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;
3850 ST.accepted = accepted;
3854 PerlIO_printf( Perl_debug_log,
3855 "%*s %sgot %"IVdf" possible matches%s\n",
3856 REPORT_CODE_OFF + depth * 2, "",
3857 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3859 goto trie_first_try; /* jump into the fail handler */
3861 assert(0); /* NOTREACHED */
3863 case TRIE_next_fail: /* we failed - try next alternative */
3867 REGCP_UNWIND(ST.cp);
3868 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3870 if (!--ST.accepted) {
3872 PerlIO_printf( Perl_debug_log,
3873 "%*s %sTRIE failed...%s\n",
3874 REPORT_CODE_OFF+depth*2, "",
3881 /* Find next-highest word to process. Note that this code
3882 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3885 U16 const nextword = ST.nextword;
3886 reg_trie_wordinfo * const wordinfo
3887 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3888 for (word=ST.topword; word; word=wordinfo[word].prev) {
3889 if (word > nextword && (!min || word < min))
3902 ST.lastparen = rex->lastparen;
3903 ST.lastcloseparen = rex->lastcloseparen;
3907 /* find start char of end of current word */
3909 U32 chars; /* how many chars to skip */
3910 reg_trie_data * const trie
3911 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3913 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3915 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3920 /* the hard option - fold each char in turn and find
3921 * its folded length (which may be different */
3922 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3930 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3938 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3943 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3959 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
3960 ? ST.jump[ST.nextword]
3964 PerlIO_printf( Perl_debug_log,
3965 "%*s %sTRIE matched word #%d, continuing%s\n",
3966 REPORT_CODE_OFF+depth*2, "",
3973 if (ST.accepted > 1 || has_cutgroup) {
3974 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
3975 assert(0); /* NOTREACHED */
3977 /* only one choice left - just continue */
3979 AV *const trie_words
3980 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3981 SV ** const tmp = av_fetch( trie_words,
3983 SV *sv= tmp ? sv_newmortal() : NULL;
3985 PerlIO_printf( Perl_debug_log,
3986 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3987 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3989 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3990 PL_colors[0], PL_colors[1],
3991 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3993 : "not compiled under -Dr",
3997 locinput = (char*)uc;
3998 continue; /* execute rest of RE */
3999 assert(0); /* NOTREACHED */
4003 case EXACT: { /* /abc/ */
4004 char *s = STRING(scan);
4006 if (utf8_target != UTF_PATTERN) {
4007 /* The target and the pattern have differing utf8ness. */
4009 const char * const e = s + ln;
4012 /* The target is utf8, the pattern is not utf8.
4013 * Above-Latin1 code points can't match the pattern;
4014 * invariants match exactly, and the other Latin1 ones need
4015 * to be downgraded to a single byte in order to do the
4016 * comparison. (If we could be confident that the target
4017 * is not malformed, this could be refactored to have fewer
4018 * tests by just assuming that if the first bytes match, it
4019 * is an invariant, but there are tests in the test suite
4020 * dealing with (??{...}) which violate this) */
4024 if (UTF8_IS_ABOVE_LATIN1(* (U8*) l)) {
4027 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4034 if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) {
4043 /* The target is not utf8, the pattern is utf8. */
4045 if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4049 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4056 if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) {
4067 /* The target and the pattern have the same utf8ness. */
4068 /* Inline the first character, for speed. */
4069 if (UCHARAT(s) != nextchr)
4071 if (PL_regeol - locinput < ln)
4073 if (ln > 1 && memNE(s, locinput, ln))
4079 case EXACTFL: { /* /abc/il */
4081 const U8 * fold_array;
4083 U32 fold_utf8_flags;
4085 PL_reg_flags |= RF_tainted;
4086 folder = foldEQ_locale;
4087 fold_array = PL_fold_locale;
4088 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
4091 case EXACTFU_SS: /* /\x{df}/iu */
4092 case EXACTFU_TRICKYFOLD: /* /\x{390}/iu */
4093 case EXACTFU: /* /abc/iu */
4094 folder = foldEQ_latin1;
4095 fold_array = PL_fold_latin1;
4096 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4099 case EXACTFA: /* /abc/iaa */
4100 folder = foldEQ_latin1;
4101 fold_array = PL_fold_latin1;
4102 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4105 case EXACTF: /* /abc/i */
4107 fold_array = PL_fold;
4108 fold_utf8_flags = 0;
4114 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
4115 /* Either target or the pattern are utf8, or has the issue where
4116 * the fold lengths may differ. */
4117 const char * const l = locinput;
4118 char *e = PL_regeol;
4120 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
4121 l, &e, 0, utf8_target, fold_utf8_flags))
4129 /* Neither the target nor the pattern are utf8 */
4130 if (UCHARAT(s) != nextchr &&
4131 UCHARAT(s) != fold_array[nextchr])
4135 if (PL_regeol - locinput < ln)
4137 if (ln > 1 && ! folder(s, locinput, ln))
4143 /* XXX Could improve efficiency by separating these all out using a
4144 * macro or in-line function. At that point regcomp.c would no longer
4145 * have to set the FLAGS fields of these */
4146 case BOUNDL: /* /\b/l */
4147 case NBOUNDL: /* /\B/l */
4148 PL_reg_flags |= RF_tainted;
4150 case BOUND: /* /\b/ */
4151 case BOUNDU: /* /\b/u */
4152 case BOUNDA: /* /\b/a */
4153 case NBOUND: /* /\B/ */
4154 case NBOUNDU: /* /\B/u */
4155 case NBOUNDA: /* /\B/a */
4156 /* was last char in word? */
4158 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4159 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4161 if (locinput == PL_bostr)
4164 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
4166 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
4168 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4169 ln = isALNUM_uni(ln);
4173 LOAD_UTF8_CHARCLASS_ALNUM();
4174 n = swash_fetch(PL_utf8_alnum, (U8*)locinput,
4179 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
4180 n = NEXTCHR_IS_EOS ? 0 : isALNUM_LC_utf8((U8*)locinput);
4185 /* Here the string isn't utf8, or is utf8 and only ascii
4186 * characters are to match \w. In the latter case looking at
4187 * the byte just prior to the current one may be just the final
4188 * byte of a multi-byte character. This is ok. There are two
4190 * 1) it is a single byte character, and then the test is doing
4191 * just what it's supposed to.
4192 * 2) it is a multi-byte character, in which case the final
4193 * byte is never mistakable for ASCII, and so the test
4194 * will say it is not a word character, which is the
4195 * correct answer. */
4196 ln = (locinput != PL_bostr) ?
4197 UCHARAT(locinput - 1) : '\n';
4198 switch (FLAGS(scan)) {
4199 case REGEX_UNICODE_CHARSET:
4200 ln = isWORDCHAR_L1(ln);
4201 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4203 case REGEX_LOCALE_CHARSET:
4204 ln = isALNUM_LC(ln);
4205 n = NEXTCHR_IS_EOS ? 0 : isALNUM_LC(nextchr);
4207 case REGEX_DEPENDS_CHARSET:
4209 n = NEXTCHR_IS_EOS ? 0 : isALNUM(nextchr);
4211 case REGEX_ASCII_RESTRICTED_CHARSET:
4212 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4213 ln = isWORDCHAR_A(ln);
4214 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4217 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4221 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4223 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4227 case ANYOFV: /* /[abx{df}]/i */
4228 case ANYOF: /* /[abc]/ */
4231 if (utf8_target || state_num == ANYOFV) {
4232 STRLEN inclasslen = PL_regeol - locinput;
4233 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
4235 locinput += inclasslen;
4239 if (!REGINCLASS(rex, scan, (U8*)locinput))
4246 /* Special char classes: \d, \w etc.
4247 * The defines start on line 166 or so */
4248 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
4249 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
4250 ALNUMU, NALNUMU, isWORDCHAR_L1,
4251 ALNUMA, NALNUMA, isWORDCHAR_A,
4254 CCC_TRY_U(SPACE, NSPACE, isSPACE,
4255 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
4256 SPACEU, NSPACEU, isSPACE_L1,
4257 SPACEA, NSPACEA, isSPACE_A,
4260 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
4261 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
4262 DIGITA, NDIGITA, isDIGIT_A,
4265 case POSIXA: /* /[[:ascii:]]/ etc */
4266 if (NEXTCHR_IS_EOS || ! _generic_isCC_A(nextchr, FLAGS(scan))) {
4269 /* Matched a utf8-invariant, so don't have to worry about utf8 */
4273 case NPOSIXA: /* /[^[:ascii:]]/ etc */
4274 if (NEXTCHR_IS_EOS || _generic_isCC_A(nextchr, FLAGS(scan))) {
4277 goto increment_locinput;
4279 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4280 a Unicode extended Grapheme Cluster */
4281 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4282 extended Grapheme Cluster is:
4285 | Prepend* Begin Extend*
4288 Begin is: ( Special_Begin | ! Control )
4289 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4290 Extend is: ( Grapheme_Extend | Spacing_Mark )
4291 Control is: [ GCB_Control CR LF ]
4292 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4294 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4297 Begin is ( Regular_Begin + Special Begin )
4299 It turns out that 98.4% of all Unicode code points match
4300 Regular_Begin. Doing it this way eliminates a table match in
4301 the previous implementation for almost all Unicode code points.
