (! prog->anchored_utf8) {
2382 to_utf8_substr(prog);
2384 must = prog->anchored_utf8;
2387 if (! prog->anchored_substr) {
2388 if (! to_byte_substr(prog)) {
2389 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2390 non_utf8_target_but_utf8_required));
2394 must = prog->anchored_substr;
2396 back_max = back_min = prog->anchored_offset;
2399 if (! prog->float_utf8) {
2400 to_utf8_substr(prog);
2402 must = prog->float_utf8;
2405 if (! prog->float_substr) {
2406 if (! to_byte_substr(prog)) {
2407 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2408 non_utf8_target_but_utf8_required));
2412 must = prog->float_substr;
2414 back_max = prog->float_max_offset;
2415 back_min = prog->float_min_offset;
2421 last = HOP3c(strend, /* Cannot start after this */
2422 -(I32)(CHR_SVLEN(must)
2423 - (SvTAIL(must) != 0) + back_min), strbeg);
2426 last1 = HOPc(s, -1);
2428 last1 = s - 1; /* bogus */
2430 /* XXXX check_substr already used to find "s", can optimize if
2431 check_substr==must. */
2433 dontbother = end_shift;
2434 strend = HOPc(strend, -dontbother);
2435 while ( (s <= last) &&
2436 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2437 (unsigned char*)strend, must,
2438 multiline ? FBMrf_MULTILINE : 0)) ) {
2439 DEBUG_EXECUTE_r( did_match = 1 );
2440 if (HOPc(s, -back_max) > last1) {
2441 last1 = HOPc(s, -back_min);
2442 s = HOPc(s, -back_max);
2445 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2447 last1 = HOPc(s, -back_min);
2451 while (s <= last1) {
2452 if (regtry(®info, &s))
2455 s++; /* to break out of outer loop */
2462 while (s <= last1) {
2463 if (regtry(®info, &s))
2469 DEBUG_EXECUTE_r(if (!did_match) {
2470 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2471 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2472 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2473 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2474 ? "anchored" : "floating"),
2475 quoted, RE_SV_TAIL(must));
2479 else if ( (c = progi->regstclass) ) {
2481 const OPCODE op = OP(progi->regstclass);
2482 /* don't bother with what can't match */
2483 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2484 strend = HOPc(strend, -(minlen - 1));
2487 SV * const prop = sv_newmortal();
2488 regprop(prog, prop, c);
2490 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2492 PerlIO_printf(Perl_debug_log,
2493 "Matching stclass %.*s against %s (%d bytes)\n",
2494 (int)SvCUR(prop), SvPVX_const(prop),
2495 quoted, (int)(strend - s));
2498 if (find_byclass(prog, c, s, strend, ®info))
2500 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2504 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2512 if (! prog->float_utf8) {
2513 to_utf8_substr(prog);
2515 float_real = prog->float_utf8;
2518 if (! prog->float_substr) {
2519 if (! to_byte_substr(prog)) {
2520 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2521 non_utf8_target_but_utf8_required));
2525 float_real = prog->float_substr;
2528 little = SvPV_const(float_real, len);
2529 if (SvTAIL(float_real)) {
2530 /* This means that float_real contains an artificial \n on
2531 * the end due to the presence of something like this:
2532 * /foo$/ where we can match both "foo" and "foo\n" at the
2533 * end of the string. So we have to compare the end of the
2534 * string first against the float_real without the \n and
2535 * then against the full float_real with the string. We
2536 * have to watch out for cases where the string might be
2537 * smaller than the float_real or the float_real without
2539 char *checkpos= strend - len;
2541 PerlIO_printf(Perl_debug_log,
2542 "%sChecking for float_real.%s\n",
2543 PL_colors[4], PL_colors[5]));
2544 if (checkpos + 1 < strbeg) {
2545 /* can't match, even if we remove the trailing \n
2546 * string is too short to match */
2548 PerlIO_printf(Perl_debug_log,
2549 "%sString shorter than required trailing substring, cannot match.%s\n",
2550 PL_colors[4], PL_colors[5]));
2552 } else if (memEQ(checkpos + 1, little, len - 1)) {
2553 /* can match, the end of the string matches without the
2555 last = checkpos + 1;
2556 } else if (checkpos < strbeg) {