4303 There is a subtlety with Prepend* which showed up in testing.
4304 Note that the Begin, and only the Begin is required in:
4305 | Prepend* Begin Extend*
4306 Also, Begin contains '! Control'. A Prepend must be a
4307 '! Control', which means it must also be a Begin. What it
4308 comes down to is that if we match Prepend* and then find no
4309 suitable Begin afterwards, that if we backtrack the last
4310 Prepend, that one will be a suitable Begin.
4315 if (! utf8_target) {
4317 /* Match either CR LF or '.', as all the other possibilities
4319 locinput++; /* Match the . or CR */
4320 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4322 && locinput < PL_regeol
4323 && UCHARAT(locinput) == '\n') locinput++;
4327 /* Utf8: See if is ( CR LF ); already know that locinput <
4328 * PL_regeol, so locinput+1 is in bounds */
4329 if ( nextchr == '\r' && locinput+1 < PL_regeol
4330 && UCHARAT(locinput + 1) == '\n')
4337 /* In case have to backtrack to beginning, then match '.' */
4338 char *starting = locinput;
4340 /* In case have to backtrack the last prepend */
4341 char *previous_prepend = 0;
4343 LOAD_UTF8_CHARCLASS_GCB();
4345 /* Match (prepend)* */
4346 while (locinput < PL_regeol
4347 && (len = is_GCB_Prepend_utf8(locinput)))
4349 previous_prepend = locinput;
4353 /* As noted above, if we matched a prepend character, but
4354 * the next thing won't match, back off the last prepend we
4355 * matched, as it is guaranteed to match the begin */
4356 if (previous_prepend
4357 && (locinput >= PL_regeol
4358 || (! swash_fetch(PL_utf8_X_regular_begin,
4359 (U8*)locinput, utf8_target)
4360 && ! is_GCB_SPECIAL_BEGIN_utf8(locinput)))
4363 locinput = previous_prepend;
4366 /* Note that here we know PL_regeol > locinput, as we
4367 * tested that upon input to this switch case, and if we
4368 * moved locinput forward, we tested the result just above
4369 * and it either passed, or we backed off so that it will
4371 if (swash_fetch(PL_utf8_X_regular_begin,
4372 (U8*)locinput, utf8_target)) {
4373 locinput += UTF8SKIP(locinput);
4375 else if (! is_GCB_SPECIAL_BEGIN_utf8(locinput)) {
4377 /* Here did not match the required 'Begin' in the
4378 * second term. So just match the very first
4379 * character, the '.' of the final term of the regex */
4380 locinput = starting + UTF8SKIP(starting);
4384 /* Here is a special begin. It can be composed of
4385 * several individual characters. One possibility is
4387 if ((len = is_GCB_RI_utf8(locinput))) {
4389 while (locinput < PL_regeol
4390 && (len = is_GCB_RI_utf8(locinput)))
4394 } else if ((len = is_GCB_T_utf8(locinput))) {
4395 /* Another possibility is T+ */
4397 while (locinput < PL_regeol
4398 && (len = is_GCB_T_utf8(locinput)))
4404 /* Here, neither RI+ nor T+; must be some other
4405 * Hangul. That means it is one of the others: L,
4406 * LV, LVT or V, and matches:
4407 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4410 while (locinput < PL_regeol
4411 && (len = is_GCB_L_utf8(locinput)))
4416 /* Here, have exhausted L*. If the next character
4417 * is not an LV, LVT nor V, it means we had to have
4418 * at least one L, so matches L+ in the original
4419 * equation, we have a complete hangul syllable.
4422 if (locinput < PL_regeol
4423 && is_GCB_LV_LVT_V_utf8(locinput))
4426 /* Otherwise keep going. Must be LV, LVT or V.
4428 if (is_utf8_X_LVT((U8*)locinput)) {
4429 locinput += UTF8SKIP(locinput);
4432 /* Must be V or LV. Take it, then match
4434 locinput += UTF8SKIP(locinput);
4435 while (locinput < PL_regeol
4436 && (len = is_GCB_V_utf8(locinput)))
4442 /* And any of LV, LVT, or V can be followed
4444 while (locinput < PL_regeol
4445 && (len = is_GCB_T_utf8(locinput)))
4453 /* Match any extender */
4454 while (locinput < PL_regeol
4455 && swash_fetch(PL_utf8_X_extend,
4456 (U8*)locinput, utf8_target))
4458 locinput += UTF8SKIP(locinput);
4462 if (locinput > PL_regeol) sayNO;
4466 case NREFFL: /* /\g{name}/il */
4467 { /* The capture buffer cases. The ones beginning with N for the
4468 named buffers just convert to the equivalent numbered and
4469 pretend they were called as the corresponding numbered buffer
4471 /* don't initialize these in the declaration, it makes C++
4476 const U8 *fold_array;
4479 PL_reg_flags |= RF_tainted;
4480 folder = foldEQ_locale;
4481 fold_array = PL_fold_locale;
4483 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4486 case NREFFA: /* /\g{name}/iaa */
4487 folder = foldEQ_latin1;
4488 fold_array = PL_fold_latin1;
4490 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4493 case NREFFU: /* /\g{name}/iu */
4494 folder = foldEQ_latin1;
4495 fold_array = PL_fold_latin1;
4497 utf8_fold_flags = 0;
4500 case NREFF: /* /\g{name}/i */
4502 fold_array = PL_fold;
4504 utf8_fold_flags = 0;
4507 case NREF: /* /\g{name}/ */
4511 utf8_fold_flags = 0;
4514 /* For the named back references, find the corresponding buffer
4516 n = reg_check_named_buff_matched(rex,scan);
4521 goto do_nref_ref_common;
4523 case REFFL: /* /\1/il */
4524 PL_reg_flags |= RF_tainted;
4525 folder = foldEQ_locale;
4526 fold_array = PL_fold_locale;
4527 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4530 case REFFA: /* /\1/iaa */
4531 folder = foldEQ_latin1;
4532 fold_array = PL_fold_latin1;
4533 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4536 case REFFU: /* /\1/iu */
4537 folder = foldEQ_latin1;
4538 fold_array = PL_fold_latin1;
4539 utf8_fold_flags = 0;
4542 case REFF: /* /\1/i */
4544 fold_array = PL_fold;
4545 utf8_fold_flags = 0;
4548 case REF: /* /\1/ */
4551 utf8_fold_flags = 0;
4555 n = ARG(scan); /* which paren pair */
4558 ln = rex->offs[n].start;
4559 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4560 if (rex->lastparen < n || ln == -1)
4561 sayNO; /* Do not match unless seen CLOSEn. */
4562 if (ln == rex->offs[n].end)
4566 if (type != REF /* REF can do byte comparison */
4567 && (utf8_target || type == REFFU))
4568 { /* XXX handle REFFL better */
4569 char * limit = PL_regeol;
4571 /* This call case insensitively compares the entire buffer
4572 * at s, with the current input starting at locinput, but
4573 * not going off the end given by PL_regeol, and returns in
4574 * limit upon success, how much of the current input was
4576 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4577 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4585 /* Not utf8: Inline the first character, for speed. */
4586 if (!NEXTCHR_IS_EOS &&
4587 UCHARAT(s) != nextchr &&
4589 UCHARAT(s) != fold_array[nextchr]))
4591 ln = rex->offs[n].end - ln;
4592 if (locinput + ln > PL_regeol)
4594 if (ln > 1 && (type == REF
4595 ? memNE(s, locinput, ln)
4596 : ! folder(s, locinput, ln)))
4602 case NOTHING: /* null op; e.g. the 'nothing' following
4603 * the '*' in m{(a+|b)*}' */
4605 case TAIL: /* placeholder while compiling (A|B|C) */
4608 case BACK: /* ??? doesn't appear to be used ??? */
4612 #define ST st->u.eval
4617 regexp_internal *rei;
4618 regnode *startpoint;
4620 case GOSTART: /* (?R) */
4621 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4622 if (cur_eval && cur_eval->locinput==locinput) {
4623 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4624 Perl_croak(aTHX_ "Infinite recursion in regex");
4625 if ( ++nochange_depth > max_nochange_depth )
4627 "Pattern subroutine nesting without pos change"
4628 " exceeded limit in regex");
4635 if (OP(scan)==GOSUB) {
4636 startpoint = scan + ARG2L(scan);
4637 ST.close_paren = ARG(scan);
4639 startpoint = rei->program+1;
4642 goto eval_recurse_doit;
4643 assert(0); /* NOTREACHED */
4645 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4646 if (cur_eval && cur_eval->locinput==locinput) {
4647 if ( ++nochange_depth > max_nochange_depth )
4648 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4653 /* execute the code in the {...} */
4657 OP * const oop = PL_op;
4658 COP * const ocurcop = PL_curcop;
4660 char *saved_regeol = PL_regeol;
4661 struct re_save_state saved_state;
4664 /* save *all* paren positions */
4666 REGCP_SET(runops_cp);
4668 /* To not corrupt the existing regex state while executing the
4669 * eval we would normally put it on the save stack, like with
4670 * save_re_context. However, re-evals have a weird scoping so we
4671 * can't just add ENTER/LEAVE here. With that, things like
4673 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4675 * would break, as they expect the localisation to be unwound
4676 * only when the re-engine backtracks through the bit that
4679 * What we do instead is just saving the state in a local c
4682 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4684 PL_reg_state.re_reparsing = FALSE;
4687 caller_cv = find_runcv(NULL);
4691 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4692 newcv = ((struct regexp *)SvANY(
4693 (REGEXP*)(rexi->data->data[n])
4696 nop = (OP*)rexi->data->data[n+1];
4698 else if (rexi->data->what[n] == 'l') { /* literal code */
4700 nop = (OP*)rexi->data->data[n];
4701 assert(CvDEPTH(newcv));
4704 /* literal with own CV */
4705 assert(rexi->data->what[n] == 'L');
4706 newcv = rex->qr_anoncv;
4707 nop = (OP*)rexi->data->data[n];
4710 /* normally if we're about to execute code from the same
4711 * CV that we used previously, we just use the existing
4712 * CX stack entry. However, its possible that in the
4713 * meantime we may have backtracked, popped from the save
4714 * stack, and undone the SAVECOMPPAD(s) associated with
4715 * PUSH_MULTICALL; in which case PL_comppad no longer
4716 * points to newcv's pad. */
4717 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4719 I32 depth = (newcv == caller_cv) ? 0 : 1;
4720 if (last_pushed_cv) {
4721 CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
4724 PUSH_MULTICALL_WITHDEPTH(newcv, depth);
4726 last_pushed_cv = newcv;
4728 last_pad = PL_comppad;
4730 /* the initial nextstate you would normally execute
4731 * at the start of an eval (which would cause error
4732 * messages to come from the eval), may be optimised
4733 * away from the execution path in the regex code blocks;
4734 * so manually set PL_curcop to it initially */
4736 OP *o = cUNOPx(nop)->op_first;
4737 assert(o->op_type == OP_NULL);
4738 if (o->op_targ == OP_SCOPE) {
4739 o = cUNOPo->op_first;
4742 assert(o->op_targ == OP_LEAVE);
4743 o = cUNOPo->op_first;
4744 assert(o->op_type == OP_ENTER);
4748 if (o->op_type != OP_STUB) {
4749 assert( o->op_type == OP_NEXTSTATE
4750 || o->op_type == OP_DBSTATE
4751 || (o->op_type == OP_NULL
4752 && ( o->op_targ == OP_NEXTSTATE
4753 || o->op_targ == OP_DBSTATE
4757 PL_curcop = (COP*)o;
4762 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4763 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4765 rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4768 SV *sv_mrk = get_sv("REGMARK", 1);
4769 sv_setsv(sv_mrk, sv_yes_mark);
4772 /* we don't use MULTICALL here as we want to call the
4773 * first op of the block of interest, rather than the
4774 * first op of the sub */
4777 CALLRUNOPS(aTHX); /* Scalar context. */
4780 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4786 /* before restoring everything, evaluate the returned
4787 * value, so that 'uninit' warnings don't use the wrong
4788 * PL_op or pad. Also need to process any magic vars
4789 * (e.g. $1) *before* parentheses are restored */
4794 if (logical == 0) /* (?{})/ */
4795 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
4796 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
4797 sw = cBOOL(SvTRUE(ret));
4800 else { /* /(??{}) */
4801 /* if its overloaded, let the regex compiler handle
4802 * it; otherwise extract regex, or stringify */
4803 if (!SvAMAGIC(ret)) {
4807 if (SvTYPE(sv) == SVt_REGEXP)
4808 re_sv = (REGEXP*) sv;
4809 else if (SvSMAGICAL(sv)) {
4810 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
4812 re_sv = (REGEXP *) mg->mg_obj;
4815 /* force any magic, undef warnings here */
4817 ret = sv_mortalcopy(ret);
4818 (void) SvPV_force_nolen(ret);
4824 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4826 /* *** Note that at this point we don't restore
4827 * PL_comppad, (or pop the CxSUB) on the assumption it may
4828 * be used again soon. This is safe as long as nothing
4829 * in the regexp code uses the pad ! */
4831 PL_curcop = ocurcop;
4832 PL_regeol = saved_regeol;
4833 S_regcp_restore(aTHX_ rex, runops_cp);
4839 /* only /(??{})/ from now on */
4842 /* extract RE object from returned value; compiling if
4846 re_sv = reg_temp_copy(NULL, re_sv);
4850 const I32 osize = PL_regsize;
4852 if (SvUTF8(ret) && IN_BYTES) {
4853 /* In use 'bytes': make a copy of the octet
4854 * sequence, but without the flag on */
4856 const char *const p = SvPV(ret, len);
4857 ret = newSVpvn_flags(p, len, SVs_TEMP);
4859 if (rex->intflags & PREGf_USE_RE_EVAL)
4860 pm_flags |= PMf_USE_RE_EVAL;
4862 /* if we got here, it should be an engine which
4863 * supports compiling code blocks and stuff */
4864 assert(rex->engine && rex->engine->op_comp);
4865 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
4866 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
4867 rex->engine, NULL, NULL,
4868 /* copy /msix etc to inner pattern */
4873 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4875 /* This isn't a first class regexp. Instead, it's
4876 caching a regexp onto an existing, Perl visible
4878 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
4881 /* safe to do now that any $1 etc has been
4882 * interpolated into the new pattern string and
4884 S_regcp_restore(aTHX_ rex, runops_cp);
4886 re = (struct regexp *)SvANY(re_sv);
4888 RXp_MATCH_COPIED_off(re);
4889 re->subbeg = rex->subbeg;
4890 re->sublen = rex->sublen;
4891 re->suboffset = rex->suboffset;
4892 re->subcoffset = rex->subcoffset;
4895 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4896 "Matching embedded");
4898 startpoint = rei->program + 1;
4899 ST.close_paren = 0; /* only used for GOSUB */
4901 eval_recurse_doit: /* Share code with GOSUB below this line */
4902 /* run the pattern returned from (??{...}) */
4903 ST.cp = regcppush(rex, 0); /* Save *all* the positions. */
4904 REGCP_SET(ST.lastcp);
4907 re->lastcloseparen = 0;
4911 /* XXXX This is too dramatic a measure... */
4914 ST.toggle_reg_flags = PL_reg_flags;
4916 PL_reg_flags |= RF_utf8;
4918 PL_reg_flags &= ~RF_utf8;
4919 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4921 ST.prev_rex = rex_sv;
4922 ST.prev_curlyx = cur_curlyx;
4924 SET_reg_curpm(rex_sv);
4929 ST.prev_eval = cur_eval;
4931 /* now continue from first node in postoned RE */
4932 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
4933 assert(0); /* NOTREACHED */
4936 case EVAL_AB: /* cleanup after a successful (??{A})B */
4937 /* note: this is called twice; first after popping B, then A */
4938 PL_reg_flags ^= ST.toggle_reg_flags;
4939 rex_sv = ST.prev_rex;
4940 SET_reg_curpm(rex_sv);
4941 rex = (struct regexp *)SvANY(rex_sv);
4942 rexi = RXi_GET(rex);
4944 cur_eval = ST.prev_eval;
4945 cur_curlyx = ST.prev_curlyx;
4947 /* XXXX This is too dramatic a measure... */
4949 if ( nochange_depth )
4954 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4955 /* note: this is called twice; first after popping B, then A */
4956 PL_reg_flags ^= ST.toggle_reg_flags;
4957 rex_sv = ST.prev_rex;
4958 SET_reg_curpm(rex_sv);
4959 rex = (struct regexp *)SvANY(rex_sv);
4960 rexi = RXi_GET(rex);
4962 REGCP_UNWIND(ST.lastcp);
4964 cur_eval = ST.prev_eval;
4965 cur_curlyx = ST.prev_curlyx;
4966 /* XXXX This is too dramatic a measure... */
4968 if ( nochange_depth )
4974 n = ARG(scan); /* which paren pair */
4975 rex->offs[n].start_tmp = locinput - PL_bostr;
4978 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
4979 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; regsize=%"UVuf"\n",
4983 (IV)rex->offs[n].start_tmp,
4989 /* XXX really need to log other places start/end are set too */
4990 #define CLOSE_CAPTURE \
4991 rex->offs[n].start = rex->offs[n].start_tmp; \
4992 rex->offs[n].end = locinput - PL_bostr; \
4993 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
4994 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
4996 PTR2UV(rex->offs), \
4998 (IV)rex->offs[n].start, \
4999 (IV)rex->offs[n].end \
5003 n = ARG(scan); /* which paren pair */
5005 /*if (n > PL_regsize)
5007 if (n > rex->lastparen)
5009 rex->lastcloseparen = n;
5010 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5015 case ACCEPT: /* (*ACCEPT) */
5019 cursor && OP(cursor)!=END;
5020 cursor=regnext(cursor))
5022 if ( OP(cursor)==CLOSE ){
5024 if ( n <= lastopen ) {
5026 /*if (n > PL_regsize)
5028 if (n > rex->lastparen)
5030 rex->lastcloseparen = n;
5031 if ( n == ARG(scan) || (cur_eval &&
5032 cur_eval->u.eval.close_paren == n))
5041 case GROUPP: /* (?(1)) */
5042 n = ARG(scan); /* which paren pair */
5043 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5046 case NGROUPP: /* (?(<name>)) */
5047 /* reg_check_named_buff_matched returns 0 for no match */
5048 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5051 case INSUBP: /* (?(R)) */
5053 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5056 case DEFINEP: /* (?(DEFINE)) */
5060 case IFTHEN: /* (?(cond)A|B) */
5061 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
5063 next = NEXTOPER(NEXTOPER(scan));
5065 next = scan + ARG(scan);
5066 if (OP(next) == IFTHEN) /* Fake one. */
5067 next = NEXTOPER(NEXTOPER(next));
5071 case LOGICAL: /* modifier for EVAL and IFMATCH */
5072 logical = scan->flags;
5075 /*******************************************************************
5077 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5078 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5079 STAR/PLUS/CURLY/CURLYN are used instead.)
5081 A*B is compiled as <CURLYX><A><WHILEM><B>
5083 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5084 state, which contains the current count, initialised to -1. It also sets
5085 cur_curlyx to point to this state, with any previous value saved in the
5088 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5089 since the pattern may possibly match zero times (i.e. it's a while {} loop
5090 rather than a do {} while loop).
5092 Each entry to WHILEM represents a successful match of A. The count in the
5093 CURLYX block is incremented, another WHILEM state is pushed, and execution
5094 passes to A or B depending on greediness and the current count.
5096 For example, if matching against the string a1a2a3b (where the aN are
5097 substrings that match /A/), then the match progresses as follows: (the
5098 pushed states are interspersed with the bits of strings matched so far):
5101 <CURLYX cnt=0><WHILEM>
5102 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5103 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5104 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5105 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5107 (Contrast this with something like CURLYM, which maintains only a single
5111 a1 <CURLYM cnt=1> a2
5112 a1 a2 <CURLYM cnt=2> a3
5113 a1 a2 a3 <CURLYM cnt=3> b
5116 Each WHILEM state block marks a point to backtrack to upon partial failure
5117 of A or B, and also contains some minor state data related to that
5118 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5119 overall state, such as the count, and pointers to the A and B ops.
5121 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5122 must always point to the *current* CURLYX block, the rules are:
5124 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5125 and set cur_curlyx to point the new block.
5127 When popping the CURLYX block after a successful or unsuccessful match,
5128 restore the previous cur_curlyx.
5130 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5131 to the outer one saved in the CURLYX block.
5133 When popping the WHILEM block after a successful or unsuccessful B match,
5134 restore the previous cur_curlyx.
5136 Here's an example for the pattern (AI* BI)*BO
5137 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5140 curlyx backtrack stack
5141 ------ ---------------
5143 CO <CO prev=NULL> <WO>
5144 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5145 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5146 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5148 At this point the pattern succeeds, and we work back down the stack to
5149 clean up, restoring as we go:
5151 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5152 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5153 CO <CO prev=NULL> <WO>
5156 *******************************************************************/
5158 #define ST st->u.curlyx
5160 case CURLYX: /* start of /A*B/ (for complex A) */
5162 /* No need to save/restore up to this paren */
5163 I32 parenfloor = scan->flags;
5165 assert(next); /* keep Coverity happy */
5166 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5169 /* XXXX Probably it is better to teach regpush to support
5170 parenfloor > PL_regsize... */
5171 if (parenfloor > (I32)rex->lastparen)
5172 parenfloor = rex->lastparen; /* Pessimization... */
5174 ST.prev_curlyx= cur_curlyx;
5176 ST.cp = PL_savestack_ix;
5178 /* these fields contain the state of the current curly.