2557 /* cant match, string is too short when the "\n" is
2560 PerlIO_printf(Perl_debug_log,
2561 "%sString does not contain required trailing substring, cannot match.%s\n",
2562 PL_colors[4], PL_colors[5]));
2564 } else if (!multiline) {
2565 /* non multiline match, so compare with the "\n" at the
2566 * end of the string */
2567 if (memEQ(checkpos, little, len)) {
2571 PerlIO_printf(Perl_debug_log,
2572 "%sString does not contain required trailing substring, cannot match.%s\n",
2573 PL_colors[4], PL_colors[5]));
2577 /* multiline match, so we have to search for a place
2578 * where the full string is located */
2584 last = rninstr(s, strend, little, little + len);
2586 last = strend; /* matching "$" */
2589 /* at one point this block contained a comment which was
2590 * probably incorrect, which said that this was a "should not
2591 * happen" case. Even if it was true when it was written I am
2592 * pretty sure it is not anymore, so I have removed the comment
2593 * and replaced it with this one. Yves */
2595 PerlIO_printf(Perl_debug_log,
2596 "String does not contain required substring, cannot match.\n"
2600 dontbother = strend - last + prog->float_min_offset;
2602 if (minlen && (dontbother < minlen))
2603 dontbother = minlen - 1;
2604 strend -= dontbother; /* this one's always in bytes! */
2605 /* We don't know much -- general case. */
2608 if (regtry(®info, &s))
2617 if (regtry(®info, &s))
2619 } while (s++ < strend);
2629 PerlIO_printf(Perl_debug_log,
2630 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2636 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2638 if (PL_reg_state.re_state_eval_setup_done)
2639 restore_pos(aTHX_ prog);
2640 if (RXp_PAREN_NAMES(prog))
2641 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2643 /* make sure $`, $&, $', and $digit will work later */
2644 if ( !(flags & REXEC_NOT_FIRST) ) {
2645 if (flags & REXEC_COPY_STR) {
2646 #ifdef PERL_OLD_COPY_ON_WRITE
2648 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2650 PerlIO_printf(Perl_debug_log,
2651 "Copy on write: regexp capture, type %d\n",
2654 RX_MATCH_COPY_FREE(rx);
2655 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2656 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2657 assert (SvPOKp(prog->saved_copy));
2658 prog->sublen = PL_regeol - strbeg;
2659 prog->suboffset = 0;
2660 prog->subcoffset = 0;
2665 I32 max = PL_regeol - strbeg;
2668 if ( (flags & REXEC_COPY_SKIP_POST)
2669 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2670 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2671 ) { /* don't copy $' part of string */
2674 /* calculate the right-most part of the string covered
2675 * by a capture. Due to look-ahead, this may be to
2676 * the right of $&, so we have to scan all captures */
2677 while (n <= prog->lastparen) {
2678 if (prog->offs[n].end > max)
2679 max = prog->offs[n].end;
2683 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2684 ? prog->offs[0].start
2686 assert(max >= 0 && max <= PL_regeol - strbeg);
2689 if ( (flags & REXEC_COPY_SKIP_PRE)
2690 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2691 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2692 ) { /* don't copy $` part of string */
2695 /* calculate the left-most part of the string covered
2696 * by a capture. Due to look-behind, this may be to
2697 * the left of $&, so we have to scan all captures */
2698 while (min && n <= prog->lastparen) {
2699 if ( prog->offs[n].start != -1
2700 && prog->offs[n].start < min)
2702 min = prog->offs[n].start;
2706 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2707 && min > prog->offs[0].end
2709 min = prog->offs[0].end;
2713 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2716 if (RX_MATCH_COPIED(rx)) {
2717 if (sublen > prog->sublen)
2719 (char*)saferealloc(prog->subbeg, sublen+1);
2722 prog->subbeg = (char*)safemalloc(sublen+1);
2723 Copy(strbeg + min, prog->subbeg, sublen, char);
2724 prog->subbeg[sublen] = '\0';
2725 prog->suboffset = min;
2726 prog->sublen = sublen;
2727 RX_MATCH_COPIED_on(rx);
2729 prog->subcoffset = prog->suboffset;
2730 if (prog->suboffset && utf8_target) {
2731 /* Convert byte offset to chars.