5179 * they are accessed by subsequent WHILEMs */
5180 ST.parenfloor = parenfloor;
5185 ST.count = -1; /* this will be updated by WHILEM */
5186 ST.lastloc = NULL; /* this will be updated by WHILEM */
5188 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5189 assert(0); /* NOTREACHED */
5192 case CURLYX_end: /* just finished matching all of A*B */
5193 cur_curlyx = ST.prev_curlyx;
5195 assert(0); /* NOTREACHED */
5197 case CURLYX_end_fail: /* just failed to match all of A*B */
5199 cur_curlyx = ST.prev_curlyx;
5201 assert(0); /* NOTREACHED */
5205 #define ST st->u.whilem
5207 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5209 /* see the discussion above about CURLYX/WHILEM */
5211 int min = ARG1(cur_curlyx->u.curlyx.me);
5212 int max = ARG2(cur_curlyx->u.curlyx.me);
5213 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5215 assert(cur_curlyx); /* keep Coverity happy */
5216 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5217 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5218 ST.cache_offset = 0;
5222 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5223 "%*s whilem: matched %ld out of %d..%d\n",
5224 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5227 /* First just match a string of min A's. */
5230 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5231 cur_curlyx->u.curlyx.lastloc = locinput;
5232 REGCP_SET(ST.lastcp);
5234 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5235 assert(0); /* NOTREACHED */
5238 /* If degenerate A matches "", assume A done. */
5240 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5241 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5242 "%*s whilem: empty match detected, trying continuation...\n",
5243 REPORT_CODE_OFF+depth*2, "")
5245 goto do_whilem_B_max;
5248 /* super-linear cache processing */
5252 if (!PL_reg_maxiter) {
5253 /* start the countdown: Postpone detection until we
5254 * know the match is not *that* much linear. */
5255 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
5256 /* possible overflow for long strings and many CURLYX's */
5257 if (PL_reg_maxiter < 0)
5258 PL_reg_maxiter = I32_MAX;
5259 PL_reg_leftiter = PL_reg_maxiter;
5262 if (PL_reg_leftiter-- == 0) {
5263 /* initialise cache */
5264 const I32 size = (PL_reg_maxiter + 7)/8;
5265 if (PL_reg_poscache) {
5266 if ((I32)PL_reg_poscache_size < size) {
5267 Renew(PL_reg_poscache, size, char);
5268 PL_reg_poscache_size = size;
5270 Zero(PL_reg_poscache, size, char);
5273 PL_reg_poscache_size = size;
5274 Newxz(PL_reg_poscache, size, char);
5276 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5277 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5278 PL_colors[4], PL_colors[5])
5282 if (PL_reg_leftiter < 0) {
5283 /* have we already failed at this position? */
5285 offset = (scan->flags & 0xf) - 1
5286 + (locinput - PL_bostr) * (scan->flags>>4);
5287 mask = 1 << (offset % 8);
5289 if (PL_reg_poscache[offset] & mask) {
5290 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5291 "%*s whilem: (cache) already tried at this position...\n",
5292 REPORT_CODE_OFF+depth*2, "")
5294 sayNO; /* cache records failure */
5296 ST.cache_offset = offset;
5297 ST.cache_mask = mask;
5301 /* Prefer B over A for minimal matching. */
5303 if (cur_curlyx->u.curlyx.minmod) {
5304 ST.save_curlyx = cur_curlyx;
5305 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5306 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor);
5307 REGCP_SET(ST.lastcp);
5308 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5310 assert(0); /* NOTREACHED */
5313 /* Prefer A over B for maximal matching. */
5315 if (n < max) { /* More greed allowed? */
5316 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5317 cur_curlyx->u.curlyx.lastloc = locinput;
5318 REGCP_SET(ST.lastcp);
5319 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5320 assert(0); /* NOTREACHED */
5322 goto do_whilem_B_max;
5324 assert(0); /* NOTREACHED */
5326 case WHILEM_B_min: /* just matched B in a minimal match */
5327 case WHILEM_B_max: /* just matched B in a maximal match */
5328 cur_curlyx = ST.save_curlyx;
5330 assert(0); /* NOTREACHED */
5332 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5333 cur_curlyx = ST.save_curlyx;
5334 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5335 cur_curlyx->u.curlyx.count--;
5337 assert(0); /* NOTREACHED */
5339 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5341 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5342 REGCP_UNWIND(ST.lastcp);
5344 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5345 cur_curlyx->u.curlyx.count--;
5347 assert(0); /* NOTREACHED */
5349 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5350 REGCP_UNWIND(ST.lastcp);
5351 regcppop(rex); /* Restore some previous $<digit>s? */
5352 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5353 "%*s whilem: failed, trying continuation...\n",
5354 REPORT_CODE_OFF+depth*2, "")
5357 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5358 && ckWARN(WARN_REGEXP)
5359 && !(PL_reg_flags & RF_warned))
5361 PL_reg_flags |= RF_warned;
5362 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5363 "Complex regular subexpression recursion limit (%d) "
5369 ST.save_curlyx = cur_curlyx;
5370 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5371 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5373 assert(0); /* NOTREACHED */
5375 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5376 cur_curlyx = ST.save_curlyx;
5377 REGCP_UNWIND(ST.lastcp);
5380 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5381 /* Maximum greed exceeded */
5382 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5383 && ckWARN(WARN_REGEXP)
5384 && !(PL_reg_flags & RF_warned))
5386 PL_reg_flags |= RF_warned;
5387 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5388 "Complex regular subexpression recursion "
5389 "limit (%d) exceeded",
5392 cur_curlyx->u.curlyx.count--;
5396 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5397 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5399 /* Try grabbing another A and see if it helps. */
5400 cur_curlyx->u.curlyx.lastloc = locinput;
5401 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5402 REGCP_SET(ST.lastcp);
5403 PUSH_STATE_GOTO(WHILEM_A_min,
5404 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5406 assert(0); /* NOTREACHED */
5409 #define ST st->u.branch
5411 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5412 next = scan + ARG(scan);
5415 scan = NEXTOPER(scan);
5418 case BRANCH: /* /(...|A|...)/ */
5419 scan = NEXTOPER(scan); /* scan now points to inner node */
5420 ST.lastparen = rex->lastparen;
5421 ST.lastcloseparen = rex->lastcloseparen;
5422 ST.next_branch = next;
5425 /* Now go into the branch */
5427 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5429 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5431 assert(0); /* NOTREACHED */
5433 case CUTGROUP: /* /(*THEN)/ */
5434 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5435 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5436 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
5437 assert(0); /* NOTREACHED */
5439 case CUTGROUP_next_fail:
5442 if (st->u.mark.mark_name)
5443 sv_commit = st->u.mark.mark_name;
5445 assert(0); /* NOTREACHED */
5449 assert(0); /* NOTREACHED */
5451 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5456 REGCP_UNWIND(ST.cp);
5457 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5458 scan = ST.next_branch;
5459 /* no more branches? */
5460 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5462 PerlIO_printf( Perl_debug_log,
5463 "%*s %sBRANCH failed...%s\n",
5464 REPORT_CODE_OFF+depth*2, "",
5470 continue; /* execute next BRANCH[J] op */
5471 assert(0); /* NOTREACHED */
5473 case MINMOD: /* next op will be non-greedy, e.g. A*? */
5478 #define ST st->u.curlym
5480 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5482 /* This is an optimisation of CURLYX that enables us to push
5483 * only a single backtracking state, no matter how many matches
5484 * there are in {m,n}. It relies on the pattern being constant
5485 * length, with no parens to influence future backrefs
5489 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5491 ST.lastparen = rex->lastparen;
5492 ST.lastcloseparen = rex->lastcloseparen;
5494 /* if paren positive, emulate an OPEN/CLOSE around A */
5496 U32 paren = ST.me->flags;
5497 if (paren > PL_regsize)
5499 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5507 ST.c1 = CHRTEST_UNINIT;
5510 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5513 curlym_do_A: /* execute the A in /A{m,n}B/ */
5514 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
5515 assert(0); /* NOTREACHED */
5517 case CURLYM_A: /* we've just matched an A */
5519 /* after first match, determine A's length: u.curlym.alen */
5520 if (ST.count == 1) {
5521 if (PL_reg_match_utf8) {
5522 char *s = st->locinput;
5523 while (s < locinput) {
5529 ST.alen = locinput - st->locinput;
5532 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5535 PerlIO_printf(Perl_debug_log,
5536 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5537 (int)(REPORT_CODE_OFF+(depth*2)), "",
5538 (IV) ST.count, (IV)ST.alen)
5541 if (cur_eval && cur_eval->u.eval.close_paren &&
5542 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5546 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5547 if ( max == REG_INFTY || ST.count < max )
5548 goto curlym_do_A; /* try to match another A */
5550 goto curlym_do_B; /* try to match B */
5552 case CURLYM_A_fail: /* just failed to match an A */
5553 REGCP_UNWIND(ST.cp);
5555 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5556 || (cur_eval && cur_eval->u.eval.close_paren &&
5557 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5560 curlym_do_B: /* execute the B in /A{m,n}B/ */
5561 if (ST.c1 == CHRTEST_UNINIT) {
5562 /* calculate c1 and c2 for possible match of 1st char
5563 * following curly */
5564 ST.c1 = ST.c2 = CHRTEST_VOID;
5565 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5566 regnode *text_node = ST.B;
5567 if (! HAS_TEXT(text_node))
5568 FIND_NEXT_IMPT(text_node);
5571 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5573 But the former is redundant in light of the latter.
5575 if this changes back then the macro for
5576 IS_TEXT and friends need to change.
5578 if (PL_regkind[OP(text_node)] == EXACT) {
5579 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ text_node,
5589 PerlIO_printf(Perl_debug_log,
5590 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5591 (int)(REPORT_CODE_OFF+(depth*2)),
5594 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
5595 const UV c = (utf8_target)
5596 ? utf8n_to_uvchr((U8*)locinput,
5597 UTF8_MAXBYTES, NULL,
5600 if (c != (UV) ST.c1 && c != (UV) ST.c2) {
5601 /* simulate B failing */
5603 PerlIO_printf(Perl_debug_log,
5604 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5605 (int)(REPORT_CODE_OFF+(depth*2)),"",
5608 state_num = CURLYM_B_fail;
5609 goto reenter_switch;
5614 /* emulate CLOSE: mark current A as captured */
5615 I32 paren = ST.me->flags;
5617 rex->offs[paren].start
5618 = HOPc(locinput, -ST.alen) - PL_bostr;
5619 rex->offs[paren].end = locinput - PL_bostr;
5620 if ((U32)paren > rex->lastparen)
5621 rex->lastparen = paren;
5622 rex->lastcloseparen = paren;
5625 rex->offs[paren].end = -1;
5626 if (cur_eval && cur_eval->u.eval.close_paren &&
5627 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5636 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
5637 assert(0); /* NOTREACHED */
5639 case CURLYM_B_fail: /* just failed to match a B */
5640 REGCP_UNWIND(ST.cp);
5641 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5643 I32 max = ARG2(ST.me);
5644 if (max != REG_INFTY && ST.count == max)
5646 goto curlym_do_A; /* try to match a further A */
5648 /* backtrack one A */
5649 if (ST.count == ARG1(ST.me) /* min */)
5652 SET_locinput(HOPc(locinput, -ST.alen));
5653 goto curlym_do_B; /* try to match B */
5656 #define ST st->u.curly
5658 #define CURLY_SETPAREN(paren, success) \
5661 rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5662 rex->offs[paren].end = locinput - PL_bostr; \
5663 if (paren > rex->lastparen) \
5664 rex->lastparen = paren; \
5665 rex->lastcloseparen = paren; \
5668 rex->offs[paren].end = -1; \
5669 rex->lastparen = ST.lastparen; \
5670 rex->lastcloseparen = ST.lastcloseparen; \
5674 case STAR: /* /A*B/ where A is width 1 char */
5678 scan = NEXTOPER(scan);
5681 case PLUS: /* /A+B/ where A is width 1 char */
5685 scan = NEXTOPER(scan);
5688 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
5689 ST.paren = scan->flags; /* Which paren to set */
5690 ST.lastparen = rex->lastparen;
5691 ST.lastcloseparen = rex->lastcloseparen;
5692 if (ST.paren > PL_regsize)
5693 PL_regsize = ST.paren;
5694 ST.min = ARG1(scan); /* min to match */
5695 ST.max = ARG2(scan); /* max to match */
5696 if (cur_eval && cur_eval->u.eval.close_paren &&
5697 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5701 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5704 case CURLY: /* /A{m,n}B/ where A is width 1 char */
5706 ST.min = ARG1(scan); /* min to match */
5707 ST.max = ARG2(scan); /* max to match */
5708 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5711 * Lookahead to avoid useless match attempts
5712 * when we know what character comes next.
5714 * Used to only do .*x and .*?x, but now it allows
5715 * for )'s, ('s and (?{ ... })'s to be in the way
5716 * of the quantifier and the EXACT-like node. -- japhy
5719 assert(ST.min <= ST.max);
5720 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
5721 ST.c1 = ST.c2 = CHRTEST_VOID;
5724 regnode *text_node = next;
5726 if (! HAS_TEXT(text_node))
5727 FIND_NEXT_IMPT(text_node);
5729 if (! HAS_TEXT(text_node))
5730 ST.c1 = ST.c2 = CHRTEST_VOID;
5732 if ( PL_regkind[OP(text_node)] != EXACT ) {
5733 ST.c1 = ST.c2 = CHRTEST_VOID;
5737 /* Currently we only get here when
5739 PL_rekind[OP(text_node)] == EXACT
5741 if this changes back then the macro for IS_TEXT and
5742 friends need to change. */
5743 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ text_node,
5755 char *li = locinput;
5757 if (ST.min && regrepeat(rex, &li, ST.A, ST.min, depth) < ST.min)
5762 if (ST.c1 == CHRTEST_VOID)
5763 goto curly_try_B_min;
5765 ST.oldloc = locinput;
5767 /* set ST.maxpos to the furthest point along the
5768 * string that could possibly match */
5769 if (ST.max == REG_INFTY) {
5770 ST.maxpos = PL_regeol - 1;
5772 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5775 else if (utf8_target) {
5776 int m = ST.max - ST.min;
5777 for (ST.maxpos = locinput;
5778 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5779 ST.maxpos += UTF8SKIP(ST.maxpos);
5782 ST.maxpos = locinput + ST.max - ST.min;
5783 if (ST.maxpos >= PL_regeol)
5784 ST.maxpos = PL_regeol - 1;
5786 goto curly_try_B_min_known;
5790 /* avoid taking address of locinput, so it can remain
5792 char *li = locinput;
5793 ST.count = regrepeat(rex, &li, ST.A, ST.max, depth);
5794 if (ST.count < ST.min)
5797 if ((ST.count > ST.min)
5798 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5800 /* A{m,n} must come at the end of the string, there's
5801 * no point in backing off ... */
5803 /* ...except that $ and \Z can match before *and* after
5804 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5805 We may back off by one in this case. */
5806 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
5810 goto curly_try_B_max;
5812 assert(0); /* NOTREACHED */
5815 case CURLY_B_min_known_fail:
5816 /* failed to find B in a non-greedy match where c1,c2 valid */
5818 REGCP_UNWIND(ST.cp);
5820 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5822 /* Couldn't or didn't -- move forward. */
5823 ST.oldloc = locinput;
5825 locinput += UTF8SKIP(locinput);
5829 curly_try_B_min_known:
5830 /* find the next place where 'B' could work, then call B */
5834 n = (ST.oldloc == locinput) ? 0 : 1;
5835 if (ST.c1 == ST.c2) {
5837 /* set n to utf8_distance(oldloc, locinput) */
5838 while (locinput <= ST.maxpos &&
5839 utf8n_to_uvchr((U8*)locinput,
5840 UTF8_MAXBYTES, &len,
5841 uniflags) != (UV)ST.c1) {
5847 /* set n to utf8_distance(oldloc, locinput) */
5848 while (locinput <= ST.maxpos) {
5850 const UV c = utf8n_to_uvchr((U8*)locinput,
5851 UTF8_MAXBYTES, &len,
5853 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5860 else { /* Not utf8_target */
5861 if (ST.c1 == ST.c2) {
5862 while (locinput <= ST.maxpos &&
5863 UCHARAT(locinput) != ST.c1)
5867 while (locinput <= ST.maxpos
5868 && UCHARAT(locinput) != ST.c1
5869 && UCHARAT(locinput) != ST.c2)
5872 n = locinput - ST.oldloc;
5874 if (locinput > ST.maxpos)
5877 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
5878 * at b; check that everything between oldloc and
5879 * locinput matches */
5880 char *li = ST.oldloc;
5882 if (regrepeat(rex, &li, ST.A, n, depth) < n)
5884 assert(n == REG_INFTY || locinput == li);
5886 CURLY_SETPAREN(ST.paren, ST.count);
5887 if (cur_eval && cur_eval->u.eval.close_paren &&
5888 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5891 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
5893 assert(0); /* NOTREACHED */
5896 case CURLY_B_min_fail:
5897 /* failed to find B in a non-greedy match where c1,c2 invalid */
5899 REGCP_UNWIND(ST.cp);
5901 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5903 /* failed -- move forward one */
5905 char *li = locinput;
5906 if (!regrepeat(rex, &li, ST.A, 1, depth)) {
5913 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5914 ST.count > 0)) /* count overflow ? */
5917 CURLY_SETPAREN(ST.paren, ST.count);
5918 if (cur_eval && cur_eval->u.eval.close_paren &&
5919 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5922 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
5926 assert(0); /* NOTREACHED */
5930 /* a successful greedy match: now try to match B */
5931 if (cur_eval && cur_eval->u.eval.close_paren &&
5932 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5937 if (ST.c1 != CHRTEST_VOID && locinput < PL_regeol)
5938 c = utf8_target ? utf8n_to_uvchr((U8*)locinput,
5939 UTF8_MAXBYTES, 0, uniflags)
5940 : (UV) UCHARAT(locinput);
5941 /* If it could work, try it. */
5942 if (ST.c1 == CHRTEST_VOID
5943 || (locinput < PL_regeol &&
5944 (c == (UV)ST.c1 || c == (UV)ST.c2)))
5946 CURLY_SETPAREN(ST.paren, ST.count);
5947 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
5948 assert(0); /* NOTREACHED */
5953 case CURLY_B_max_fail:
5954 /* failed to find B in a greedy match */
5956 REGCP_UNWIND(ST.cp);
5958 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5961 if (--ST.count < ST.min)
5963 locinput = HOPc(locinput, -1);
5964 goto curly_try_B_max;
5968 case END: /* last op of main pattern */
5971 /* we've just finished A in /(??{A})B/; now continue with B */
5972 st->u.eval.toggle_reg_flags
5973 = cur_eval->u.eval.toggle_reg_flags;
5974 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5976 st->u.eval.prev_rex = rex_sv; /* inner */
5977 st->u.eval.cp = regcppush(rex, 0); /* Save *all* the positions. */
5978 rex_sv = cur_eval->u.eval.prev_rex;
5979 SET_reg_curpm(rex_sv);
5980 rex = (struct regexp *)SvANY(rex_sv);
5981 rexi = RXi_GET(rex);
5982 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5984 REGCP_SET(st->u.eval.lastcp);
5986 /* Restore parens of the outer rex without popping the
5988 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp);
5990 st->u.eval.prev_eval = cur_eval;
5991 cur_eval = cur_eval->u.eval.prev_eval;
5993 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5994 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5995 if ( nochange_depth )
5998 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
5999 locinput); /* match B */
6002 if (locinput < reginfo->till) {
6003 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6004 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6006 (long)(locinput - PL_reg_starttry),
6007 (long)(reginfo->till - PL_reg_starttry),
6010 sayNO_SILENT; /* Cannot match: too short. */
6012 sayYES; /* Success! */
6014 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6016 PerlIO_printf(Perl_debug_log,
6017 "%*s %ssubpattern success...%s\n",
6018 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6019 sayYES; /* Success! */
6022 #define ST st->u.ifmatch
6027 case SUSPEND: /* (?>A) */
6029 newstart = locinput;
6032 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6034 goto ifmatch_trivial_fail_test;
6036 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6038 ifmatch_trivial_fail_test:
6040 char * const s = HOPBACKc(locinput, scan->flags);
6045 sw = 1 - cBOOL(ST.wanted);
6049 next = scan + ARG(scan);
6057 newstart = locinput;
6061 ST.logical = logical;
6062 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6064 /* execute body of (?...A) */
6065 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6066 assert(0); /* NOTREACHED */
6069 case IFMATCH_A_fail: /* body of (?...A) failed */
6070 ST.wanted = !ST.wanted;
6073 case IFMATCH_A: /* body of (?...A) succeeded */
6075 sw = cBOOL(ST.wanted);
6077 else if (!ST.wanted)
6080 if (OP(ST.me) != SUSPEND) {
6081 /* restore old position except for (?>...) */
6082 locinput = st->locinput;
6084 scan = ST.me + ARG(ST.me);
6087 continue; /* execute B */
6091 case LONGJMP: /* alternative with many branches compiles to
6092 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6093 next = scan + ARG(scan);
6098 case COMMIT: /* (*COMMIT) */
6099 reginfo->cutpoint = PL_regeol;
6102 case PRUNE: /* (*PRUNE) */
6104 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6105 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6106 assert(0); /* NOTREACHED */
6108 case COMMIT_next_fail:
6112 case OPFAIL: /* (*FAIL) */
6114 assert(0); /* NOTREACHED */
6116 #define ST st->u.mark
6117 case MARKPOINT: /* (*MARK:foo) */
6118 ST.prev_mark = mark_state;
6119 ST.mark_name = sv_commit = sv_yes_mark
6120 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6122 ST.mark_loc = locinput;
6123 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6124 assert(0); /* NOTREACHED */
6126 case MARKPOINT_next:
6127 mark_state = ST.prev_mark;
6129 assert(0); /* NOTREACHED */
6131 case MARKPOINT_next_fail:
6132 if (popmark && sv_eq(ST.mark_name,popmark))
6134 if (ST.mark_loc > startpoint)
6135 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6136 popmark = NULL; /* we found our mark */
6137 sv_commit = ST.mark_name;
6140 PerlIO_printf(Perl_debug_log,
6141 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6142 REPORT_CODE_OFF+depth*2, "",
6143 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6146 mark_state = ST.prev_mark;
6147 sv_yes_mark = mark_state ?