2732 * XXX ideally should only compute this if @-/@+
2733 * has been seen, a la PL_sawampersand ??? */
2735 /* If there's a direct correspondence between the
2736 * string which we're matching and the original SV,
2737 * then we can use the utf8 len cache associated with
2738 * the SV. In particular, it means that under //g,
2739 * sv_pos_b2u() will use the previously cached
2740 * position to speed up working out the new length of
2741 * subcoffset, rather than counting from the start of
2742 * the string each time. This stops
2743 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2744 * from going quadratic */
2745 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2746 sv_pos_b2u(sv, &(prog->subcoffset));
2748 prog->subcoffset = utf8_length((U8*)strbeg,
2749 (U8*)(strbeg+prog->suboffset));
2753 RX_MATCH_COPY_FREE(rx);
2754 prog->subbeg = strbeg;
2755 prog->suboffset = 0;
2756 prog->subcoffset = 0;
2757 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2764 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2765 PL_colors[4], PL_colors[5]));
2766 if (PL_reg_state.re_state_eval_setup_done)
2767 restore_pos(aTHX_ prog);
2769 /* we failed :-( roll it back */
2770 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2771 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2776 Safefree(prog->offs);
2783 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2784 * Do inc before dec, in case old and new rex are the same */
2785 #define SET_reg_curpm(Re2) \
2786 if (PL_reg_state.re_state_eval_setup_done) { \
2787 (void)ReREFCNT_inc(Re2); \
2788 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2789 PM_SETRE((PL_reg_curpm), (Re2)); \
2794 - regtry - try match at specific point
2796 STATIC I32 /* 0 failure, 1 success */
2797 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2801 REGEXP *const rx = reginfo->prog;
2802 regexp *const prog = (struct regexp *)SvANY(rx);
2804 RXi_GET_DECL(prog,progi);
2805 GET_RE_DEBUG_FLAGS_DECL;
2807 PERL_ARGS_ASSERT_REGTRY;
2809 reginfo->cutpoint=NULL;
2811 if ((prog->extflags & RXf_EVAL_SEEN)
2812 && !PL_reg_state.re_state_eval_setup_done)
2816 PL_reg_state.re_state_eval_setup_done = TRUE;
2818 /* Make $_ available to executed code. */
2819 if (reginfo->sv != DEFSV) {
2821 DEFSV_set(reginfo->sv);
2824 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2825 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2826 /* prepare for quick setting of pos */
2827 #ifdef PERL_OLD_COPY_ON_WRITE
2828 if (SvIsCOW(reginfo->sv))
2829 sv_force_normal_flags(reginfo->sv, 0);
2831 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2832 &PL_vtbl_mglob, NULL, 0);
2836 PL_reg_oldpos = mg->mg_len;
2837 SAVEDESTRUCTOR_X(restore_pos, prog);
2839 if (!PL_reg_curpm) {
2840 Newxz(PL_reg_curpm, 1, PMOP);
2843 SV* const repointer = &PL_sv_undef;
2844 /* this regexp is also owned by the new PL_reg_curpm, which
2845 will try to free it. */
2846 av_push(PL_regex_padav, repointer);
2847 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2848 PL_regex_pad = AvARRAY(PL_regex_padav);
2853 PL_reg_oldcurpm = PL_curpm;
2854 PL_curpm = PL_reg_curpm;
2855 if (RXp_MATCH_COPIED(prog)) {
2856 /* Here is a serious problem: we cannot rewrite subbeg,
2857 since it may be needed if this match fails. Thus
2858 $` inside (?{}) could fail... */
2859 PL_reg_oldsaved = prog->subbeg;
2860 PL_reg_oldsavedlen = prog->sublen;
2861 PL_reg_oldsavedoffset = prog->suboffset;
2862 PL_reg_oldsavedcoffset = prog->suboffset;
2863 #ifdef PERL_OLD_COPY_ON_WRITE
2864 PL_nrs = prog->saved_copy;
2866 RXp_MATCH_COPIED_off(prog);
2869 PL_reg_oldsaved = NULL;
2870 prog->subbeg = PL_bostr;
2871 prog->suboffset = 0;
2872 prog->subcoffset = 0;
2873 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2876 PL_reg_starttry = *startposp;
2878 prog->offs[0].start = *startposp - PL_bostr;
2879 prog->lastparen = 0;
2880 prog->lastcloseparen = 0;
2883 /* XXXX What this code is doing here?!!! There should be no need
2884 to do this again and again, prog->lastparen should take care of
2887 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2888 * Actually, the code in regcppop() (which Ilya may be meaning by
2889 * prog->lastparen), is not needed at all by the test suite
2890 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2891 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2892 * Meanwhile, this code *is* needed for the
2893 * above-mentioned test suite tests to succeed. The common theme
2894 * on those tests seems to be returning null fields from matches.
2895 * --jhi updated by dapm */
2897 if (prog->nparens) {
2898 regexp_paren_pair *pp = prog->offs;
2900 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2908 result = regmatch(reginfo, *startposp, progi->program + 1);
2910 prog->offs[0].end = result;
2913 if (reginfo->cutpoint)
2914 *startposp= reginfo->cutpoint;
2915 REGCP_UNWIND(lastcp);
2920 #define sayYES goto yes
2921 #define sayNO goto no
2922 #define sayNO_SILENT goto no_silent
2924 /* we dont use STMT_START/END here because it leads to
2925 "unreachable code" warnings, which are bogus, but distracting. */
2926 #define CACHEsayNO \
2927 if (ST.cache_mask) \
2928 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2931 /* this is used to determine how far from the left messages like
2932 'failed...' are printed. It should be set such that messages
2933 are inline with the regop output that created them.