6148 mark_state->u.mark.mark_name : NULL;
6150 assert(0); /* NOTREACHED */
6152 case SKIP: /* (*SKIP) */
6154 /* (*SKIP) : if we fail we cut here*/
6155 ST.mark_name = NULL;
6156 ST.mark_loc = locinput;
6157 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6159 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6160 otherwise do nothing. Meaning we need to scan
6162 regmatch_state *cur = mark_state;
6163 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6166 if ( sv_eq( cur->u.mark.mark_name,
6169 ST.mark_name = find;
6170 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6172 cur = cur->u.mark.prev_mark;
6175 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6178 case SKIP_next_fail:
6180 /* (*CUT:NAME) - Set up to search for the name as we
6181 collapse the stack*/
6182 popmark = ST.mark_name;
6184 /* (*CUT) - No name, we cut here.*/
6185 if (ST.mark_loc > startpoint)
6186 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6187 /* but we set sv_commit to latest mark_name if there
6188 is one so they can test to see how things lead to this
6191 sv_commit=mark_state->u.mark.mark_name;
6195 assert(0); /* NOTREACHED */
6198 case LNBREAK: /* \R */
6199 if ((n=is_LNBREAK_safe(locinput, PL_regeol, utf8_target))) {
6205 #define CASE_CLASS(nAmE) \
6207 if (NEXTCHR_IS_EOS) \
6209 if ((n=is_##nAmE(locinput,utf8_target))) { \
6215 if (NEXTCHR_IS_EOS) \
6217 if ((n=is_##nAmE(locinput,utf8_target))) { \
6220 locinput += UTF8SKIP(locinput); \
6224 CASE_CLASS(VERTWS); /* \v \V */
6225 CASE_CLASS(HORIZWS); /* \h \H */
6229 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6230 PTR2UV(scan), OP(scan));
6231 Perl_croak(aTHX_ "regexp memory corruption");
6233 /* this is a point to jump to in order to increment
6234 * locinput by one character */
6237 locinput += PL_utf8skip[nextchr];
6238 /* locinput is allowed to go 1 char off the end, but not 2+ */
6239 if (locinput > PL_regeol)
6248 /* switch break jumps here */
6249 scan = next; /* prepare to execute the next op and ... */
6250 continue; /* ... jump back to the top, reusing st */
6251 assert(0); /* NOTREACHED */
6254 /* push a state that backtracks on success */
6255 st->u.yes.prev_yes_state = yes_state;
6259 /* push a new regex state, then continue at scan */
6261 regmatch_state *newst;
6264 regmatch_state *cur = st;
6265 regmatch_state *curyes = yes_state;
6267 regmatch_slab *slab = PL_regmatch_slab;
6268 for (;curd > -1;cur--,curd--) {
6269 if (cur < SLAB_FIRST(slab)) {
6271 cur = SLAB_LAST(slab);
6273 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6274 REPORT_CODE_OFF + 2 + depth * 2,"",
6275 curd, PL_reg_name[cur->resume_state],
6276 (curyes == cur) ? "yes" : ""
6279 curyes = cur->u.yes.prev_yes_state;
6282 DEBUG_STATE_pp("push")
6285 st->locinput = locinput;
6287 if (newst > SLAB_LAST(PL_regmatch_slab))
6288 newst = S_push_slab(aTHX);
6289 PL_regmatch_state = newst;
6291 locinput = pushinput;
6294 assert(0); /* NOTREACHED */
6299 * We get here only if there's trouble -- normally "case END" is
6300 * the terminating point.
6302 Perl_croak(aTHX_ "corrupted regexp pointers");
6308 /* we have successfully completed a subexpression, but we must now
6309 * pop to the state marked by yes_state and continue from there */
6310 assert(st != yes_state);
6312 while (st != yes_state) {
6314 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6315 PL_regmatch_slab = PL_regmatch_slab->prev;
6316 st = SLAB_LAST(PL_regmatch_slab);
6320 DEBUG_STATE_pp("pop (no final)");
6322 DEBUG_STATE_pp("pop (yes)");
6328 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6329 || yes_state > SLAB_LAST(PL_regmatch_slab))
6331 /* not in this slab, pop slab */
6332 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6333 PL_regmatch_slab = PL_regmatch_slab->prev;
6334 st = SLAB_LAST(PL_regmatch_slab);
6336 depth -= (st - yes_state);
6339 yes_state = st->u.yes.prev_yes_state;
6340 PL_regmatch_state = st;
6343 locinput= st->locinput;
6344 state_num = st->resume_state + no_final;
6345 goto reenter_switch;
6348 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6349 PL_colors[4], PL_colors[5]));
6351 if (PL_reg_state.re_state_eval_setup_done) {
6352 /* each successfully executed (?{...}) block does the equivalent of
6353 * local $^R = do {...}
6354 * When popping the save stack, all these locals would be undone;
6355 * bypass this by setting the outermost saved $^R to the latest
6357 if (oreplsv != GvSV(PL_replgv))
6358 sv_setsv(oreplsv, GvSV(PL_replgv));
6365 PerlIO_printf(Perl_debug_log,
6366 "%*s %sfailed...%s\n",
6367 REPORT_CODE_OFF+depth*2, "",
6368 PL_colors[4], PL_colors[5])
6380 /* there's a previous state to backtrack to */
6382 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6383 PL_regmatch_slab = PL_regmatch_slab->prev;
6384 st = SLAB_LAST(PL_regmatch_slab);
6386 PL_regmatch_state = st;
6387 locinput= st->locinput;
6389 DEBUG_STATE_pp("pop");
6391 if (yes_state == st)
6392 yes_state = st->u.yes.prev_yes_state;
6394 state_num = st->resume_state + 1; /* failure = success + 1 */
6395 goto reenter_switch;
6400 if (rex->intflags & PREGf_VERBARG_SEEN) {
6401 SV *sv_err = get_sv("REGERROR", 1);
6402 SV *sv_mrk = get_sv("REGMARK", 1);
6404 sv_commit = &PL_sv_no;
6406 sv_yes_mark = &PL_sv_yes;
6409 sv_commit = &PL_sv_yes;
6410 sv_yes_mark = &PL_sv_no;
6412 sv_setsv(sv_err, sv_commit);
6413 sv_setsv(sv_mrk, sv_yes_mark);
6417 if (last_pushed_cv) {
6420 PERL_UNUSED_VAR(SP);
6423 /* clean up; in particular, free all slabs above current one */
6424 LEAVE_SCOPE(oldsave);
6426 assert(!result || locinput - PL_bostr >= 0);
6427 return result ? locinput - PL_bostr : -1;
6431 - regrepeat - repeatedly match something simple, report how many
6433 * startposp - pointer a pointer to the start position. This is updated
6434 * to point to the byte following the highest successful
6436 * p - the regnode to be repeatedly matched against.
6437 * max - maximum number of characters to match.
6438 * depth - (for debugging) backtracking depth.
6441 S_regrepeat(pTHX_ const regexp *prog, char **startposp, const regnode *p, I32 max, int depth)
6446 char *loceol = PL_regeol;
6448 bool utf8_target = PL_reg_match_utf8;
6451 PERL_UNUSED_ARG(depth);
6454 PERL_ARGS_ASSERT_REGREPEAT;
6457 if (max == REG_INFTY)
6459 else if (max < loceol - scan)
6460 loceol = scan + max;
6465 while (scan < loceol && hardcount < max && *scan != '\n') {
6466 scan += UTF8SKIP(scan);
6470 while (scan < loceol && *scan != '\n')
6477 while (scan < loceol && hardcount < max) {
6478 scan += UTF8SKIP(scan);
6491 /* Can use a simple loop if the pattern char to match on is invariant
6492 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
6493 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
6494 * true iff it doesn't matter if the argument is in UTF-8 or not */
6495 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! UTF_PATTERN)) {
6496 while (scan < loceol && UCHARAT(scan) == c) {
6500 else if (UTF_PATTERN) {
6502 STRLEN scan_char_len;
6505 /* When both target and pattern are UTF-8, we have to do s
6507 while (hardcount < max
6508 && scan + (scan_char_len = UTF8SKIP(scan)) <= loceol
6509 && scan_char_len <= STR_LEN(p)
6510 && memEQ(scan, STRING(p), scan_char_len))
6512 scan += scan_char_len;
6516 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
6518 /* Target isn't utf8; convert the character in the UTF-8
6519 * pattern to non-UTF8, and do a simple loop */
6520 c = TWO_BYTE_UTF8_TO_UNI(c, *(STRING(p) + 1));
6521 while (scan < loceol && UCHARAT(scan) == c) {
6524 } /* else pattern char is above Latin1, can't possibly match the
6529 /* Here, the string must be utf8; pattern isn't, and <c> is
6530 * different in utf8 than not, so can't compare them directly.