2935 #define REPORT_CODE_OFF 32
2938 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2939 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2941 #define SLAB_FIRST(s) (&(s)->states[0])
2942 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2944 /* grab a new slab and return the first slot in it */
2946 STATIC regmatch_state *
2949 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2952 regmatch_slab *s = PL_regmatch_slab->next;
2954 Newx(s, 1, regmatch_slab);
2955 s->prev = PL_regmatch_slab;
2957 PL_regmatch_slab->next = s;
2959 PL_regmatch_slab = s;
2960 return SLAB_FIRST(s);
2964 /* push a new state then goto it */
2966 #define PUSH_STATE_GOTO(state, node, input) \
2967 pushinput = input; \
2969 st->resume_state = state; \
2972 /* push a new state with success backtracking, then goto it */
2974 #define PUSH_YES_STATE_GOTO(state, node, input) \
2975 pushinput = input; \
2977 st->resume_state = state; \
2978 goto push_yes_state;
2985 regmatch() - main matching routine
2987 This is basically one big switch statement in a loop. We execute an op,
2988 set 'next' to point the next op, and continue. If we come to a point which
2989 we may need to backtrack to on failure such as (A|B|C), we push a
2990 backtrack state onto the backtrack stack. On failure, we pop the top
2991 state, and re-enter the loop at the state indicated. If there are no more
2992 states to pop, we return failure.
2994 Sometimes we also need to backtrack on success; for example /A+/, where
2995 after successfully matching one A, we need to go back and try to
2996 match another one; similarly for lookahead assertions: if the assertion
2997 completes successfully, we backtrack to the state just before the assertion
2998 and then carry on. In these cases, the pushed state is marked as
2999 'backtrack on success too'. This marking is in fact done by a chain of
3000 pointers, each pointing to the previous 'yes' state. On success, we pop to
3001 the nearest yes state, discarding any intermediate failure-only states.
3002 Sometimes a yes state is pushed just to force some cleanup code to be
3003 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3004 it to free the inner regex.
3006 Note that failure backtracking rewinds the cursor position, while
3007 success backtracking leaves it alone.
3009 A pattern is complete when the END op is executed, while a subpattern
3010 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3011 ops trigger the "pop to last yes state if any, otherwise return true"
3014 A common convention in this function is to use A and B to refer to the two
3015 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3016 the subpattern to be matched possibly multiple times, while B is the entire
3017 rest of the pattern. Variable and state names reflect this convention.
3019 The states in the main switch are the union of ops and failure/success of
3020 substates associated with with that op. For example, IFMATCH is the op
3021 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3022 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3023 successfully matched A and IFMATCH_A_fail is a state saying that we have
3024 just failed to match A. Resume states always come in pairs. The backtrack
3025 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3026 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3027 on success or failure.
3029 The struct that holds a backtracking state is actually a big union, with
3030 one variant for each major type of op. The variable st points to the
3031 top-most backtrack struct. To make the code clearer, within each
3032 block of code we #define ST to alias the relevant union.
3034 Here's a concrete example of a (vastly oversimplified) IFMATCH
3040 #define ST st->u.ifmatch
3042 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3043 ST.foo = ...; // some state we wish to save
3045 // push a yes backtrack state with a resume value of
3046 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3048 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3051 case IFMATCH_A: // we have successfully executed A; now continue with B
3053 bar = ST.foo; // do something with the preserved value
3056 case IFMATCH_A_fail: // A failed, so the assertion failed
3057 ...; // do some housekeeping, then ...
3058 sayNO; // propagate the failure
3065 For any old-timers reading this who are familiar with the old recursive
3066 approach, the code above is equivalent to:
3068 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3077 ...; // do some housekeeping, then ...
3078 sayNO; // propagate the failure
3081 The topmost backtrack state, pointed to by st, is usually free. If you
3082 want to claim it, populate any ST.foo fields in it with values you wish to
3083 save, then do one of
3085 PUSH_STATE_GOTO(resume_state, node, newinput);
3086 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3088 which sets that backtrack state's resume value to 'resume_state', pushes a
3089 new free entry to the top of the backtrack stack, then goes to 'node'.
3090 On backtracking, the free slot is popped, and the saved state becomes the
3091 new free state. An ST.foo field in this new top state can be temporarily
3092 accessed to retrieve values, but once the main loop is re-entered, it
3093 becomes available for reuse.
3095 Note that the depth of the backtrack stack constantly increases during the
3096 left-to-right execution of the pattern, rather than going up and down with
3097 the pattern nesting. For example the stack is at its maximum at Z at the
3098 end of the pattern, rather than at X in the following:
3100 /(((X)+)+)+....(Y)+....Z/
3102 The only exceptions to this are lookahead/behind assertions and the cut,
3103 (?>A), which pop all the backtrack states associated with A before
3106 Backtrack state structs are allocated in slabs of about 4K in size.
3107 PL_regmatch_state and st always point to the currently active state,
3108 and PL_regmatch_slab points to the slab currently containing
3109 PL_regmatch_state. The first time regmatch() is called, the first slab is
3110 allocated, and is never freed until interpreter destruction. When the slab
3111 is full, a new one is allocated and chained to the end. At exit from
3112 regmatch(), slabs allocated since entry are freed.
3117 #define DEBUG_STATE_pp(pp) \
3119 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3120 PerlIO_printf(Perl_debug_log, \
3121 " %*s"pp" %s%s%s%s%s\n", \
3123 PL_reg_name[st->resume_state], \
3124 ((st==yes_state||st==mark_state) ? "[" : ""), \
3125 ((st==yes_state) ? "Y" : ""), \
3126 ((st==mark_state) ? "M" : ""), \
3127 ((st==yes_state||st==mark_state) ? "]" : "") \
3132 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3137 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3138 const char *start, const char *end, const char *blurb)
3140 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3142 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3147 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3148 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3150 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3151 start, end - start, 60);
3153 PerlIO_printf(Perl_debug_log,
3154 "%s%s REx%s %s against %s\n",
3155 PL_colors[4], blurb, PL_colors[5], s0, s1);
3157 if (utf8_target||utf8_pat)
3158 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3159 utf8_pat ? "pattern" : "",
3160 utf8_pat && utf8_target ? " and " : "",
3161 utf8_target ? "string" : ""
3167 S_dump_exec_pos(pTHX_ const char *locinput,
3168 const regnode *scan,
3169 const char *loc_regeol,
3170 const char *loc_bostr,
3171 const char *loc_reg_starttry,
3172 const bool utf8_target)
3174 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3175 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3176 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3177 /* The part of the string before starttry has one color
3178 (pref0_len chars), between starttry and current
3179 position another one (pref_len - pref0_len chars),
3180 after the current position the third one.
3181 We assume that pref0_len <= pref_len, otherwise we
3182 decrease pref0_len. */
3183 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3184 ? (5 + taill) - l : locinput - loc_bostr;
3187 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3189 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3191 pref0_len = pref_len - (locinput - loc_reg_starttry);
3192 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3193 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3194 ? (5 + taill) - pref_len : loc_regeol - locinput);
3195 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3199 if (pref0_len > pref_len)
3200 pref0_len = pref_len;
3202 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3204 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3205 (locinput - pref_len),pref0_len, 60, 4, 5);
3207 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3208 (locinput - pref_len + pref0_len),
3209 pref_len - pref0_len, 60, 2, 3);
3211 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3212 locinput, loc_regeol - locinput, 10, 0, 1);
3214 const STRLEN tlen=len0+len1+len2;
3215 PerlIO_printf(Perl_debug_log,
3216 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3217 (IV)(locinput - loc_bostr),
3220 (docolor ? "" : "> <"),
3222 (int)(tlen > 19 ? 0 : 19 - tlen),
3229 /* reg_check_named_buff_matched()
3230 * Checks to see if a named buffer has matched. The data array of
3231 * buffer numbers corresponding to the buffer is expected to reside
3232 * in the regexp->data->data array in the slot stored in the ARG() of
3233 * node involved. Note that this routine doesn't actually care about the
3234 * name, that information is not preserved from compilation to execution.
3235 * Returns the index of the leftmost defined buffer with the given name
3236 * or 0 if non of the buffers matched.
3239 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3242 RXi_GET_DECL(rex,rexi);
3243 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3244 I32 *nums=(I32*)SvPVX(sv_dat);
3246 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3248 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3249 if ((I32)rex->lastparen >= nums[n] &&
3250 rex->offs[nums[n]].end != -1)
3259 /* free all slabs above current one - called during LEAVE_SCOPE */
3262 S_clear_backtrack_stack(pTHX_ void *p)
3264 regmatch_slab *s = PL_regmatch_slab->next;
3269 PL_regmatch_slab->next = NULL;
3271 regmatch_slab * const osl = s;
3277 S_setup_EXACTISH_ST_c1_c2(pTHX_ regnode *text_node, I32 *c1, I32 *c2)
3279 /* This sets up a relatively quick check for the initial part of what must
3280 * match after a CURLY-type operation condition (the "B" in A*B), where B
3281 * starts with an EXACTish node, <text_node>. If this check is not met,
3282 * the caller knows that it should continue with the loop. If the check is
3283 * met, the caller must see if all of B is met, before making the decision.
3285 * This function sets *<c1> and *<c2> to be the first code point of B. If
3286 * there are two possible such code points (as when the text_node is
3287 * folded), *<c2> is set to the second. If there are more than two (which
3288 * happens for some folds), or there is some other complication, these
3289 * parameters are set to CHRTEST_VOID, to indicate not to do a quick check:
3290 * just try all of B after every time through the loop.
3292 * If the routine determines that there is no possible way for there to be
3293 * a match, it returns FALSE.