6531 * Outside the loop, find the two utf8 bytes that represent c, and
6532 * then look for those in sequence in the utf8 string */
6533 U8 high = UTF8_TWO_BYTE_HI(c);
6534 U8 low = UTF8_TWO_BYTE_LO(c);
6537 while (hardcount < max
6538 && scan + 1 < loceol
6539 && UCHARAT(scan) == high
6540 && UCHARAT(scan + 1) == low)
6549 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6553 PL_reg_flags |= RF_tainted;
6554 utf8_flags = FOLDEQ_UTF8_LOCALE;
6562 case EXACTFU_TRICKYFOLD:
6564 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6570 || OP(p) == EXACTFU_SS
6571 || (UTF_PATTERN && ! UTF8_IS_INVARIANT(c)))
6573 /* Use full Unicode fold matching */
6574 char *tmpeol = loceol;
6575 STRLEN pat_len = (UTF_PATTERN) ? UTF8SKIP(STRING(p)) : 1;
6576 while (hardcount < max
6577 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6578 STRING(p), NULL, pat_len, cBOOL(UTF_PATTERN), utf8_flags))
6585 /* XXX Note that the above handles properly the German sharp s in
6586 * the pattern matching ss in the string. But it doesn't handle
6587 * properly cases where the string contains say 'LIGATURE ff' and
6588 * the pattern is 'f+'. This would require, say, a new function or
6589 * revised interface to foldEQ_utf8(), in which the maximum number
6590 * of characters to match could be passed and it would return how
6591 * many actually did. This is just one of many cases where
6592 * multi-char folds don't work properly, and so the fix is being
6598 /* Here, the string isn't utf8; and either the pattern isn't utf8
6599 * or c is an invariant, so its utf8ness doesn't affect c. Can
6600 * just do simple comparisons for exact or fold matching. */
6602 case EXACTF: folded = PL_fold[c]; break;
6604 case EXACTFU_TRICKYFOLD:
6605 case EXACTFU: folded = PL_fold_latin1[c]; break;
6606 case EXACTFL: folded = PL_fold_locale[c]; break;
6607 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6609 while (scan < loceol &&
6610 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6618 if (utf8_target || OP(p) == ANYOFV) {
6621 inclasslen = loceol - scan;
6622 while (hardcount < max
6623 && ((inclasslen = loceol - scan) > 0)
6624 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6630 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6638 LOAD_UTF8_CHARCLASS_ALNUM();
6639 while (hardcount < max && scan < loceol &&
6640 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6642 scan += UTF8SKIP(scan);
6646 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6654 while (scan < loceol && isALNUM((U8) *scan)) {
6659 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6664 PL_reg_flags |= RF_tainted;
6667 while (hardcount < max && scan < loceol &&
6668 isALNUM_LC_utf8((U8*)scan)) {
6669 scan += UTF8SKIP(scan);
6673 while (scan < loceol && isALNUM_LC(*scan))
6683 LOAD_UTF8_CHARCLASS_ALNUM();
6684 while (hardcount < max && scan < loceol &&
6685 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6687 scan += UTF8SKIP(scan);
6691 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6698 goto utf8_Nwordchar;
6699 while (scan < loceol && ! isALNUM((U8) *scan)) {
6705 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
6711 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6712 scan += UTF8SKIP(scan);
6716 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6723 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6724 scan += UTF8SKIP(scan);
6728 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6734 PL_reg_flags |= RF_tainted;
6737 while (hardcount < max && scan < loceol &&
6738 !isALNUM_LC_utf8((U8*)scan)) {
6739 scan += UTF8SKIP(scan);
6743 while (scan < loceol && !isALNUM_LC(*scan))
6753 LOAD_UTF8_CHARCLASS_SPACE();
6754 while (hardcount < max && scan < loceol &&
6756 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6758 scan += UTF8SKIP(scan);
6764 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6773 while (scan < loceol && isSPACE((U8) *scan)) {
6778 while (scan < loceol && isSPACE_A((U8) *scan)) {
6783 PL_reg_flags |= RF_tainted;
6786 while (hardcount < max && scan < loceol &&
6787 isSPACE_LC_utf8((U8*)scan)) {
6788 scan += UTF8SKIP(scan);
6792 while (scan < loceol && isSPACE_LC(*scan))
6802 LOAD_UTF8_CHARCLASS_SPACE();
6803 while (hardcount < max && scan < loceol &&
6805 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6807 scan += UTF8SKIP(scan);
6813 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6822 while (scan < loceol && ! isSPACE((U8) *scan)) {
6828 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6829 scan += UTF8SKIP(scan);
6833 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6839 PL_reg_flags |= RF_tainted;
6842 while (hardcount < max && scan < loceol &&
6843 !isSPACE_LC_utf8((U8*)scan)) {
6844 scan += UTF8SKIP(scan);
6848 while (scan < loceol && !isSPACE_LC(*scan))
6855 LOAD_UTF8_CHARCLASS_DIGIT();
6856 while (hardcount < max && scan < loceol &&
6857 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6858 scan += UTF8SKIP(scan);
6862 while (scan < loceol && isDIGIT(*scan))
6867 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6872 PL_reg_flags |= RF_tainted;
6875 while (hardcount < max && scan < loceol &&
6876 isDIGIT_LC_utf8((U8*)scan)) {
6877 scan += UTF8SKIP(scan);
6881 while (scan < loceol && isDIGIT_LC(*scan))
6888 LOAD_UTF8_CHARCLASS_DIGIT();
6889 while (hardcount < max && scan < loceol &&
6890 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6891 scan += UTF8SKIP(scan);
6895 while (scan < loceol && !isDIGIT(*scan))
6901 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6902 scan += UTF8SKIP(scan);
6906 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6912 PL_reg_flags |= RF_tainted;
6915 while (hardcount < max && scan < loceol &&
6916 !isDIGIT_LC_utf8((U8*)scan)) {
6917 scan += UTF8SKIP(scan);
6921 while (scan < loceol && !isDIGIT_LC(*scan))
6928 while (hardcount < max && scan < loceol &&
6929 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
6935 LNBREAK can match two latin chars, which is ok,
6936 because we have a null terminated string, but we
6937 have to use hardcount in this situation
6939 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
6948 while (hardcount < max && scan < loceol &&
6949 (c=is_HORIZWS_utf8_safe(scan, loceol)))
6955 while (scan < loceol && is_HORIZWS_latin1_safe(scan, loceol))
6962 while (hardcount < max && scan < loceol &&
6963 !is_HORIZWS_utf8_safe(scan, loceol))
6965 scan += UTF8SKIP(scan);
6969 while (scan < loceol && !is_HORIZWS_latin1_safe(scan, loceol))
6977 while (hardcount < max && scan < loceol &&
6978 (c=is_VERTWS_utf8_safe(scan, loceol)))
6984 while (scan < loceol && is_VERTWS_latin1_safe(scan, loceol))
6992 while (hardcount < max && scan < loceol &&
6993 !is_VERTWS_utf8_safe(scan, loceol))
6995 scan += UTF8SKIP(scan);
6999 while (scan < loceol && !is_VERTWS_latin1_safe(scan, loceol))
7005 default: /* Called on something of 0 width. */
7006 break; /* So match right here or not at all. */
7012 c = scan - *startposp;
7016 GET_RE_DEBUG_FLAGS_DECL;
7018 SV * const prop = sv_newmortal();
7019 regprop(prog, prop, p);
7020 PerlIO_printf(Perl_debug_log,
7021 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7022 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7030 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7032 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7033 create a copy so that changes the caller makes won't change the shared one
7036 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7038 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7039 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp, altsvp));
7044 S_core_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7046 /* Returns the swash for the input 'node' in the regex 'prog'.
7047 * If <doinit> is true, will attempt to create the swash if not already
7049 * If <listsvp> is non-null, will return the swash initialization string in
7051 * If <altsvp> is non-null, will return the alternates to the regular swash
7053 * Tied intimately to how regcomp.c sets up the data structure */
7061 RXi_GET_DECL(prog,progi);
7062 const struct reg_data * const data = prog ? progi->data : NULL;
7064 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
7066 assert(ANYOF_NONBITMAP(node));
7068 if (data && data->count) {
7069 const U32 n = ARG(node);
7071 if (data->what[n] == 's') {
7072 SV * const rv = MUTABLE_SV(data->data[n]);
7073 AV * const av = MUTABLE_AV(SvRV(rv));
7074 SV **const ary = AvARRAY(av);
7075 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7077 si = *ary; /* ary[0] = the string to initialize the swash with */
7079 /* Elements 3 and 4 are either both present or both absent. [3] is
7080 * any inversion list generated at compile time; [4] indicates if
7081 * that inversion list has any user-defined properties in it. */
7082 if (av_len(av) >= 3) {
7085 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7092 /* Element [1] is reserved for the set-up swash. If already there,
7093 * return it; if not, create it and store it there */
7094 if (SvROK(ary[1])) {
7097 else if (si && doinit) {
7099 sw = _core_swash_init("utf8", /* the utf8 package */
7103 0, /* not from tr/// */
7106 (void)av_store(av, 1, sw);
7109 /* Element [2] is for any multi-char folds. Note that is a
7110 * fundamentally flawed design, because can't backtrack and try
7111 * again. See [perl #89774] */
7112 if (SvTYPE(ary[2]) == SVt_PVAV) {
7119 SV* matches_string = newSVpvn("", 0);
7121 /* Use the swash, if any, which has to have incorporated into it all
7123 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7124 && (si && si != &PL_sv_undef))
7127 /* If no swash, use the input initialization string, if available */
7128 sv_catsv(matches_string, si);
7131 /* Add the inversion list to whatever we have. This may have come from
7132 * the swash, or from an input parameter */
7134 sv_catsv(matches_string, _invlist_contents(invlist));
7136 *listsvp = matches_string;
7146 - reginclass - determine if a character falls into a character class
7148 n is the ANYOF regnode
7149 p is the target string
7150 lenp is pointer to the maximum number of bytes of how far to go in p
7151 (This is assumed wthout checking to always be at least the current
7153 utf8_target tells whether p is in UTF-8.
7155 Returns true if matched; false otherwise. If lenp is not NULL, on return
7156 from a successful match, the value it points to will be updated to how many
7157 bytes in p were matched. If there was no match, the value is undefined,
7158 possibly changed from the input.
7160 Note that this can be a synthetic start class, a combination of various
7161 nodes, so things you think might be mutually exclusive, such as locale,
7162 aren't. It can match both locale and non-locale
7167 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
7170 const char flags = ANYOF_FLAGS(n);
7176 PERL_ARGS_ASSERT_REGINCLASS;
7178 /* If c is not already the code point, get it */
7179 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
7180 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
7181 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7182 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7183 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7184 * UTF8_ALLOW_FFFF */
7185 if (c_len == (STRLEN)-1)
7186 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7192 /* Use passed in max length, or one character if none passed in or less
7193 * than one character. And assume will match just one character. This is
7194 * overwritten later if matched more. */
7196 maxlen = (*lenp > c_len) ? *lenp : c_len;
7204 /* If this character is potentially in the bitmap, check it */
7206 if (ANYOF_BITMAP_TEST(n, c))
7208 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
7215 else if (flags & ANYOF_LOCALE) {
7216 PL_reg_flags |= RF_tainted;
7218 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
7219 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7223 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
7224 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
7225 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
7226 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
7227 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
7228 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
7229 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
7230 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
7231 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
7232 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
7233 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
7234 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII_LC(c)) ||
7235 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII_LC(c)) ||
7236 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
7237 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
7238 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
7239 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
7240 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
7241 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
7242 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
7243 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
7244 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
7245 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
7246 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
7247 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
7248 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
7249 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
7250 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
7251 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
7252 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK_LC(c)) ||
7253 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK_LC(c))
7254 ) /* How's that for a conditional? */
7261 /* If the bitmap didn't (or couldn't) match, and something outside the
7262 * bitmap could match, try that. Locale nodes specifiy completely the
7263 * behavior of code points in the bit map (otherwise, a utf8 target would
7264 * cause them to be treated as Unicode and not locale), except in
7265 * the very unlikely event when this node is a synthetic start class, which
7266 * could be a combination of locale and non-locale nodes. So allow locale
7267 * to match for the synthetic start class, which will give a false
7268 * positive that will be resolved when the match is done again as not part
7269 * of the synthetic start class */
7271 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
7272 match = TRUE; /* Everything above 255 matches */
7274 else if (ANYOF_NONBITMAP(n)
7275 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
7278 || (! (flags & ANYOF_LOCALE))
7279 || (flags & ANYOF_IS_SYNTHETIC)))))
7282 SV * const sw = core_regclass_swash(prog, n, TRUE, 0, (SV**)&av);
7290 /* Not utf8. Convert as much of the string as available up
7291 * to the limit of how far the (single) character in the
7292 * pattern can possibly match (no need to go further). If
7293 * the node is a straight ANYOF or not folding, it can't
7294 * match more than one. Otherwise, It can match up to how
7295 * far a single char can fold to. Since not utf8, each
7296 * character is a single byte, so the max it can be in
7297 * bytes is the same as the max it can be in characters */
7298 STRLEN len = (OP(n) == ANYOF
7299 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
7301 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
7303 : UTF8_MAX_FOLD_CHAR_EXPAND;
7304 utf8_p = bytes_to_utf8(p, &len);
7307 if (swash_fetch(sw, utf8_p, TRUE))
7309 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
7311 /* Here, we need to test if the fold of the target string
7312 * matches. The non-multi char folds have all been moved to
7313 * the compilation phase, and the multi-char folds have
7314 * been stored by regcomp into 'av'; we linearly check to
7315 * see if any match the target string (folded). We know
7316 * that the originals were each one character, but we don't
7317 * currently know how many characters/bytes each folded to,
7318 * except we do know that there are small limits imposed by
7319 * Unicode. XXX A performance enhancement would be to have
7320 * regcomp.c store the max number of chars/bytes that are
7321 * in an av entry, as, say the 0th element. Even better
7322 * would be to have a hash of the few characters that can
7323 * start a multi-char fold to the max number of chars of
7326 * If there is a match, we will need to advance (if lenp is
7327 * specified) the match pointer in the target string. But
7328 * what we are comparing here isn't that string directly,
7329 * but its fold, whose length may differ from the original.