3296 const bool utf8_target = PL_reg_match_utf8;
3297 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3300 /* First byte from the EXACTish node */
3301 U8 *pat_byte = (U8*)STRING(text_node);
3303 if (! UTF_PATTERN) { /* Not UTF-8: the code point is the byte */
3305 if (OP(text_node) == EXACT) {
3308 else if (utf8_target
3309 && HAS_NONLATIN1_FOLD_CLOSURE(*c1)
3310 && (OP(text_node) != EXACTFA || ! isASCII(*c1)))
3312 /* Here, there could be something above Latin1 in the target which
3313 * folds to this character in the pattern, which means there are
3314 * more than two possible beginnings of B. */
3315 *c1 = *c2 = CHRTEST_VOID;
3317 else { /* Here nothing above Latin1 can fold to the pattern character */
3318 switch (OP(text_node)) {
3320 case EXACTFL: /* /l rules */
3321 *c2 = PL_fold_locale[*c1];
3324 case EXACTFU_SS: /* This requires special handling: Don't
3326 *c1 = *c2 = CHRTEST_VOID;
3330 if (! utf8_target) { /* /d rules */
3335 /* /u rules for all these. This happens to work for
3336 * EXACTFA in the ASCII range as nothing in Latin1 folds to
3339 case EXACTFU_TRICKYFOLD:
3341 *c2 = PL_fold_latin1[*c1];
3344 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3348 else { /* UTF_PATTERN */
3349 if (OP(text_node) == EXACT) {
3350 *c2 = *c1 = utf8n_to_uvchr(pat_byte, UTF8_MAXBYTES, 0, uniflags);
3351 if (*c1 < 0) { /* Overflowed what we can handle */
3352 *c1 = *c2 = CHRTEST_VOID;
3354 else if (*c1 > 255 && ! utf8_target) {
3355 return FALSE; /* Can't possibly match */
3359 if (UTF8_IS_ABOVE_LATIN1(*pat_byte)) {
3361 /* A multi-character fold is complicated, probably has more
3362 * than two possibilities */
3363 if (is_MULTI_CHAR_FOLD_utf8_safe((char*) pat_byte,
3364 (char*) pat_byte + STR_LEN(text_node)))
3366 *c1 = *c2 = CHRTEST_VOID;
3368 else { /* Not a multi-char fold */
3370 /* Load the folds hash, if not already done */
3372 if (! PL_utf8_foldclosures) {
3373 if (! PL_utf8_tofold) {
3374 U8 dummy[UTF8_MAXBYTES+1];
3377 /* Force loading this by folding an above-Latin1
3379 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, &dummy_len);
3380 assert(PL_utf8_tofold); /* Verify that worked */
3382 PL_utf8_foldclosures =
3383 _swash_inversion_hash(PL_utf8_tofold);
3386 /* The fold closures data structure is a hash with the keys
3387 * being every character that is folded to, like 'k', and
3388 * the values each an array of everything that folds to its
3389 * key. e.g. [ 'k', 'K', KELVIN_SIGN ] */
3390 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3395 /* Not found in the hash, therefore there are no folds
3396 * containing it, so there is only a single char
3397 * possible for beginning B */
3398 *c2 = *c1 = utf8n_to_uvchr(pat_byte, STR_LEN(text_node),
3400 if (*c1 < 0) { /* Overflowed what we can handle */
3401 *c1 = *c2 = CHRTEST_VOID;
3405 AV* list = (AV*) *listp;
3406 if (av_len(list) != 1) { /* If there aren't exactly
3407 two folds to this, have
3408 to test B completely */
3409 *c1 = *c2 = CHRTEST_VOID;
3411 else { /* There are two. Set *c1 and *c2 to them */
3412 SV** c_p = av_fetch(list, 0, FALSE);
3414 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3417 c_p = av_fetch(list, 1, FALSE);
3419 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3427 /* Get the character represented by the UTF-8-encoded byte */
3428 U8 c = (UTF8_IS_INVARIANT(*pat_byte))
3430 : TWO_BYTE_UTF8_TO_UNI(*pat_byte, *(pat_byte+1));
3432 if (HAS_NONLATIN1_FOLD_CLOSURE(c)
3433 && (OP(text_node) != EXACTFA || ! isASCII(c)))
3434 { /* Something above Latin1 folds to this; hence there are
3435 more than 2 possibilities for B to begin with */
3436 *c1 = *c2 = CHRTEST_VOID;
3440 *c2 = (OP(text_node) == EXACTFL)
3441 ? PL_fold_locale[*c1]
3442 : PL_fold_latin1[*c1];
3451 /* returns -1 on failure, $+[0] on success */
3453 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3455 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3459 const bool utf8_target = PL_reg_match_utf8;
3460 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3461 REGEXP *rex_sv = reginfo->prog;
3462 regexp *rex = (struct regexp *)SvANY(rex_sv);
3463 RXi_GET_DECL(rex,rexi);
3465 /* the current state. This is a cached copy of PL_regmatch_state */
3467 /* cache heavy used fields of st in registers */
3470 U32 n = 0; /* general value; init to avoid compiler warning */
3471 I32 ln = 0; /* len or last; init to avoid compiler warning */
3472 char *locinput = startpos;
3473 char *pushinput; /* where to continue after a PUSH */
3474 I32 nextchr; /* is always set to UCHARAT(locinput) */
3476 bool result = 0; /* return value of S_regmatch */
3477 int depth = 0; /* depth of backtrack stack */
3478 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3479 const U32 max_nochange_depth =
3480 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3481 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3482 regmatch_state *yes_state = NULL; /* state to pop to on success of
3484 /* mark_state piggy backs on the yes_state logic so that when we unwind
3485 the stack on success we can update the mark_state as we go */
3486 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3487 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3488 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3490 bool no_final = 0; /* prevent failure from backtracking? */
3491 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3492 char *startpoint = locinput;
3493 SV *popmark = NULL; /* are we looking for a mark? */
3494 SV *sv_commit = NULL; /* last mark name seen in failure */
3495 SV *sv_yes_mark = NULL; /* last mark name we have seen
3496 during a successful match */
3497 U32 lastopen = 0; /* last open we saw */
3498 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3499 SV* const oreplsv = GvSV(PL_replgv);
3500 /* these three flags are set by various ops to signal information to
3501 * the very next op. They have a useful lifetime of exactly one loop
3502 * iteration, and are not preserved or restored by state pushes/pops
3504 bool sw = 0; /* the condition value in (?(cond)a|b) */
3505 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3506 int logical = 0; /* the following EVAL is:
3510 or the following IFMATCH/UNLESSM is:
3511 false: plain (?=foo)
3512 true: used as a condition: (?(?=foo))
3514 PAD* last_pad = NULL;
3516 I32 gimme = G_SCALAR;
3517 CV *caller_cv = NULL; /* who called us */
3518 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3519 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3522 GET_RE_DEBUG_FLAGS_DECL;
3525 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3526 multicall_oldcatch = 0;
3527 multicall_cv = NULL;
3529 PERL_UNUSED_VAR(multicall_cop);
3530 PERL_UNUSED_VAR(newsp);
3533 PERL_ARGS_ASSERT_REGMATCH;
3535 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3536 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3538 /* on first ever call to regmatch, allocate first slab */
3539 if (!PL_regmatch_slab) {
3540 Newx(PL_regmatch_slab, 1, regmatch_slab);
3541 PL_regmatch_slab->prev = NULL;
3542 PL_regmatch_slab->next = NULL;
3543 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3546 oldsave = PL_savestack_ix;
3547 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3548 SAVEVPTR(PL_regmatch_slab);
3549 SAVEVPTR(PL_regmatch_state);
3551 /* grab next free state slot */
3552 st = ++PL_regmatch_state;
3553 if (st > SLAB_LAST(PL_regmatch_slab))
3554 st = PL_regmatch_state = S_push_slab(aTHX);
3556 /* Note that nextchr is a byte even in UTF */
3559 while (scan != NULL) {
3562 SV * const prop = sv_newmortal();
3563 regnode *rnext=regnext(scan);
3564 DUMP_EXEC_POS( locinput, scan, utf8_target );
3565 regprop(rex, prop, scan);
3567 PerlIO_printf(Perl_debug_log,
3568 "%3"IVdf":%*s%s(%"IVdf")\n",
3569 (IV)(scan - rexi->program), depth*2, "",
3571 (PL_regkind[OP(scan)] == END || !rnext) ?
3572 0 : (IV)(rnext - rexi->program));
3575 next = scan + NEXT_OFF(scan);
3578 state_num = OP(scan);
3584 switch (state_num) {
3585 case BOL: /* /^../ */
3586 if (locinput == PL_bostr)
3588 /* reginfo->till = reginfo->bol; */
3593 case MBOL: /* /^../m */
3594 if (locinput == PL_bostr ||
3595 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3601 case SBOL: /* /^../s */
3602 if (locinput == PL_bostr)
3607 if (locinput == reginfo->ganch)
3611 case KEEPS: /* \K */
3612 /* update the startpoint */
3613 st->u.keeper.val = rex->offs[0].start;
3614 rex->offs[0].start = locinput - PL_bostr;
3615 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3617 case KEEPS_next_fail:
3618 /* rollback the start point change */
3619 rex->offs[0].start = st->u.keeper.val;
3623 case EOL: /* /..$/ */
3626 case MEOL: /* /..$/m */
3627 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3631 case SEOL: /* /..$/s */
3633 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3635 if (PL_regeol - locinput > 1)
3640 if (!NEXTCHR_IS_EOS)
3644 case SANY: /* /./s */
3647 goto increment_locinput;
3655 case REG_ANY: /* /./ */
3656 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3658 goto increment_locinput;
3662 #define ST st->u.trie
3663 case TRIEC: /* (ab|cd) with known charclass */
3664 /* In this case the charclass data is available inline so
3665 we can fail fast without a lot of extra overhead.
3667 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3669 PerlIO_printf(Perl_debug_log,
3670 "%*s %sfailed to match trie start class...%s\n",
3671 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3674 assert(0); /* NOTREACHED */
3677 case TRIE: /* (ab|cd) */
3678 /* the basic plan of execution of the trie is:
3679 * At the beginning, run though all the states, and
3680 * find the longest-matching word. Also remember the position
3681 * of the shortest matching word. For example, this pattern:
3684 * when matched against the string "abcde", will generate
3685 * accept states for all words except 3, with the longest
3686 * matching word being 4, and the shortest being 2 (with
3687 * the position being after char 1 of the string).
3689 * Then for each matching word, in word order (i.e. 1,2,4,5),
3690 * we run the remainder of the pattern; on each try setting
3691 * the current position to the character following the word,
3692 * returning to try the next word on failure.
3694 * We avoid having to build a list of words at runtime by
3695 * using a compile-time structure, wordinfo[].prev, which
3696 * gives, for each word, the previous accepting word (if any).
3697 * In the case above it would contain the mappings 1->2, 2->0,
3698 * 3->0, 4->5, 5->1. We can use this table to generate, from
3699 * the longest word (4 above), a list of all words, by
3700 * following the list of prev pointers; this gives us the
3701 * unordered list 4,5,1,2. Then given the current word we have
3702 * just tried, we can go through the list and find the
3703 * next-biggest word to try (so if we just failed on word 2,
3704 * the next in the list is 4).
3706 * Since at runtime we don't record the matching position in
3707 * the string for each word, we have to work that out for
3708 * each word we're about to process. The wordinfo table holds
3709 * the character length of each word; given that we recorded
3710 * at the start: the position of the shortest word and its
3711 * length in chars, we just need to move the pointer the
3712 * difference between the two char lengths. Depending on
3713 * Unicode status and folding, that's cheap or expensive.
3715 * This algorithm is optimised for the case where are only a
3716 * small number of accept states, i.e. 0,1, or maybe 2.
3717 * With lots of accepts states, and having to try all of them,
3718 * it becomes quadratic on number of accept states to find all
3723 /* what type of TRIE am I? (utf8 makes this contextual) */
3724 DECL_TRIE_TYPE(scan);
3726 /* what trie are we using right now */
3727 reg_trie_data * const trie
3728 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3729 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3730 U32 state = trie->startstate;
3733 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3735 if (trie->states[ state ].wordnum) {
3737 PerlIO_printf(Perl_debug_log,
3738 "%*s %smatched empty string...%s\n",
3739 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3745 PerlIO_printf(Perl_debug_log,
3746 "%*s %sfailed to match trie start class...%s\n",
3747 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3754 U8 *uc = ( U8* )locinput;
3758 U8 *uscan = (U8*)NULL;
3759 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3760 U32 charcount = 0; /* how many input chars we have matched */
3761 U32 accepted = 0; /* have we seen any accepting states? */
3763 ST.jump = trie->jump;
3766 ST.longfold = FALSE; /* char longer if folded => it's harder */
3769 /* fully traverse the TRIE; note the position of the
3770 shortest accept state and the wordnum of the longest
3773 while ( state && uc <= (U8*)PL_regeol ) {
3774 U32 base = trie->states[ state ].trans.base;
3778 wordnum = trie->states[ state ].wordnum;
3780 if (wordnum) { /* it's an accept state */
3783 /* record first match position */
3785 ST.firstpos = (U8*)locinput;
3790 ST.firstchars = charcount;
3793 if (!ST.nextword || wordnum < ST.nextword)
3794 ST.nextword = wordnum;
3795 ST.topword = wordnum;
3798 DEBUG_TRIE_EXECUTE_r({
3799 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3800 PerlIO_printf( Perl_debug_log,
3801 "%*s %sState: %4"UVxf" Accepted: %c ",
3802 2+depth * 2, "", PL_colors[4],
3803 (UV)state, (accepted ? 'Y' : 'N'));
3806 /* read a char and goto next state */
3807 if ( base && (foldlen || uc < (U8*)PL_regeol)) {
3809 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3810 uscan, len, uvc, charid, foldlen,
3817 base + charid - 1 - trie->uniquecharcount)) >= 0)
3819 && ((U32)offset < trie->lasttrans)
3820 && trie->trans[offset].check == state)
3822 state = trie->trans[offset].next;
3833 DEBUG_TRIE_EXECUTE_r(
3834 PerlIO_printf( Perl_debug_log,
3835 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3836 charid, uvc, (UV)state, PL_colors[5] );
3842 /* calculate total number of accept states */
3847 w = trie->wordinfo[w].prev;