7330 * As we go along in constructing the fold, therefore, we
7331 * create a map so that we know how many bytes in the
7332 * source to advance given that we have matched a certain
7333 * number of bytes in the fold. This map is stored in
7334 * 'map_fold_len_back'. Let n mean the number of bytes in
7335 * the fold of the first character that we are folding.
7336 * Then map_fold_len_back[n] is set to the number of bytes
7337 * in that first character. Similarly let m be the
7338 * corresponding number for the second character to be
7339 * folded. Then map_fold_len_back[n+m] is set to the
7340 * number of bytes occupied by the first two source
7341 * characters. ... */
7342 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
7343 U8 folded[UTF8_MAXBYTES_CASE+1];
7344 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
7345 STRLEN total_foldlen = 0; /* num bytes in fold of all
7348 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
7350 /* Here, only need to fold the first char of the target
7351 * string. It the source wasn't utf8, is 1 byte long */
7352 to_utf8_fold(utf8_p, folded, &foldlen);
7353 total_foldlen = foldlen;
7354 map_fold_len_back[foldlen] = (utf8_target)
7360 /* Here, need to fold more than the first char. Do so
7361 * up to the limits */
7362 U8* source_ptr = utf8_p; /* The source for the fold
7365 U8* folded_ptr = folded;
7366 U8* e = utf8_p + maxlen; /* Can't go beyond last
7367 available byte in the
7371 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
7375 /* Fold the next character */
7376 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
7377 STRLEN this_char_foldlen;
7378 to_utf8_fold(source_ptr,
7380 &this_char_foldlen);
7382 /* Bail if it would exceed the byte limit for
7383 * folding a single char. */
7384 if (this_char_foldlen + folded_ptr - folded >
7390 /* Add the fold of this character */
7391 Copy(this_char_folded,
7395 source_ptr += UTF8SKIP(source_ptr);
7396 folded_ptr += this_char_foldlen;
7397 total_foldlen = folded_ptr - folded;
7399 /* Create map from the number of bytes in the fold
7400 * back to the number of bytes in the source. If
7401 * the source isn't utf8, the byte count is just
7402 * the number of characters so far */
7403 map_fold_len_back[total_foldlen]
7405 ? source_ptr - utf8_p
7412 /* Do the linear search to see if the fold is in the list
7413 * of multi-char folds. */
7416 for (i = 0; i <= av_len(av); i++) {
7417 SV* const sv = *av_fetch(av, i, FALSE);
7419 const char * const s = SvPV_const(sv, len);
7421 if (len <= total_foldlen
7422 && memEQ(s, (char*)folded, len)
7424 /* If 0, means matched a partial char. See
7426 && map_fold_len_back[len])
7429 /* Advance the target string ptr to account for
7430 * this fold, but have to translate from the
7431 * folded length to the corresponding source
7434 *lenp = map_fold_len_back[len];
7443 /* If we allocated a string above, free it */
7444 if (! utf8_target) Safefree(utf8_p);
7448 if (UNICODE_IS_SUPER(c)
7449 && (flags & ANYOF_WARN_SUPER)
7450 && ckWARN_d(WARN_NON_UNICODE))
7452 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7453 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7457 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7458 return cBOOL(flags & ANYOF_INVERT) ^ match;
7462 S_reghop3(U8 *s, I32 off, const U8* lim)
7464 /* return the position 'off' UTF-8 characters away from 's', forward if
7465 * 'off' >= 0, backwards if negative. But don't go outside of position
7466 * 'lim', which better be < s if off < 0 */
7470 PERL_ARGS_ASSERT_REGHOP3;
7473 while (off-- && s < lim) {
7474 /* XXX could check well-formedness here */
7479 while (off++ && s > lim) {
7481 if (UTF8_IS_CONTINUED(*s)) {
7482 while (s > lim && UTF8_IS_CONTINUATION(*s))
7485 /* XXX could check well-formedness here */
7492 /* there are a bunch of places where we use two reghop3's that should
7493 be replaced with this routine. but since thats not done yet
7494 we ifdef it out - dmq
7497 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7501 PERL_ARGS_ASSERT_REGHOP4;
7504 while (off-- && s < rlim) {
7505 /* XXX could check well-formedness here */
7510 while (off++ && s > llim) {
7512 if (UTF8_IS_CONTINUED(*s)) {
7513 while (s > llim && UTF8_IS_CONTINUATION(*s))
7516 /* XXX could check well-formedness here */
7524 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7528 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7531 while (off-- && s < lim) {
7532 /* XXX could check well-formedness here */
7539 while (off++ && s > lim) {
7541 if (UTF8_IS_CONTINUED(*s)) {
7542 while (s > lim && UTF8_IS_CONTINUATION(*s))
7545 /* XXX could check well-formedness here */
7554 restore_pos(pTHX_ void *arg)
7557 regexp * const rex = (regexp *)arg;
7558 if (PL_reg_state.re_state_eval_setup_done) {
7559 if (PL_reg_oldsaved) {
7560 rex->subbeg = PL_reg_oldsaved;
7561 rex->sublen = PL_reg_oldsavedlen;
7562 rex->suboffset = PL_reg_oldsavedoffset;
7563 rex->subcoffset = PL_reg_oldsavedcoffset;
7564 #ifdef PERL_OLD_COPY_ON_WRITE
7565 rex->saved_copy = PL_nrs;
7567 RXp_MATCH_COPIED_on(rex);
7569 PL_reg_magic->mg_len = PL_reg_oldpos;
7570 PL_reg_state.re_state_eval_setup_done = FALSE;
7571 PL_curpm = PL_reg_oldcurpm;
7576 S_to_utf8_substr(pTHX_ register regexp *prog)
7578 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
7579 * on the converted value */
7583 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7586 if (prog->substrs->data[i].substr
7587 && !prog->substrs->data[i].utf8_substr) {
7588 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7589 prog->substrs->data[i].utf8_substr = sv;
7590 sv_utf8_upgrade(sv);
7591 if (SvVALID(prog->substrs->data[i].substr)) {
7592 if (SvTAIL(prog->substrs->data[i].substr)) {
7593 /* Trim the trailing \n that fbm_compile added last
7595 SvCUR_set(sv, SvCUR(sv) - 1);
7596 /* Whilst this makes the SV technically "invalid" (as its
7597 buffer is no longer followed by "\0") when fbm_compile()
7598 adds the "\n" back, a "\0" is restored. */
7599 fbm_compile(sv, FBMcf_TAIL);
7603 if (prog->substrs->data[i].substr == prog->check_substr)
7604 prog->check_utf8 = sv;
7610 S_to_byte_substr(pTHX_ register regexp *prog)
7612 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
7613 * on the converted value; returns FALSE if can't be converted. */
7618 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7621 if (prog->substrs->data[i].utf8_substr
7622 && !prog->substrs->data[i].substr) {
7623 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7624 if (! sv_utf8_downgrade(sv, TRUE)) {
7627 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7628 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7629 /* Trim the trailing \n that fbm_compile added last
7631 SvCUR_set(sv, SvCUR(sv) - 1);
7632 fbm_compile(sv, FBMcf_TAIL);
7636 prog->substrs->data[i].substr = sv;
7637 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7638 prog->check_substr = sv;
7645 /* These constants are for finding GCB=LV and GCB=LVT. These are for the
7646 * pre-composed Hangul syllables, which are all in a contiguous block and
7647 * arranged there in such a way so as to facilitate alorithmic determination of
7648 * their characteristics. As such, they don't need a swash, but can be
7649 * determined by simple arithmetic. Almost all are GCB=LVT, but every 28th one
7651 #define SBASE 0xAC00 /* Start of block */
7652 #define SCount 11172 /* Length of block */
7655 #if 0 /* This routine is not currently used */
7656 PERL_STATIC_INLINE bool
7657 S_is_utf8_X_LV(pTHX_ const U8 *p)
7659 /* Unlike most other similarly named routines here, this does not create a
7660 * swash, so swash_fetch() cannot be used on PL_utf8_X_LV. */
7664 UV cp = valid_utf8_to_uvchr(p, NULL);
7666 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
7668 /* The earliest Unicode releases did not have these precomposed Hangul
7669 * syllables. Set to point to undef in that case, so will return false on
7671 if (! PL_utf8_X_LV) { /* Set up if this is the first time called */
7672 PL_utf8_X_LV = swash_init("utf8", "_X_GCB_LV", &PL_sv_undef, 1, 0);
7673 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LV)) == 0) {
7674 SvREFCNT_dec(PL_utf8_X_LV);
7675 PL_utf8_X_LV = &PL_sv_undef;
7679 return (PL_utf8_X_LV != &PL_sv_undef
7680 && cp >= SBASE && cp < SBASE + SCount
7681 && (cp - SBASE) % TCount == 0); /* Only every TCount one is LV */
7685 PERL_STATIC_INLINE bool
7686 S_is_utf8_X_LVT(pTHX_ const U8 *p)
7688 /* Unlike most other similarly named routines here, this does not create a
7689 * swash, so swash_fetch() cannot be used on PL_utf8_X_LVT. */
7693 UV cp = valid_utf8_to_uvchr(p, NULL);
7695 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
7697 /* The earliest Unicode releases did not have these precomposed Hangul
7698 * syllables. Set to point to undef in that case, so will return false on
7700 if (! PL_utf8_X_LVT) { /* Set up if this is the first time called */
7701 PL_utf8_X_LVT = swash_init("utf8", "_X_GCB_LVT", &PL_sv_undef, 1, 0);
7702 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LVT)) == 0) {
7703 SvREFCNT_dec(PL_utf8_X_LVT);
7704 PL_utf8_X_LVT = &PL_sv_undef;
7708 return (PL_utf8_X_LVT != &PL_sv_undef
7709 && cp >= SBASE && cp < SBASE + SCount
7710 && (cp - SBASE) % TCount != 0); /* All but every TCount one is LV */
7715 * c-indentation-style: bsd
7717 * indent-tabs-mode: nil
7720 * ex: set ts=8 sts=4 sw=4 et: