5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "inline_invlist.c"
84 #include "unicode_constants.h"
86 #define RF_tainted 1 /* tainted information used? e.g. locale */
87 #define RF_warned 2 /* warned about big count? */
89 #define RF_utf8 8 /* Pattern contains multibyte chars? */
91 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
97 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
98 * call if there are no complications: i.e., if everything matchable is
99 * straight forward in the bitmap */
100 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
101 : ANYOF_BITMAP_TEST(p,*(c)))
107 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
108 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
110 #define HOPc(pos,off) \
111 (char *)(PL_reg_match_utf8 \
112 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
114 #define HOPBACKc(pos, off) \
115 (char*)(PL_reg_match_utf8\
116 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
117 : (pos - off >= PL_bostr) \
121 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
122 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
124 /* these are unrolled below in the CCC_TRY_XXX defined */
125 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
126 if (!CAT2(PL_utf8_,class)) { \
128 ENTER; save_re_context(); \
129 ok=CAT2(is_utf8_,class)((const U8*)str); \
130 PERL_UNUSED_VAR(ok); \
131 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
132 /* Doesn't do an assert to verify that is correct */
133 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
134 if (!CAT2(PL_utf8_,class)) { \
135 bool throw_away PERL_UNUSED_DECL; \
136 ENTER; save_re_context(); \
137 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
140 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
141 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
142 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
144 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
145 /* No asserts are done for some of these, in case called on a */ \
146 /* Unicode version in which they map to nothing */ \
147 LOAD_UTF8_CHARCLASS(X_regular_begin, HYPHEN_UTF8); \
148 LOAD_UTF8_CHARCLASS(X_extend, COMBINING_GRAVE_ACCENT_UTF8); \
150 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
152 /* The actual code for CCC_TRY, which uses several variables from the routine
153 * it's callable from. It is designed to be the bulk of a case statement.
154 * FUNC is the macro or function to call on non-utf8 targets that indicate if
155 * nextchr matches the class.
156 * UTF8_TEST is the whole test string to use for utf8 targets
157 * LOAD is what to use to test, and if not present to load in the swash for the
159 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
161 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
162 * utf8 and a variant, load the swash if necessary and test using the utf8
163 * test. Advance to the next character if test is ok, otherwise fail; If not
164 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
165 * fails, or advance to the next character */
167 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
168 if (locinput >= PL_regeol) { \
171 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
172 LOAD_UTF8_CHARCLASS(CLASS, STR); \
173 if (POS_OR_NEG (UTF8_TEST)) { \
176 locinput += PL_utf8skip[nextchr]; \
177 nextchr = UCHARAT(locinput); \
180 if (POS_OR_NEG (FUNC(nextchr))) { \
183 nextchr = UCHARAT(++locinput); \
186 /* Handle the non-locale cases for a character class and its complement. It
187 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
188 * This is because that code fails when the test succeeds, so we want to have
189 * the test fail so that the code succeeds. The swash is stored in a
190 * predictable PL_ place */
191 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
194 _CCC_TRY_CODE( !, FUNC, \
195 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
196 (U8*)locinput, TRUE)), \
199 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
200 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
201 (U8*)locinput, TRUE)), \
204 /* Generate the case statements for both locale and non-locale character
205 * classes in regmatch for classes that don't have special unicode semantics.
206 * Locales don't use an immediate swash, but an intermediary special locale
207 * function that is called on the pointer to the current place in the input
208 * string. That function will resolve to needing the same swash. One might
209 * think that because we don't know what the locale will match, we shouldn't
210 * check with the swash loading function that it loaded properly; ie, that we
211 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
212 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
214 #define CCC_TRY(NAME, NNAME, FUNC, \
215 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
216 NAMEA, NNAMEA, FUNCA, \
219 PL_reg_flags |= RF_tainted; \
220 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
222 PL_reg_flags |= RF_tainted; \
223 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
226 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
229 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
230 nextchr = UCHARAT(++locinput); \
233 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
237 locinput += PL_utf8skip[nextchr]; \
238 nextchr = UCHARAT(locinput); \
241 nextchr = UCHARAT(++locinput); \
244 /* Generate the non-locale cases */ \
245 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
247 /* This is like CCC_TRY, but has an extra set of parameters for generating case
248 * statements to handle separate Unicode semantics nodes */
249 #define CCC_TRY_U(NAME, NNAME, FUNC, \
250 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
251 NAMEU, NNAMEU, FUNCU, \
252 NAMEA, NNAMEA, FUNCA, \
254 CCC_TRY(NAME, NNAME, FUNC, \
255 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
256 NAMEA, NNAMEA, FUNCA, \
258 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
260 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
262 /* for use after a quantifier and before an EXACT-like node -- japhy */
263 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
265 * NOTE that *nothing* that affects backtracking should be in here, specifically
266 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
267 * node that is in between two EXACT like nodes when ascertaining what the required
268 * "follow" character is. This should probably be moved to regex compile time
269 * although it may be done at run time beause of the REF possibility - more
270 * investigation required. -- demerphq
272 #define JUMPABLE(rn) ( \
274 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
276 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
277 OP(rn) == PLUS || OP(rn) == MINMOD || \
279 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
281 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
283 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
286 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
287 we don't need this definition. */
288 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
289 #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 )
290 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
293 /* ... so we use this as its faster. */
294 #define IS_TEXT(rn) ( OP(rn)==EXACT )
295 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
296 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
297 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
302 Search for mandatory following text node; for lookahead, the text must
303 follow but for lookbehind (rn->flags != 0) we skip to the next step.
305 #define FIND_NEXT_IMPT(rn) STMT_START { \
306 while (JUMPABLE(rn)) { \
307 const OPCODE type = OP(rn); \
308 if (type == SUSPEND || PL_regkind[type] == CURLY) \
309 rn = NEXTOPER(NEXTOPER(rn)); \
310 else if (type == PLUS) \
312 else if (type == IFMATCH) \
313 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
314 else rn += NEXT_OFF(rn); \
319 static void restore_pos(pTHX_ void *arg);
321 #define REGCP_PAREN_ELEMS 3
322 #define REGCP_OTHER_ELEMS 3
323 #define REGCP_FRAME_ELEMS 1
324 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
325 * are needed for the regexp context stack bookkeeping. */
328 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor)
331 const int retval = PL_savestack_ix;
332 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
333 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
334 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
336 GET_RE_DEBUG_FLAGS_DECL;
338 PERL_ARGS_ASSERT_REGCPPUSH;
340 if (paren_elems_to_push < 0)
341 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
342 paren_elems_to_push);
344 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
345 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
346 " out of range (%lu-%ld)",
347 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
349 SSGROW(total_elems + REGCP_FRAME_ELEMS);
352 if ((int)PL_regsize > (int)parenfloor)
353 PerlIO_printf(Perl_debug_log,
354 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
359 for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
360 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
361 SSPUSHINT(rex->offs[p].end);
362 SSPUSHINT(rex->offs[p].start);
363 SSPUSHINT(rex->offs[p].start_tmp);
364 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
365 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
367 (IV)rex->offs[p].start,
368 (IV)rex->offs[p].start_tmp,
372 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
373 SSPUSHINT(PL_regsize);
374 SSPUSHINT(rex->lastparen);
375 SSPUSHINT(rex->lastcloseparen);
376 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
381 /* These are needed since we do not localize EVAL nodes: */
382 #define REGCP_SET(cp) \
384 PerlIO_printf(Perl_debug_log, \
385 " Setting an EVAL scope, savestack=%"IVdf"\n", \
386 (IV)PL_savestack_ix)); \
389 #define REGCP_UNWIND(cp) \
391 if (cp != PL_savestack_ix) \
392 PerlIO_printf(Perl_debug_log, \
393 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
394 (IV)(cp), (IV)PL_savestack_ix)); \
397 #define UNWIND_PAREN(lp, lcp) \
398 for (n = rex->lastparen; n > lp; n--) \
399 rex->offs[n].end = -1; \
400 rex->lastparen = n; \
401 rex->lastcloseparen = lcp;
405 S_regcppop(pTHX_ regexp *rex)
410 GET_RE_DEBUG_FLAGS_DECL;
412 PERL_ARGS_ASSERT_REGCPPOP;
414 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
416 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
417 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
418 rex->lastcloseparen = SSPOPINT;
419 rex->lastparen = SSPOPINT;
420 PL_regsize = SSPOPINT;
422 i -= REGCP_OTHER_ELEMS;
423 /* Now restore the parentheses context. */
425 if (i || rex->lastparen + 1 <= rex->nparens)
426 PerlIO_printf(Perl_debug_log,
427 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
433 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
435 rex->offs[paren].start_tmp = SSPOPINT;
436 rex->offs[paren].start = SSPOPINT;
438 if (paren <= rex->lastparen)
439 rex->offs[paren].end = tmps;
440 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
441 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
443 (IV)rex->offs[paren].start,
444 (IV)rex->offs[paren].start_tmp,
445 (IV)rex->offs[paren].end,
446 (paren > rex->lastparen ? "(skipped)" : ""));
451 /* It would seem that the similar code in regtry()
452 * already takes care of this, and in fact it is in
453 * a better location to since this code can #if 0-ed out
454 * but the code in regtry() is needed or otherwise tests
455 * requiring null fields (pat.t#187 and split.t#{13,14}
456 * (as of patchlevel 7877) will fail. Then again,
457 * this code seems to be necessary or otherwise
458 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
459 * --jhi updated by dapm */
460 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
462 rex->offs[i].start = -1;
463 rex->offs[i].end = -1;
464 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
465 " \\%"UVuf": %s ..-1 undeffing\n",
467 (i > PL_regsize) ? "-1" : " "
473 /* restore the parens and associated vars at savestack position ix,
474 * but without popping the stack */
477 S_regcp_restore(pTHX_ regexp *rex, I32 ix)
479 I32 tmpix = PL_savestack_ix;
480 PL_savestack_ix = ix;
482 PL_savestack_ix = tmpix;
485 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
488 * pregexec and friends
491 #ifndef PERL_IN_XSUB_RE
493 - pregexec - match a regexp against a string
496 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
497 char *strbeg, I32 minend, SV *screamer, U32 nosave)
498 /* stringarg: the point in the string at which to begin matching */
499 /* strend: pointer to null at end of string */
500 /* strbeg: real beginning of string */
501 /* minend: end of match must be >= minend bytes after stringarg. */
502 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
503 * itself is accessed via the pointers above */
504 /* nosave: For optimizations. */
506 PERL_ARGS_ASSERT_PREGEXEC;
509 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
510 nosave ? 0 : REXEC_COPY_STR);
515 * Need to implement the following flags for reg_anch:
517 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
519 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
520 * INTUIT_AUTORITATIVE_ML
521 * INTUIT_ONCE_NOML - Intuit can match in one location only.
524 * Another flag for this function: SECOND_TIME (so that float substrs
525 * with giant delta may be not rechecked).
528 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
530 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
531 Otherwise, only SvCUR(sv) is used to get strbeg. */
533 /* XXXX We assume that strpos is strbeg unless sv. */
535 /* XXXX Some places assume that there is a fixed substring.
536 An update may be needed if optimizer marks as "INTUITable"
537 RExen without fixed substrings. Similarly, it is assumed that
538 lengths of all the strings are no more than minlen, thus they
539 cannot come from lookahead.
540 (Or minlen should take into account lookahead.)
541 NOTE: Some of this comment is not correct. minlen does now take account
542 of lookahead/behind. Further research is required. -- demerphq
546 /* A failure to find a constant substring means that there is no need to make
547 an expensive call to REx engine, thus we celebrate a failure. Similarly,
548 finding a substring too deep into the string means that less calls to
549 regtry() should be needed.
551 REx compiler's optimizer found 4 possible hints:
552 a) Anchored substring;
554 c) Whether we are anchored (beginning-of-line or \G);
555 d) First node (of those at offset 0) which may distinguish positions;
556 We use a)b)d) and multiline-part of c), and try to find a position in the
557 string which does not contradict any of them.
560 /* Most of decisions we do here should have been done at compile time.
561 The nodes of the REx which we used for the search should have been
562 deleted from the finite automaton. */
565 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
566 char *strend, const U32 flags, re_scream_pos_data *data)
569 struct regexp *const prog = (struct regexp *)SvANY(rx);
571 /* Should be nonnegative! */
577 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
579 char *other_last = NULL; /* other substr checked before this */
580 char *check_at = NULL; /* check substr found at this pos */
581 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
582 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
583 RXi_GET_DECL(prog,progi);
585 const char * const i_strpos = strpos;
587 GET_RE_DEBUG_FLAGS_DECL;
589 PERL_ARGS_ASSERT_RE_INTUIT_START;
590 PERL_UNUSED_ARG(flags);
591 PERL_UNUSED_ARG(data);
593 RX_MATCH_UTF8_set(rx,utf8_target);
596 PL_reg_flags |= RF_utf8;
599 debug_start_match(rx, utf8_target, strpos, strend,
600 sv ? "Guessing start of match in sv for"
601 : "Guessing start of match in string for");
604 /* CHR_DIST() would be more correct here but it makes things slow. */
605 if (prog->minlen > strend - strpos) {
606 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
607 "String too short... [re_intuit_start]\n"));
611 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
614 if (!prog->check_utf8 && prog->check_substr)
615 to_utf8_substr(prog);
616 check = prog->check_utf8;
618 if (!prog->check_substr && prog->check_utf8)
619 to_byte_substr(prog);
620 check = prog->check_substr;
622 if (check == &PL_sv_undef) {
623 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
624 "Non-utf8 string cannot match utf8 check string\n"));
627 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
628 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
629 || ( (prog->extflags & RXf_ANCH_BOL)
630 && !multiline ) ); /* Check after \n? */
633 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
634 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
635 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
637 && (strpos != strbeg)) {
638 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
641 if (prog->check_offset_min == prog->check_offset_max &&
642 !(prog->extflags & RXf_CANY_SEEN)) {
643 /* Substring at constant offset from beg-of-str... */
646 s = HOP3c(strpos, prog->check_offset_min, strend);
649 slen = SvCUR(check); /* >= 1 */
651 if ( strend - s > slen || strend - s < slen - 1
652 || (strend - s == slen && strend[-1] != '\n')) {
653 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
656 /* Now should match s[0..slen-2] */
658 if (slen && (*SvPVX_const(check) != *s
660 && memNE(SvPVX_const(check), s, slen)))) {
662 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
666 else if (*SvPVX_const(check) != *s
667 || ((slen = SvCUR(check)) > 1
668 && memNE(SvPVX_const(check), s, slen)))
671 goto success_at_start;
674 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
676 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
677 end_shift = prog->check_end_shift;
680 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
681 - (SvTAIL(check) != 0);
682 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
684 if (end_shift < eshift)
688 else { /* Can match at random position */
691 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
692 end_shift = prog->check_end_shift;
694 /* end shift should be non negative here */
697 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
699 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
700 (IV)end_shift, RX_PRECOMP(prog));
704 /* Find a possible match in the region s..strend by looking for
705 the "check" substring in the region corrected by start/end_shift. */
708 I32 srch_start_shift = start_shift;
709 I32 srch_end_shift = end_shift;
712 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
713 srch_end_shift -= ((strbeg - s) - srch_start_shift);
714 srch_start_shift = strbeg - s;
716 DEBUG_OPTIMISE_MORE_r({
717 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
718 (IV)prog->check_offset_min,
719 (IV)srch_start_shift,
721 (IV)prog->check_end_shift);
724 if (prog->extflags & RXf_CANY_SEEN) {
725 start_point= (U8*)(s + srch_start_shift);
726 end_point= (U8*)(strend - srch_end_shift);
728 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
729 end_point= HOP3(strend, -srch_end_shift, strbeg);
731 DEBUG_OPTIMISE_MORE_r({
732 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
733 (int)(end_point - start_point),
734 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
738 s = fbm_instr( start_point, end_point,
739 check, multiline ? FBMrf_MULTILINE : 0);
741 /* Update the count-of-usability, remove useless subpatterns,
745 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
746 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
747 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
748 (s ? "Found" : "Did not find"),
749 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
750 ? "anchored" : "floating"),
753 (s ? " at offset " : "...\n") );
758 /* Finish the diagnostic message */
759 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
761 /* XXX dmq: first branch is for positive lookbehind...
762 Our check string is offset from the beginning of the pattern.
763 So we need to do any stclass tests offset forward from that
772 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
773 Start with the other substr.
774 XXXX no SCREAM optimization yet - and a very coarse implementation
775 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
776 *always* match. Probably should be marked during compile...
777 Probably it is right to do no SCREAM here...
780 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
781 : (prog->float_substr && prog->anchored_substr))
783 /* Take into account the "other" substring. */
784 /* XXXX May be hopelessly wrong for UTF... */
787 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
790 char * const last = HOP3c(s, -start_shift, strbeg);
792 char * const saved_s = s;
795 t = s - prog->check_offset_max;
796 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
798 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
803 t = HOP3c(t, prog->anchored_offset, strend);
804 if (t < other_last) /* These positions already checked */
806 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
809 /* XXXX It is not documented what units *_offsets are in.
810 We assume bytes, but this is clearly wrong.
811 Meaning this code needs to be carefully reviewed for errors.
815 /* On end-of-str: see comment below. */
816 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
817 if (must == &PL_sv_undef) {
819 DEBUG_r(must = prog->anchored_utf8); /* for debug */
824 HOP3(HOP3(last1, prog->anchored_offset, strend)
825 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
827 multiline ? FBMrf_MULTILINE : 0
830 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
831 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
832 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
833 (s ? "Found" : "Contradicts"),
834 quoted, RE_SV_TAIL(must));
839 if (last1 >= last2) {
840 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
841 ", giving up...\n"));
844 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
845 ", trying floating at offset %ld...\n",
846 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
847 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
848 s = HOP3c(last, 1, strend);
852 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
853 (long)(s - i_strpos)));
854 t = HOP3c(s, -prog->anchored_offset, strbeg);
855 other_last = HOP3c(s, 1, strend);
863 else { /* Take into account the floating substring. */
865 char * const saved_s = s;
868 t = HOP3c(s, -start_shift, strbeg);
870 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
871 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
872 last = HOP3c(t, prog->float_max_offset, strend);
873 s = HOP3c(t, prog->float_min_offset, strend);
876 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
877 must = utf8_target ? prog->float_utf8 : prog->float_substr;
878 /* fbm_instr() takes into account exact value of end-of-str
879 if the check is SvTAIL(ed). Since false positives are OK,
880 and end-of-str is not later than strend we are OK. */
881 if (must == &PL_sv_undef) {
883 DEBUG_r(must = prog->float_utf8); /* for debug message */
886 s = fbm_instr((unsigned char*)s,
887 (unsigned char*)last + SvCUR(must)
889 must, multiline ? FBMrf_MULTILINE : 0);
891 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
892 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
893 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
894 (s ? "Found" : "Contradicts"),
895 quoted, RE_SV_TAIL(must));
899 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
900 ", giving up...\n"));
903 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
904 ", trying anchored starting at offset %ld...\n",
905 (long)(saved_s + 1 - i_strpos)));
907 s = HOP3c(t, 1, strend);
911 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
912 (long)(s - i_strpos)));
913 other_last = s; /* Fix this later. --Hugo */
923 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
925 DEBUG_OPTIMISE_MORE_r(
926 PerlIO_printf(Perl_debug_log,
927 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
928 (IV)prog->check_offset_min,
929 (IV)prog->check_offset_max,
937 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
939 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
942 /* Fixed substring is found far enough so that the match
943 cannot start at strpos. */
945 if (ml_anch && t[-1] != '\n') {
946 /* Eventually fbm_*() should handle this, but often
947 anchored_offset is not 0, so this check will not be wasted. */
948 /* XXXX In the code below we prefer to look for "^" even in
949 presence of anchored substrings. And we search even
950 beyond the found float position. These pessimizations
951 are historical artefacts only. */
953 while (t < strend - prog->minlen) {
955 if (t < check_at - prog->check_offset_min) {
956 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
957 /* Since we moved from the found position,
958 we definitely contradict the found anchored
959 substr. Due to the above check we do not
960 contradict "check" substr.
961 Thus we can arrive here only if check substr
962 is float. Redo checking for "other"=="fixed".
965 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
966 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
967 goto do_other_anchored;
969 /* We don't contradict the found floating substring. */
970 /* XXXX Why not check for STCLASS? */
972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
973 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
976 /* Position contradicts check-string */
977 /* XXXX probably better to look for check-string
978 than for "\n", so one should lower the limit for t? */
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
980 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
981 other_last = strpos = s = t + 1;
986 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
987 PL_colors[0], PL_colors[1]));
991 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
992 PL_colors[0], PL_colors[1]));
996 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
999 /* The found string does not prohibit matching at strpos,
1000 - no optimization of calling REx engine can be performed,
1001 unless it was an MBOL and we are not after MBOL,
1002 or a future STCLASS check will fail this. */
1004 /* Even in this situation we may use MBOL flag if strpos is offset
1005 wrt the start of the string. */
1006 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1007 && (strpos != strbeg) && strpos[-1] != '\n'
1008 /* May be due to an implicit anchor of m{.*foo} */
1009 && !(prog->intflags & PREGf_IMPLICIT))
1014 DEBUG_EXECUTE_r( if (ml_anch)
1015 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1016 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1019 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1021 prog->check_utf8 /* Could be deleted already */
1022 && --BmUSEFUL(prog->check_utf8) < 0
1023 && (prog->check_utf8 == prog->float_utf8)
1025 prog->check_substr /* Could be deleted already */
1026 && --BmUSEFUL(prog->check_substr) < 0
1027 && (prog->check_substr == prog->float_substr)
1030 /* If flags & SOMETHING - do not do it many times on the same match */
1031 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1032 /* XXX Does the destruction order has to change with utf8_target? */
1033 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1034 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1035 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1036 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1037 check = NULL; /* abort */
1039 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1040 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1041 if (prog->intflags & PREGf_IMPLICIT)
1042 prog->extflags &= ~RXf_ANCH_MBOL;
1043 /* XXXX This is a remnant of the old implementation. It
1044 looks wasteful, since now INTUIT can use many
1045 other heuristics. */
1046 prog->extflags &= ~RXf_USE_INTUIT;
1047 /* XXXX What other flags might need to be cleared in this branch? */
1053 /* Last resort... */
1054 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1055 /* trie stclasses are too expensive to use here, we are better off to
1056 leave it to regmatch itself */
1057 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1058 /* minlen == 0 is possible if regstclass is \b or \B,
1059 and the fixed substr is ''$.
1060 Since minlen is already taken into account, s+1 is before strend;
1061 accidentally, minlen >= 1 guaranties no false positives at s + 1
1062 even for \b or \B. But (minlen? 1 : 0) below assumes that
1063 regstclass does not come from lookahead... */
1064 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1065 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1066 const U8* const str = (U8*)STRING(progi->regstclass);
1067 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1068 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1071 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1072 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1073 else if (prog->float_substr || prog->float_utf8)
1074 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1078 if (checked_upto < s)
1080 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1081 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1084 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1089 const char *what = NULL;
1091 if (endpos == strend) {
1092 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1093 "Could not match STCLASS...\n") );
1096 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1097 "This position contradicts STCLASS...\n") );
1098 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1100 checked_upto = HOPBACKc(endpos, start_shift);
1101 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1102 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1103 /* Contradict one of substrings */
1104 if (prog->anchored_substr || prog->anchored_utf8) {
1105 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1106 DEBUG_EXECUTE_r( what = "anchored" );
1108 s = HOP3c(t, 1, strend);
1109 if (s + start_shift + end_shift > strend) {
1110 /* XXXX Should be taken into account earlier? */
1111 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1112 "Could not match STCLASS...\n") );
1117 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1118 "Looking for %s substr starting at offset %ld...\n",
1119 what, (long)(s + start_shift - i_strpos)) );
1122 /* Have both, check_string is floating */
1123 if (t + start_shift >= check_at) /* Contradicts floating=check */
1124 goto retry_floating_check;
1125 /* Recheck anchored substring, but not floating... */
1129 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1130 "Looking for anchored substr starting at offset %ld...\n",
1131 (long)(other_last - i_strpos)) );
1132 goto do_other_anchored;
1134 /* Another way we could have checked stclass at the
1135 current position only: */
1140 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1141 "Looking for /%s^%s/m starting at offset %ld...\n",
1142 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1145 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1147 /* Check is floating substring. */
1148 retry_floating_check:
1149 t = check_at - start_shift;
1150 DEBUG_EXECUTE_r( what = "floating" );
1151 goto hop_and_restart;
1154 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1155 "By STCLASS: moving %ld --> %ld\n",
1156 (long)(t - i_strpos), (long)(s - i_strpos))
1160 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1161 "Does not contradict STCLASS...\n");
1166 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1167 PL_colors[4], (check ? "Guessed" : "Giving up"),
1168 PL_colors[5], (long)(s - i_strpos)) );
1171 fail_finish: /* Substring not found */
1172 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1173 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1175 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1176 PL_colors[4], PL_colors[5]));
1180 #define DECL_TRIE_TYPE(scan) \
1181 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1182 trie_type = ((scan->flags == EXACT) \
1183 ? (utf8_target ? trie_utf8 : trie_plain) \
1184 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1186 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1187 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1189 switch (trie_type) { \
1190 case trie_utf8_fold: \
1191 if ( foldlen>0 ) { \
1192 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1197 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1198 len = UTF8SKIP(uc); \
1199 skiplen = UNISKIP( uvc ); \
1200 foldlen -= skiplen; \
1201 uscan = foldbuf + skiplen; \
1204 case trie_latin_utf8_fold: \
1205 if ( foldlen>0 ) { \
1206 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1212 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1213 skiplen = UNISKIP( uvc ); \
1214 foldlen -= skiplen; \
1215 uscan = foldbuf + skiplen; \
1219 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1226 charid = trie->charmap[ uvc ]; \
1230 if (widecharmap) { \
1231 SV** const svpp = hv_fetch(widecharmap, \
1232 (char*)&uvc, sizeof(UV), 0); \
1234 charid = (U16)SvIV(*svpp); \
1239 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1243 && (ln == 1 || folder(s, pat_string, ln)) \
1244 && (!reginfo || regtry(reginfo, &s)) ) \
1250 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1252 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1258 #define REXEC_FBC_SCAN(CoDe) \
1260 while (s < strend) { \
1266 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1267 REXEC_FBC_UTF8_SCAN( \
1269 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1278 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1281 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1290 #define REXEC_FBC_TRYIT \
1291 if ((!reginfo || regtry(reginfo, &s))) \
1294 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1295 if (utf8_target) { \
1296 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1299 REXEC_FBC_CLASS_SCAN(CoNd); \
1302 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1303 if (utf8_target) { \
1305 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1308 REXEC_FBC_CLASS_SCAN(CoNd); \
1311 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1312 PL_reg_flags |= RF_tainted; \
1313 if (utf8_target) { \
1314 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1317 REXEC_FBC_CLASS_SCAN(CoNd); \
1320 #define DUMP_EXEC_POS(li,s,doutf8) \
1321 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1324 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1325 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1326 tmp = TEST_NON_UTF8(tmp); \
1327 REXEC_FBC_UTF8_SCAN( \
1328 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1337 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1338 if (s == PL_bostr) { \
1342 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1343 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1346 LOAD_UTF8_CHARCLASS_ALNUM(); \
1347 REXEC_FBC_UTF8_SCAN( \
1348 if (tmp == ! (TeSt2_UtF8)) { \
1357 /* The only difference between the BOUND and NBOUND cases is that
1358 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1359 * NBOUND. This is accomplished by passing it in either the if or else clause,
1360 * with the other one being empty */
1361 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1362 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1364 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1365 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1367 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1368 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1370 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1371 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1374 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1375 * be passed in completely with the variable name being tested, which isn't
1376 * such a clean interface, but this is easier to read than it was before. We
1377 * are looking for the boundary (or non-boundary between a word and non-word
1378 * character. The utf8 and non-utf8 cases have the same logic, but the details
1379 * must be different. Find the "wordness" of the character just prior to this
1380 * one, and compare it with the wordness of this one. If they differ, we have
1381 * a boundary. At the beginning of the string, pretend that the previous
1382 * character was a new-line */
1383 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1384 if (utf8_target) { \
1387 else { /* Not utf8 */ \
1388 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1389 tmp = TEST_NON_UTF8(tmp); \
1391 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1400 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1403 /* We know what class REx starts with. Try to find this position... */
1404 /* if reginfo is NULL, its a dryrun */
1405 /* annoyingly all the vars in this routine have different names from their counterparts
1406 in regmatch. /grrr */
1409 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1410 const char *strend, regmatch_info *reginfo)
1413 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1414 char *pat_string; /* The pattern's exactish string */
1415 char *pat_end; /* ptr to end char of pat_string */
1416 re_fold_t folder; /* Function for computing non-utf8 folds */
1417 const U8 *fold_array; /* array for folding ords < 256 */
1424 I32 tmp = 1; /* Scratch variable? */
1425 const bool utf8_target = PL_reg_match_utf8;
1426 UV utf8_fold_flags = 0;
1427 RXi_GET_DECL(prog,progi);
1429 PERL_ARGS_ASSERT_FIND_BYCLASS;
1431 /* We know what class it must start with. */
1435 if (utf8_target || OP(c) == ANYOFV) {
1436 STRLEN inclasslen = strend - s;
1437 REXEC_FBC_UTF8_CLASS_SCAN(
1438 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1441 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1446 if (tmp && (!reginfo || regtry(reginfo, &s)))
1454 if (UTF_PATTERN || utf8_target) {
1455 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1456 goto do_exactf_utf8;
1458 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1459 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1460 goto do_exactf_non_utf8; /* isn't dealt with by these */
1465 /* regcomp.c already folded this if pattern is in UTF-8 */
1466 utf8_fold_flags = 0;
1467 goto do_exactf_utf8;
1469 fold_array = PL_fold;
1471 goto do_exactf_non_utf8;
1474 if (UTF_PATTERN || utf8_target) {
1475 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1476 goto do_exactf_utf8;
1478 fold_array = PL_fold_locale;
1479 folder = foldEQ_locale;
1480 goto do_exactf_non_utf8;
1484 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1486 goto do_exactf_utf8;
1488 case EXACTFU_TRICKYFOLD:
1490 if (UTF_PATTERN || utf8_target) {
1491 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1492 goto do_exactf_utf8;
1495 /* Any 'ss' in the pattern should have been replaced by regcomp,
1496 * so we don't have to worry here about this single special case
1497 * in the Latin1 range */
1498 fold_array = PL_fold_latin1;
1499 folder = foldEQ_latin1;
1503 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1504 are no glitches with fold-length differences
1505 between the target string and pattern */
1507 /* The idea in the non-utf8 EXACTF* cases is to first find the
1508 * first character of the EXACTF* node and then, if necessary,
1509 * case-insensitively compare the full text of the node. c1 is the
1510 * first character. c2 is its fold. This logic will not work for
1511 * Unicode semantics and the german sharp ss, which hence should
1512 * not be compiled into a node that gets here. */
1513 pat_string = STRING(c);
1514 ln = STR_LEN(c); /* length to match in octets/bytes */
1516 /* We know that we have to match at least 'ln' bytes (which is the
1517 * same as characters, since not utf8). If we have to match 3
1518 * characters, and there are only 2 availabe, we know without
1519 * trying that it will fail; so don't start a match past the
1520 * required minimum number from the far end */
1521 e = HOP3c(strend, -((I32)ln), s);
1523 if (!reginfo && e < s) {
1524 e = s; /* Due to minlen logic of intuit() */
1528 c2 = fold_array[c1];
1529 if (c1 == c2) { /* If char and fold are the same */
1530 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1533 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1542 /* If one of the operands is in utf8, we can't use the simpler
1543 * folding above, due to the fact that many different characters
1544 * can have the same fold, or portion of a fold, or different-
1546 pat_string = STRING(c);
1547 ln = STR_LEN(c); /* length to match in octets/bytes */
1548 pat_end = pat_string + ln;
1549 lnc = (UTF_PATTERN) /* length to match in characters */
1550 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1553 /* We have 'lnc' characters to match in the pattern, but because of
1554 * multi-character folding, each character in the target can match
1555 * up to 3 characters (Unicode guarantees it will never exceed
1556 * this) if it is utf8-encoded; and up to 2 if not (based on the
1557 * fact that the Latin 1 folds are already determined, and the
1558 * only multi-char fold in that range is the sharp-s folding to
1559 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1560 * string character. Adjust lnc accordingly, rounding up, so that
1561 * if we need to match at least 4+1/3 chars, that really is 5. */
1562 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1563 lnc = (lnc + expansion - 1) / expansion;
1565 /* As in the non-UTF8 case, if we have to match 3 characters, and
1566 * only 2 are left, it's guaranteed to fail, so don't start a
1567 * match that would require us to go beyond the end of the string
1569 e = HOP3c(strend, -((I32)lnc), s);
1571 if (!reginfo && e < s) {
1572 e = s; /* Due to minlen logic of intuit() */
1575 /* XXX Note that we could recalculate e to stop the loop earlier,
1576 * as the worst case expansion above will rarely be met, and as we
1577 * go along we would usually find that e moves further to the left.
1578 * This would happen only after we reached the point in the loop
1579 * where if there were no expansion we should fail. Unclear if
1580 * worth the expense */
1583 char *my_strend= (char *)strend;
1584 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1585 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1586 && (!reginfo || regtry(reginfo, &s)) )
1590 s += (utf8_target) ? UTF8SKIP(s) : 1;
1595 PL_reg_flags |= RF_tainted;
1596 FBC_BOUND(isALNUM_LC,
1597 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1598 isALNUM_LC_utf8((U8*)s));
1601 PL_reg_flags |= RF_tainted;
1602 FBC_NBOUND(isALNUM_LC,
1603 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1604 isALNUM_LC_utf8((U8*)s));
1607 FBC_BOUND(isWORDCHAR,
1609 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1612 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1614 isWORDCHAR_A((U8*)s));
1617 FBC_NBOUND(isWORDCHAR,
1619 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1622 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1624 isWORDCHAR_A((U8*)s));
1627 FBC_BOUND(isWORDCHAR_L1,
1629 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1632 FBC_NBOUND(isWORDCHAR_L1,
1634 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1637 REXEC_FBC_CSCAN_TAINT(
1638 isALNUM_LC_utf8((U8*)s),
1643 REXEC_FBC_CSCAN_PRELOAD(
1644 LOAD_UTF8_CHARCLASS_ALNUM(),
1645 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1646 isWORDCHAR_L1((U8) *s)
1650 REXEC_FBC_CSCAN_PRELOAD(
1651 LOAD_UTF8_CHARCLASS_ALNUM(),
1652 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1657 /* Don't need to worry about utf8, as it can match only a single
1658 * byte invariant character */
1659 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1662 REXEC_FBC_CSCAN_PRELOAD(
1663 LOAD_UTF8_CHARCLASS_ALNUM(),
1664 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1665 ! isWORDCHAR_L1((U8) *s)
1669 REXEC_FBC_CSCAN_PRELOAD(
1670 LOAD_UTF8_CHARCLASS_ALNUM(),
1671 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1682 REXEC_FBC_CSCAN_TAINT(
1683 !isALNUM_LC_utf8((U8*)s),
1688 REXEC_FBC_CSCAN_PRELOAD(
1689 LOAD_UTF8_CHARCLASS_SPACE(),
1690 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1695 REXEC_FBC_CSCAN_PRELOAD(
1696 LOAD_UTF8_CHARCLASS_SPACE(),
1697 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1702 /* Don't need to worry about utf8, as it can match only a single
1703 * byte invariant character */
1704 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1707 REXEC_FBC_CSCAN_TAINT(
1708 isSPACE_LC_utf8((U8*)s),
1713 REXEC_FBC_CSCAN_PRELOAD(
1714 LOAD_UTF8_CHARCLASS_SPACE(),
1715 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1716 ! isSPACE_L1((U8) *s)
1720 REXEC_FBC_CSCAN_PRELOAD(
1721 LOAD_UTF8_CHARCLASS_SPACE(),
1722 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1733 REXEC_FBC_CSCAN_TAINT(
1734 !isSPACE_LC_utf8((U8*)s),
1739 REXEC_FBC_CSCAN_PRELOAD(
1740 LOAD_UTF8_CHARCLASS_DIGIT(),
1741 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1746 /* Don't need to worry about utf8, as it can match only a single
1747 * byte invariant character */
1748 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1751 REXEC_FBC_CSCAN_TAINT(
1752 isDIGIT_LC_utf8((U8*)s),
1757 REXEC_FBC_CSCAN_PRELOAD(
1758 LOAD_UTF8_CHARCLASS_DIGIT(),
1759 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1770 REXEC_FBC_CSCAN_TAINT(
1771 !isDIGIT_LC_utf8((U8*)s),
1778 is_LNBREAK_latin1(s)
1790 !is_VERTWS_latin1(s)
1796 is_HORIZWS_latin1(s)
1801 !is_HORIZWS_utf8(s),
1802 !is_HORIZWS_latin1(s)
1806 /* Don't need to worry about utf8, as it can match only a single
1807 * byte invariant character. The flag in this node type is the
1808 * class number to pass to _generic_isCC() to build a mask for
1809 * searching in PL_charclass[] */
1810 REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
1814 !_generic_isCC_A(*s, FLAGS(c)),
1815 !_generic_isCC_A(*s, FLAGS(c))
1823 /* what trie are we using right now */
1825 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1827 = (reg_trie_data*)progi->data->data[ aho->trie ];
1828 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1830 const char *last_start = strend - trie->minlen;
1832 const char *real_start = s;
1834 STRLEN maxlen = trie->maxlen;
1836 U8 **points; /* map of where we were in the input string
1837 when reading a given char. For ASCII this
1838 is unnecessary overhead as the relationship
1839 is always 1:1, but for Unicode, especially
1840 case folded Unicode this is not true. */
1841 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1845 GET_RE_DEBUG_FLAGS_DECL;
1847 /* We can't just allocate points here. We need to wrap it in
1848 * an SV so it gets freed properly if there is a croak while
1849 * running the match */
1852 sv_points=newSV(maxlen * sizeof(U8 *));
1853 SvCUR_set(sv_points,
1854 maxlen * sizeof(U8 *));
1855 SvPOK_on(sv_points);
1856 sv_2mortal(sv_points);
1857 points=(U8**)SvPV_nolen(sv_points );
1858 if ( trie_type != trie_utf8_fold
1859 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1862 bitmap=(U8*)trie->bitmap;
1864 bitmap=(U8*)ANYOF_BITMAP(c);
1866 /* this is the Aho-Corasick algorithm modified a touch
1867 to include special handling for long "unknown char"
1868 sequences. The basic idea being that we use AC as long
1869 as we are dealing with a possible matching char, when
1870 we encounter an unknown char (and we have not encountered
1871 an accepting state) we scan forward until we find a legal
1873 AC matching is basically that of trie matching, except
1874 that when we encounter a failing transition, we fall back
1875 to the current states "fail state", and try the current char
1876 again, a process we repeat until we reach the root state,
1877 state 1, or a legal transition. If we fail on the root state
1878 then we can either terminate if we have reached an accepting
1879 state previously, or restart the entire process from the beginning
1883 while (s <= last_start) {
1884 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1892 U8 *uscan = (U8*)NULL;
1893 U8 *leftmost = NULL;
1895 U32 accepted_word= 0;
1899 while ( state && uc <= (U8*)strend ) {
1901 U32 word = aho->states[ state ].wordnum;
1905 DEBUG_TRIE_EXECUTE_r(
1906 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1907 dump_exec_pos( (char *)uc, c, strend, real_start,
1908 (char *)uc, utf8_target );
1909 PerlIO_printf( Perl_debug_log,
1910 " Scanning for legal start char...\n");
1914 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1918 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1924 if (uc >(U8*)last_start) break;
1928 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1929 if (!leftmost || lpos < leftmost) {
1930 DEBUG_r(accepted_word=word);
1936 points[pointpos++ % maxlen]= uc;
1937 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1938 uscan, len, uvc, charid, foldlen,
1940 DEBUG_TRIE_EXECUTE_r({
1941 dump_exec_pos( (char *)uc, c, strend, real_start,
1943 PerlIO_printf(Perl_debug_log,
1944 " Charid:%3u CP:%4"UVxf" ",
1950 word = aho->states[ state ].wordnum;
1952 base = aho->states[ state ].trans.base;
1954 DEBUG_TRIE_EXECUTE_r({
1956 dump_exec_pos( (char *)uc, c, strend, real_start,
1958 PerlIO_printf( Perl_debug_log,
1959 "%sState: %4"UVxf", word=%"UVxf,
1960 failed ? " Fail transition to " : "",
1961 (UV)state, (UV)word);
1967 ( ((offset = base + charid
1968 - 1 - trie->uniquecharcount)) >= 0)
1969 && ((U32)offset < trie->lasttrans)
1970 && trie->trans[offset].check == state
1971 && (tmp=trie->trans[offset].next))
1973 DEBUG_TRIE_EXECUTE_r(
1974 PerlIO_printf( Perl_debug_log," - legal\n"));
1979 DEBUG_TRIE_EXECUTE_r(
1980 PerlIO_printf( Perl_debug_log," - fail\n"));
1982 state = aho->fail[state];
1986 /* we must be accepting here */
1987 DEBUG_TRIE_EXECUTE_r(
1988 PerlIO_printf( Perl_debug_log," - accepting\n"));
1997 if (!state) state = 1;
2000 if ( aho->states[ state ].wordnum ) {
2001 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2002 if (!leftmost || lpos < leftmost) {
2003 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2008 s = (char*)leftmost;
2009 DEBUG_TRIE_EXECUTE_r({
2011 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2012 (UV)accepted_word, (IV)(s - real_start)
2015 if (!reginfo || regtry(reginfo, &s)) {
2021 DEBUG_TRIE_EXECUTE_r({
2022 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2025 DEBUG_TRIE_EXECUTE_r(
2026 PerlIO_printf( Perl_debug_log,"No match.\n"));
2035 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2045 - regexec_flags - match a regexp against a string
2048 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2049 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2050 /* stringarg: the point in the string at which to begin matching */
2051 /* strend: pointer to null at end of string */
2052 /* strbeg: real beginning of string */
2053 /* minend: end of match must be >= minend bytes after stringarg. */
2054 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2055 * itself is accessed via the pointers above */
2056 /* data: May be used for some additional optimizations.
2057 Currently its only used, with a U32 cast, for transmitting
2058 the ganch offset when doing a /g match. This will change */
2059 /* nosave: For optimizations. */
2063 struct regexp *const prog = (struct regexp *)SvANY(rx);
2064 /*register*/ char *s;
2066 /*register*/ char *startpos = stringarg;
2067 I32 minlen; /* must match at least this many chars */
2068 I32 dontbother = 0; /* how many characters not to try at end */
2069 I32 end_shift = 0; /* Same for the end. */ /* CC */
2070 I32 scream_pos = -1; /* Internal iterator of scream. */
2071 char *scream_olds = NULL;
2072 const bool utf8_target = cBOOL(DO_UTF8(sv));
2074 RXi_GET_DECL(prog,progi);
2075 regmatch_info reginfo; /* create some info to pass to regtry etc */
2076 regexp_paren_pair *swap = NULL;
2077 GET_RE_DEBUG_FLAGS_DECL;
2079 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2080 PERL_UNUSED_ARG(data);
2082 /* Be paranoid... */
2083 if (prog == NULL || startpos == NULL) {
2084 Perl_croak(aTHX_ "NULL regexp parameter");
2088 multiline = prog->extflags & RXf_PMf_MULTILINE;
2089 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2091 RX_MATCH_UTF8_set(rx, utf8_target);
2093 debug_start_match(rx, utf8_target, startpos, strend,
2097 minlen = prog->minlen;
2099 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2100 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2101 "String too short [regexec_flags]...\n"));
2106 /* Check validity of program. */
2107 if (UCHARAT(progi->program) != REG_MAGIC) {
2108 Perl_croak(aTHX_ "corrupted regexp program");
2112 PL_reg_state.re_state_eval_setup_done = FALSE;
2116 PL_reg_flags |= RF_utf8;
2118 /* Mark beginning of line for ^ and lookbehind. */
2119 reginfo.bol = startpos; /* XXX not used ??? */
2123 /* Mark end of line for $ (and such) */
2126 /* see how far we have to get to not match where we matched before */
2127 reginfo.till = startpos+minend;
2129 /* If there is a "must appear" string, look for it. */
2132 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2134 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2135 reginfo.ganch = startpos + prog->gofs;
2136 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2137 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2138 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2140 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2141 && mg->mg_len >= 0) {
2142 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2143 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2144 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2146 if (prog->extflags & RXf_ANCH_GPOS) {
2147 if (s > reginfo.ganch)
2149 s = reginfo.ganch - prog->gofs;
2150 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2151 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2157 reginfo.ganch = strbeg + PTR2UV(data);
2158 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2159 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2161 } else { /* pos() not defined */
2162 reginfo.ganch = strbeg;
2163 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2164 "GPOS: reginfo.ganch = strbeg\n"));
2167 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2168 /* We have to be careful. If the previous successful match
2169 was from this regex we don't want a subsequent partially
2170 successful match to clobber the old results.
2171 So when we detect this possibility we add a swap buffer
2172 to the re, and switch the buffer each match. If we fail
2173 we switch it back, otherwise we leave it swapped.
2176 /* do we need a save destructor here for eval dies? */
2177 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2178 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2179 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2185 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2186 re_scream_pos_data d;
2188 d.scream_olds = &scream_olds;
2189 d.scream_pos = &scream_pos;
2190 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2192 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2193 goto phooey; /* not present */
2199 /* Simplest case: anchored match need be tried only once. */
2200 /* [unless only anchor is BOL and multiline is set] */
2201 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2202 if (s == startpos && regtry(®info, &startpos))
2204 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2205 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2210 dontbother = minlen - 1;
2211 end = HOP3c(strend, -dontbother, strbeg) - 1;
2212 /* for multiline we only have to try after newlines */
2213 if (prog->check_substr || prog->check_utf8) {
2214 /* because of the goto we can not easily reuse the macros for bifurcating the
2215 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2218 goto after_try_utf8;
2220 if (regtry(®info, &s)) {
2227 if (prog->extflags & RXf_USE_INTUIT) {
2228 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2237 } /* end search for check string in unicode */
2239 if (s == startpos) {
2240 goto after_try_latin;
2243 if (regtry(®info, &s)) {
2250 if (prog->extflags & RXf_USE_INTUIT) {
2251 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2260 } /* end search for check string in latin*/
2261 } /* end search for check string */
2262 else { /* search for newline */
2264 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2267 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2268 while (s <= end) { /* note it could be possible to match at the end of the string */
2269 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2270 if (regtry(®info, &s))
2274 } /* end search for newline */
2275 } /* end anchored/multiline check string search */
2277 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2279 /* the warning about reginfo.ganch being used without initialization
2280 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2281 and we only enter this block when the same bit is set. */
2282 char *tmp_s = reginfo.ganch - prog->gofs;
2284 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2289 /* Messy cases: unanchored match. */
2290 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2291 /* we have /x+whatever/ */
2292 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2297 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2298 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2299 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2304 DEBUG_EXECUTE_r( did_match = 1 );
2305 if (regtry(®info, &s)) goto got_it;
2307 while (s < strend && *s == ch)
2315 DEBUG_EXECUTE_r( did_match = 1 );
2316 if (regtry(®info, &s)) goto got_it;
2318 while (s < strend && *s == ch)
2323 DEBUG_EXECUTE_r(if (!did_match)
2324 PerlIO_printf(Perl_debug_log,
2325 "Did not find anchored character...\n")
2328 else if (prog->anchored_substr != NULL
2329 || prog->anchored_utf8 != NULL
2330 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2331 && prog->float_max_offset < strend - s)) {
2336 char *last1; /* Last position checked before */
2340 if (prog->anchored_substr || prog->anchored_utf8) {
2341 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2342 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2343 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2344 back_max = back_min = prog->anchored_offset;
2346 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2347 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2348 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2349 back_max = prog->float_max_offset;
2350 back_min = prog->float_min_offset;
2354 if (must == &PL_sv_undef)
2355 /* could not downgrade utf8 check substring, so must fail */
2361 last = HOP3c(strend, /* Cannot start after this */
2362 -(I32)(CHR_SVLEN(must)
2363 - (SvTAIL(must) != 0) + back_min), strbeg);
2366 last1 = HOPc(s, -1);
2368 last1 = s - 1; /* bogus */
2370 /* XXXX check_substr already used to find "s", can optimize if
2371 check_substr==must. */
2373 dontbother = end_shift;
2374 strend = HOPc(strend, -dontbother);
2375 while ( (s <= last) &&
2376 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2377 (unsigned char*)strend, must,
2378 multiline ? FBMrf_MULTILINE : 0)) ) {
2379 DEBUG_EXECUTE_r( did_match = 1 );
2380 if (HOPc(s, -back_max) > last1) {
2381 last1 = HOPc(s, -back_min);
2382 s = HOPc(s, -back_max);
2385 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2387 last1 = HOPc(s, -back_min);
2391 while (s <= last1) {
2392 if (regtry(®info, &s))
2398 while (s <= last1) {
2399 if (regtry(®info, &s))
2405 DEBUG_EXECUTE_r(if (!did_match) {
2406 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2407 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2408 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2409 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2410 ? "anchored" : "floating"),
2411 quoted, RE_SV_TAIL(must));
2415 else if ( (c = progi->regstclass) ) {
2417 const OPCODE op = OP(progi->regstclass);
2418 /* don't bother with what can't match */
2419 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2420 strend = HOPc(strend, -(minlen - 1));
2423 SV * const prop = sv_newmortal();
2424 regprop(prog, prop, c);
2426 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2428 PerlIO_printf(Perl_debug_log,
2429 "Matching stclass %.*s against %s (%d bytes)\n",
2430 (int)SvCUR(prop), SvPVX_const(prop),
2431 quoted, (int)(strend - s));
2434 if (find_byclass(prog, c, s, strend, ®info))
2436 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2440 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2447 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2448 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2449 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2451 little = SvPV_const(float_real, len);
2452 if (SvTAIL(float_real)) {
2453 /* This means that float_real contains an artificial \n on the end
2454 * due to the presence of something like this: /foo$/
2455 * where we can match both "foo" and "foo\n" at the end of the string.
2456 * So we have to compare the end of the string first against the float_real
2457 * without the \n and then against the full float_real with the string.
2458 * We have to watch out for cases where the string might be smaller
2459 * than the float_real or the float_real without the \n.
2461 char *checkpos= strend - len;
2463 PerlIO_printf(Perl_debug_log,
2464 "%sChecking for float_real.%s\n",
2465 PL_colors[4], PL_colors[5]));
2466 if (checkpos + 1 < strbeg) {
2467 /* can't match, even if we remove the trailing \n string is too short to match */
2469 PerlIO_printf(Perl_debug_log,
2470 "%sString shorter than required trailing substring, cannot match.%s\n",
2471 PL_colors[4], PL_colors[5]));
2473 } else if (memEQ(checkpos + 1, little, len - 1)) {
2474 /* can match, the end of the string matches without the "\n" */
2475 last = checkpos + 1;
2476 } else if (checkpos < strbeg) {
2477 /* cant match, string is too short when the "\n" is included */
2479 PerlIO_printf(Perl_debug_log,
2480 "%sString does not contain required trailing substring, cannot match.%s\n",
2481 PL_colors[4], PL_colors[5]));
2483 } else if (!multiline) {
2484 /* non multiline match, so compare with the "\n" at the end of the string */
2485 if (memEQ(checkpos, little, len)) {
2489 PerlIO_printf(Perl_debug_log,
2490 "%sString does not contain required trailing substring, cannot match.%s\n",
2491 PL_colors[4], PL_colors[5]));
2495 /* multiline match, so we have to search for a place where the full string is located */
2501 last = rninstr(s, strend, little, little + len);
2503 last = strend; /* matching "$" */
2506 /* at one point this block contained a comment which was probably
2507 * incorrect, which said that this was a "should not happen" case.
2508 * Even if it was true when it was written I am pretty sure it is
2509 * not anymore, so I have removed the comment and replaced it with
2512 PerlIO_printf(Perl_debug_log,
2513 "String does not contain required substring, cannot match.\n"
2517 dontbother = strend - last + prog->float_min_offset;
2519 if (minlen && (dontbother < minlen))
2520 dontbother = minlen - 1;
2521 strend -= dontbother; /* this one's always in bytes! */
2522 /* We don't know much -- general case. */
2525 if (regtry(®info, &s))
2534 if (regtry(®info, &s))
2536 } while (s++ < strend);
2546 PerlIO_printf(Perl_debug_log,
2547 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2553 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2555 if (PL_reg_state.re_state_eval_setup_done)
2556 restore_pos(aTHX_ prog);
2557 if (RXp_PAREN_NAMES(prog))
2558 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2560 /* make sure $`, $&, $', and $digit will work later */
2561 if ( !(flags & REXEC_NOT_FIRST) ) {
2562 if (flags & REXEC_COPY_STR) {
2563 #ifdef PERL_OLD_COPY_ON_WRITE
2565 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2567 PerlIO_printf(Perl_debug_log,
2568 "Copy on write: regexp capture, type %d\n",
2571 RX_MATCH_COPY_FREE(rx);
2572 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2573 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2574 assert (SvPOKp(prog->saved_copy));
2575 prog->sublen = PL_regeol - strbeg;
2576 prog->suboffset = 0;
2577 prog->subcoffset = 0;
2582 I32 max = PL_regeol - strbeg;
2585 if ( (flags & REXEC_COPY_SKIP_POST)
2586 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2587 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2588 ) { /* don't copy $' part of string */
2591 /* calculate the right-most part of the string covered
2592 * by a capture. Due to look-ahead, this may be to
2593 * the right of $&, so we have to scan all captures */
2594 while (n <= prog->lastparen) {
2595 if (prog->offs[n].end > max)
2596 max = prog->offs[n].end;
2600 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2601 ? prog->offs[0].start
2603 assert(max >= 0 && max <= PL_regeol - strbeg);
2606 if ( (flags & REXEC_COPY_SKIP_PRE)
2607 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2608 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2609 ) { /* don't copy $` part of string */
2612 /* calculate the left-most part of the string covered
2613 * by a capture. Due to look-behind, this may be to
2614 * the left of $&, so we have to scan all captures */
2615 while (min && n <= prog->lastparen) {
2616 if ( prog->offs[n].start != -1
2617 && prog->offs[n].start < min)
2619 min = prog->offs[n].start;
2623 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2624 && min > prog->offs[0].end
2626 min = prog->offs[0].end;
2630 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2633 if (RX_MATCH_COPIED(rx)) {
2634 if (sublen > prog->sublen)
2636 (char*)saferealloc(prog->subbeg, sublen+1);
2639 prog->subbeg = (char*)safemalloc(sublen+1);
2640 Copy(strbeg + min, prog->subbeg, sublen, char);
2641 prog->subbeg[sublen] = '\0';
2642 prog->suboffset = min;
2643 prog->sublen = sublen;
2644 RX_MATCH_COPIED_on(rx);
2646 prog->subcoffset = prog->suboffset;
2647 if (prog->suboffset && utf8_target) {
2648 /* Convert byte offset to chars.
2649 * XXX ideally should only compute this if @-/@+
2650 * has been seen, a la PL_sawampersand ??? */
2652 /* If there's a direct correspondence between the
2653 * string which we're matching and the original SV,
2654 * then we can use the utf8 len cache associated with
2655 * the SV. In particular, it means that under //g,
2656 * sv_pos_b2u() will use the previously cached
2657 * position to speed up working out the new length of
2658 * subcoffset, rather than counting from the start of
2659 * the string each time. This stops
2660 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2661 * from going quadratic */
2662 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2663 sv_pos_b2u(sv, &(prog->subcoffset));
2665 prog->subcoffset = utf8_length((U8*)strbeg,
2666 (U8*)(strbeg+prog->suboffset));
2670 RX_MATCH_COPY_FREE(rx);
2671 prog->subbeg = strbeg;
2672 prog->suboffset = 0;
2673 prog->subcoffset = 0;
2674 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2681 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2682 PL_colors[4], PL_colors[5]));
2683 if (PL_reg_state.re_state_eval_setup_done)
2684 restore_pos(aTHX_ prog);
2686 /* we failed :-( roll it back */
2687 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2688 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2693 Safefree(prog->offs);
2701 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2702 * Do inc before dec, in case old and new rex are the same */
2703 #define SET_reg_curpm(Re2) \
2704 if (PL_reg_state.re_state_eval_setup_done) { \
2705 (void)ReREFCNT_inc(Re2); \
2706 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2707 PM_SETRE((PL_reg_curpm), (Re2)); \
2712 - regtry - try match at specific point
2714 STATIC I32 /* 0 failure, 1 success */
2715 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2719 REGEXP *const rx = reginfo->prog;
2720 regexp *const prog = (struct regexp *)SvANY(rx);
2722 RXi_GET_DECL(prog,progi);
2723 GET_RE_DEBUG_FLAGS_DECL;
2725 PERL_ARGS_ASSERT_REGTRY;
2727 reginfo->cutpoint=NULL;
2729 if ((prog->extflags & RXf_EVAL_SEEN)
2730 && !PL_reg_state.re_state_eval_setup_done)
2734 PL_reg_state.re_state_eval_setup_done = TRUE;
2736 /* Make $_ available to executed code. */
2737 if (reginfo->sv != DEFSV) {
2739 DEFSV_set(reginfo->sv);
2742 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2743 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2744 /* prepare for quick setting of pos */
2745 #ifdef PERL_OLD_COPY_ON_WRITE
2746 if (SvIsCOW(reginfo->sv))
2747 sv_force_normal_flags(reginfo->sv, 0);
2749 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2750 &PL_vtbl_mglob, NULL, 0);
2754 PL_reg_oldpos = mg->mg_len;
2755 SAVEDESTRUCTOR_X(restore_pos, prog);
2757 if (!PL_reg_curpm) {
2758 Newxz(PL_reg_curpm, 1, PMOP);
2761 SV* const repointer = &PL_sv_undef;
2762 /* this regexp is also owned by the new PL_reg_curpm, which
2763 will try to free it. */
2764 av_push(PL_regex_padav, repointer);
2765 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2766 PL_regex_pad = AvARRAY(PL_regex_padav);
2771 PL_reg_oldcurpm = PL_curpm;
2772 PL_curpm = PL_reg_curpm;
2773 if (RXp_MATCH_COPIED(prog)) {
2774 /* Here is a serious problem: we cannot rewrite subbeg,
2775 since it may be needed if this match fails. Thus
2776 $` inside (?{}) could fail... */
2777 PL_reg_oldsaved = prog->subbeg;
2778 PL_reg_oldsavedlen = prog->sublen;
2779 PL_reg_oldsavedoffset = prog->suboffset;
2780 PL_reg_oldsavedcoffset = prog->suboffset;
2781 #ifdef PERL_OLD_COPY_ON_WRITE
2782 PL_nrs = prog->saved_copy;
2784 RXp_MATCH_COPIED_off(prog);
2787 PL_reg_oldsaved = NULL;
2788 prog->subbeg = PL_bostr;
2789 prog->suboffset = 0;
2790 prog->subcoffset = 0;
2791 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2794 PL_reg_starttry = *startposp;
2796 prog->offs[0].start = *startposp - PL_bostr;
2797 prog->lastparen = 0;
2798 prog->lastcloseparen = 0;
2801 /* XXXX What this code is doing here?!!! There should be no need
2802 to do this again and again, prog->lastparen should take care of
2805 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2806 * Actually, the code in regcppop() (which Ilya may be meaning by
2807 * prog->lastparen), is not needed at all by the test suite
2808 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2809 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2810 * Meanwhile, this code *is* needed for the
2811 * above-mentioned test suite tests to succeed. The common theme
2812 * on those tests seems to be returning null fields from matches.
2813 * --jhi updated by dapm */
2815 if (prog->nparens) {
2816 regexp_paren_pair *pp = prog->offs;
2818 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2826 result = regmatch(reginfo, *startposp, progi->program + 1);
2828 prog->offs[0].end = result;
2831 if (reginfo->cutpoint)
2832 *startposp= reginfo->cutpoint;
2833 REGCP_UNWIND(lastcp);
2838 #define sayYES goto yes
2839 #define sayNO goto no
2840 #define sayNO_SILENT goto no_silent
2842 /* we dont use STMT_START/END here because it leads to
2843 "unreachable code" warnings, which are bogus, but distracting. */
2844 #define CACHEsayNO \
2845 if (ST.cache_mask) \
2846 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2849 /* this is used to determine how far from the left messages like
2850 'failed...' are printed. It should be set such that messages
2851 are inline with the regop output that created them.
2853 #define REPORT_CODE_OFF 32
2856 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2857 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2859 #define SLAB_FIRST(s) (&(s)->states[0])
2860 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2862 /* grab a new slab and return the first slot in it */
2864 STATIC regmatch_state *
2867 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2870 regmatch_slab *s = PL_regmatch_slab->next;
2872 Newx(s, 1, regmatch_slab);
2873 s->prev = PL_regmatch_slab;
2875 PL_regmatch_slab->next = s;
2877 PL_regmatch_slab = s;
2878 return SLAB_FIRST(s);
2882 /* push a new state then goto it */
2884 #define PUSH_STATE_GOTO(state, node, input) \
2885 pushinput = input; \
2887 st->resume_state = state; \
2890 /* push a new state with success backtracking, then goto it */
2892 #define PUSH_YES_STATE_GOTO(state, node, input) \
2893 pushinput = input; \
2895 st->resume_state = state; \
2896 goto push_yes_state;
2903 regmatch() - main matching routine
2905 This is basically one big switch statement in a loop. We execute an op,
2906 set 'next' to point the next op, and continue. If we come to a point which
2907 we may need to backtrack to on failure such as (A|B|C), we push a
2908 backtrack state onto the backtrack stack. On failure, we pop the top
2909 state, and re-enter the loop at the state indicated. If there are no more
2910 states to pop, we return failure.
2912 Sometimes we also need to backtrack on success; for example /A+/, where
2913 after successfully matching one A, we need to go back and try to
2914 match another one; similarly for lookahead assertions: if the assertion
2915 completes successfully, we backtrack to the state just before the assertion
2916 and then carry on. In these cases, the pushed state is marked as
2917 'backtrack on success too'. This marking is in fact done by a chain of
2918 pointers, each pointing to the previous 'yes' state. On success, we pop to
2919 the nearest yes state, discarding any intermediate failure-only states.
2920 Sometimes a yes state is pushed just to force some cleanup code to be
2921 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2922 it to free the inner regex.
2924 Note that failure backtracking rewinds the cursor position, while
2925 success backtracking leaves it alone.
2927 A pattern is complete when the END op is executed, while a subpattern
2928 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2929 ops trigger the "pop to last yes state if any, otherwise return true"
2932 A common convention in this function is to use A and B to refer to the two
2933 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2934 the subpattern to be matched possibly multiple times, while B is the entire
2935 rest of the pattern. Variable and state names reflect this convention.
2937 The states in the main switch are the union of ops and failure/success of
2938 substates associated with with that op. For example, IFMATCH is the op
2939 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2940 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2941 successfully matched A and IFMATCH_A_fail is a state saying that we have
2942 just failed to match A. Resume states always come in pairs. The backtrack
2943 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2944 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2945 on success or failure.
2947 The struct that holds a backtracking state is actually a big union, with
2948 one variant for each major type of op. The variable st points to the
2949 top-most backtrack struct. To make the code clearer, within each
2950 block of code we #define ST to alias the relevant union.
2952 Here's a concrete example of a (vastly oversimplified) IFMATCH
2958 #define ST st->u.ifmatch
2960 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2961 ST.foo = ...; // some state we wish to save
2963 // push a yes backtrack state with a resume value of
2964 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2966 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
2969 case IFMATCH_A: // we have successfully executed A; now continue with B
2971 bar = ST.foo; // do something with the preserved value
2974 case IFMATCH_A_fail: // A failed, so the assertion failed
2975 ...; // do some housekeeping, then ...
2976 sayNO; // propagate the failure
2983 For any old-timers reading this who are familiar with the old recursive
2984 approach, the code above is equivalent to:
2986 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2995 ...; // do some housekeeping, then ...
2996 sayNO; // propagate the failure
2999 The topmost backtrack state, pointed to by st, is usually free. If you
3000 want to claim it, populate any ST.foo fields in it with values you wish to
3001 save, then do one of
3003 PUSH_STATE_GOTO(resume_state, node, newinput);
3004 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3006 which sets that backtrack state's resume value to 'resume_state', pushes a
3007 new free entry to the top of the backtrack stack, then goes to 'node'.
3008 On backtracking, the free slot is popped, and the saved state becomes the
3009 new free state. An ST.foo field in this new top state can be temporarily
3010 accessed to retrieve values, but once the main loop is re-entered, it
3011 becomes available for reuse.
3013 Note that the depth of the backtrack stack constantly increases during the
3014 left-to-right execution of the pattern, rather than going up and down with
3015 the pattern nesting. For example the stack is at its maximum at Z at the
3016 end of the pattern, rather than at X in the following:
3018 /(((X)+)+)+....(Y)+....Z/
3020 The only exceptions to this are lookahead/behind assertions and the cut,
3021 (?>A), which pop all the backtrack states associated with A before
3024 Backtrack state structs are allocated in slabs of about 4K in size.
3025 PL_regmatch_state and st always point to the currently active state,
3026 and PL_regmatch_slab points to the slab currently containing
3027 PL_regmatch_state. The first time regmatch() is called, the first slab is
3028 allocated, and is never freed until interpreter destruction. When the slab
3029 is full, a new one is allocated and chained to the end. At exit from
3030 regmatch(), slabs allocated since entry are freed.
3035 #define DEBUG_STATE_pp(pp) \
3037 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3038 PerlIO_printf(Perl_debug_log, \
3039 " %*s"pp" %s%s%s%s%s\n", \
3041 PL_reg_name[st->resume_state], \
3042 ((st==yes_state||st==mark_state) ? "[" : ""), \
3043 ((st==yes_state) ? "Y" : ""), \
3044 ((st==mark_state) ? "M" : ""), \
3045 ((st==yes_state||st==mark_state) ? "]" : "") \
3050 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3055 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3056 const char *start, const char *end, const char *blurb)
3058 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3060 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3065 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3066 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3068 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3069 start, end - start, 60);
3071 PerlIO_printf(Perl_debug_log,
3072 "%s%s REx%s %s against %s\n",
3073 PL_colors[4], blurb, PL_colors[5], s0, s1);
3075 if (utf8_target||utf8_pat)
3076 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3077 utf8_pat ? "pattern" : "",
3078 utf8_pat && utf8_target ? " and " : "",
3079 utf8_target ? "string" : ""
3085 S_dump_exec_pos(pTHX_ const char *locinput,
3086 const regnode *scan,
3087 const char *loc_regeol,
3088 const char *loc_bostr,
3089 const char *loc_reg_starttry,
3090 const bool utf8_target)
3092 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3093 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3094 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3095 /* The part of the string before starttry has one color
3096 (pref0_len chars), between starttry and current
3097 position another one (pref_len - pref0_len chars),
3098 after the current position the third one.
3099 We assume that pref0_len <= pref_len, otherwise we
3100 decrease pref0_len. */
3101 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3102 ? (5 + taill) - l : locinput - loc_bostr;
3105 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3107 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3109 pref0_len = pref_len - (locinput - loc_reg_starttry);
3110 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3111 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3112 ? (5 + taill) - pref_len : loc_regeol - locinput);
3113 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3117 if (pref0_len > pref_len)
3118 pref0_len = pref_len;
3120 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3122 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3123 (locinput - pref_len),pref0_len, 60, 4, 5);
3125 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3126 (locinput - pref_len + pref0_len),
3127 pref_len - pref0_len, 60, 2, 3);
3129 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3130 locinput, loc_regeol - locinput, 10, 0, 1);
3132 const STRLEN tlen=len0+len1+len2;
3133 PerlIO_printf(Perl_debug_log,
3134 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3135 (IV)(locinput - loc_bostr),
3138 (docolor ? "" : "> <"),
3140 (int)(tlen > 19 ? 0 : 19 - tlen),
3147 /* reg_check_named_buff_matched()
3148 * Checks to see if a named buffer has matched. The data array of
3149 * buffer numbers corresponding to the buffer is expected to reside
3150 * in the regexp->data->data array in the slot stored in the ARG() of
3151 * node involved. Note that this routine doesn't actually care about the
3152 * name, that information is not preserved from compilation to execution.
3153 * Returns the index of the leftmost defined buffer with the given name
3154 * or 0 if non of the buffers matched.
3157 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3160 RXi_GET_DECL(rex,rexi);
3161 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3162 I32 *nums=(I32*)SvPVX(sv_dat);
3164 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3166 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3167 if ((I32)rex->lastparen >= nums[n] &&
3168 rex->offs[nums[n]].end != -1)
3177 /* free all slabs above current one - called during LEAVE_SCOPE */
3180 S_clear_backtrack_stack(pTHX_ void *p)
3182 regmatch_slab *s = PL_regmatch_slab->next;
3187 PL_regmatch_slab->next = NULL;
3189 regmatch_slab * const osl = s;
3196 /* returns -1 on failure, $+[0] on success */
3198 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3200 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3204 const bool utf8_target = PL_reg_match_utf8;
3205 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3206 REGEXP *rex_sv = reginfo->prog;
3207 regexp *rex = (struct regexp *)SvANY(rex_sv);
3208 RXi_GET_DECL(rex,rexi);
3210 /* the current state. This is a cached copy of PL_regmatch_state */
3212 /* cache heavy used fields of st in registers */
3215 U32 n = 0; /* general value; init to avoid compiler warning */
3216 I32 ln = 0; /* len or last; init to avoid compiler warning */
3217 char *locinput = startpos;
3218 char *pushinput; /* where to continue after a PUSH */
3219 I32 nextchr; /* is always set to UCHARAT(locinput) */
3221 bool result = 0; /* return value of S_regmatch */
3222 int depth = 0; /* depth of backtrack stack */
3223 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3224 const U32 max_nochange_depth =
3225 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3226 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3227 regmatch_state *yes_state = NULL; /* state to pop to on success of
3229 /* mark_state piggy backs on the yes_state logic so that when we unwind
3230 the stack on success we can update the mark_state as we go */
3231 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3232 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3233 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3235 bool no_final = 0; /* prevent failure from backtracking? */
3236 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3237 char *startpoint = locinput;
3238 SV *popmark = NULL; /* are we looking for a mark? */
3239 SV *sv_commit = NULL; /* last mark name seen in failure */
3240 SV *sv_yes_mark = NULL; /* last mark name we have seen
3241 during a successful match */
3242 U32 lastopen = 0; /* last open we saw */
3243 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3244 SV* const oreplsv = GvSV(PL_replgv);
3245 /* these three flags are set by various ops to signal information to
3246 * the very next op. They have a useful lifetime of exactly one loop
3247 * iteration, and are not preserved or restored by state pushes/pops
3249 bool sw = 0; /* the condition value in (?(cond)a|b) */
3250 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3251 int logical = 0; /* the following EVAL is:
3255 or the following IFMATCH/UNLESSM is:
3256 false: plain (?=foo)
3257 true: used as a condition: (?(?=foo))
3259 PAD* last_pad = NULL;
3261 I32 gimme = G_SCALAR;
3262 CV *caller_cv = NULL; /* who called us */
3263 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3264 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3267 GET_RE_DEBUG_FLAGS_DECL;
3270 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3271 multicall_oldcatch = 0;
3272 multicall_cv = NULL;
3274 PERL_UNUSED_VAR(multicall_cop);
3275 PERL_UNUSED_VAR(newsp);
3278 PERL_ARGS_ASSERT_REGMATCH;
3280 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3281 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3283 /* on first ever call to regmatch, allocate first slab */
3284 if (!PL_regmatch_slab) {
3285 Newx(PL_regmatch_slab, 1, regmatch_slab);
3286 PL_regmatch_slab->prev = NULL;
3287 PL_regmatch_slab->next = NULL;
3288 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3291 oldsave = PL_savestack_ix;
3292 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3293 SAVEVPTR(PL_regmatch_slab);
3294 SAVEVPTR(PL_regmatch_state);
3296 /* grab next free state slot */
3297 st = ++PL_regmatch_state;
3298 if (st > SLAB_LAST(PL_regmatch_slab))
3299 st = PL_regmatch_state = S_push_slab(aTHX);
3301 /* Note that nextchr is a byte even in UTF */
3302 nextchr = UCHARAT(locinput);
3304 while (scan != NULL) {
3307 SV * const prop = sv_newmortal();
3308 regnode *rnext=regnext(scan);
3309 DUMP_EXEC_POS( locinput, scan, utf8_target );
3310 regprop(rex, prop, scan);
3312 PerlIO_printf(Perl_debug_log,
3313 "%3"IVdf":%*s%s(%"IVdf")\n",
3314 (IV)(scan - rexi->program), depth*2, "",
3316 (PL_regkind[OP(scan)] == END || !rnext) ?
3317 0 : (IV)(rnext - rexi->program));
3320 next = scan + NEXT_OFF(scan);
3323 state_num = OP(scan);
3327 switch (state_num) {
3329 if (locinput == PL_bostr)
3331 /* reginfo->till = reginfo->bol; */
3336 if (locinput == PL_bostr ||
3337 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3343 if (locinput == PL_bostr)
3347 if (locinput == reginfo->ganch)
3352 /* update the startpoint */
3353 st->u.keeper.val = rex->offs[0].start;
3354 rex->offs[0].start = locinput - PL_bostr;
3355 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3357 case KEEPS_next_fail:
3358 /* rollback the start point change */
3359 rex->offs[0].start = st->u.keeper.val;
3365 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3370 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3372 if (PL_regeol - locinput > 1)
3376 if (PL_regeol != locinput)
3380 if (!nextchr && locinput >= PL_regeol)
3383 locinput += PL_utf8skip[nextchr];
3384 if (locinput > PL_regeol)
3386 nextchr = UCHARAT(locinput);
3389 nextchr = UCHARAT(++locinput);
3392 if (!nextchr && locinput >= PL_regeol)
3394 nextchr = UCHARAT(++locinput);
3397 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3400 locinput += PL_utf8skip[nextchr];
3401 if (locinput > PL_regeol)
3403 nextchr = UCHARAT(locinput);
3406 nextchr = UCHARAT(++locinput);
3410 #define ST st->u.trie
3412 /* In this case the charclass data is available inline so
3413 we can fail fast without a lot of extra overhead.
3415 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3417 PerlIO_printf(Perl_debug_log,
3418 "%*s %sfailed to match trie start class...%s\n",
3419 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3422 assert(0); /* NOTREACHED */
3426 /* the basic plan of execution of the trie is:
3427 * At the beginning, run though all the states, and
3428 * find the longest-matching word. Also remember the position
3429 * of the shortest matching word. For example, this pattern:
3432 * when matched against the string "abcde", will generate
3433 * accept states for all words except 3, with the longest
3434 * matching word being 4, and the shortest being 1 (with
3435 * the position being after char 1 of the string).
3437 * Then for each matching word, in word order (i.e. 1,2,4,5),
3438 * we run the remainder of the pattern; on each try setting
3439 * the current position to the character following the word,
3440 * returning to try the next word on failure.
3442 * We avoid having to build a list of words at runtime by
3443 * using a compile-time structure, wordinfo[].prev, which
3444 * gives, for each word, the previous accepting word (if any).
3445 * In the case above it would contain the mappings 1->2, 2->0,
3446 * 3->0, 4->5, 5->1. We can use this table to generate, from
3447 * the longest word (4 above), a list of all words, by
3448 * following the list of prev pointers; this gives us the
3449 * unordered list 4,5,1,2. Then given the current word we have
3450 * just tried, we can go through the list and find the
3451 * next-biggest word to try (so if we just failed on word 2,
3452 * the next in the list is 4).
3454 * Since at runtime we don't record the matching position in
3455 * the string for each word, we have to work that out for
3456 * each word we're about to process. The wordinfo table holds
3457 * the character length of each word; given that we recorded
3458 * at the start: the position of the shortest word and its
3459 * length in chars, we just need to move the pointer the
3460 * difference between the two char lengths. Depending on
3461 * Unicode status and folding, that's cheap or expensive.
3463 * This algorithm is optimised for the case where are only a
3464 * small number of accept states, i.e. 0,1, or maybe 2.
3465 * With lots of accepts states, and having to try all of them,
3466 * it becomes quadratic on number of accept states to find all
3471 /* what type of TRIE am I? (utf8 makes this contextual) */
3472 DECL_TRIE_TYPE(scan);
3474 /* what trie are we using right now */
3475 reg_trie_data * const trie
3476 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3477 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3478 U32 state = trie->startstate;
3480 if (trie->bitmap && !TRIE_BITMAP_TEST(trie,*locinput) ) {
3481 if (trie->states[ state ].wordnum) {
3483 PerlIO_printf(Perl_debug_log,
3484 "%*s %smatched empty string...%s\n",
3485 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3491 PerlIO_printf(Perl_debug_log,
3492 "%*s %sfailed to match trie start class...%s\n",
3493 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3500 U8 *uc = ( U8* )locinput;
3504 U8 *uscan = (U8*)NULL;
3505 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3506 U32 charcount = 0; /* how many input chars we have matched */
3507 U32 accepted = 0; /* have we seen any accepting states? */
3509 ST.jump = trie->jump;
3512 ST.longfold = FALSE; /* char longer if folded => it's harder */
3515 /* fully traverse the TRIE; note the position of the
3516 shortest accept state and the wordnum of the longest
3519 while ( state && uc <= (U8*)PL_regeol ) {
3520 U32 base = trie->states[ state ].trans.base;
3524 wordnum = trie->states[ state ].wordnum;
3526 if (wordnum) { /* it's an accept state */
3529 /* record first match position */
3531 ST.firstpos = (U8*)locinput;
3536 ST.firstchars = charcount;
3539 if (!ST.nextword || wordnum < ST.nextword)
3540 ST.nextword = wordnum;
3541 ST.topword = wordnum;
3544 DEBUG_TRIE_EXECUTE_r({
3545 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3546 PerlIO_printf( Perl_debug_log,
3547 "%*s %sState: %4"UVxf" Accepted: %c ",
3548 2+depth * 2, "", PL_colors[4],
3549 (UV)state, (accepted ? 'Y' : 'N'));
3552 /* read a char and goto next state */
3555 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3556 uscan, len, uvc, charid, foldlen,
3563 base + charid - 1 - trie->uniquecharcount)) >= 0)
3565 && ((U32)offset < trie->lasttrans)
3566 && trie->trans[offset].check == state)
3568 state = trie->trans[offset].next;
3579 DEBUG_TRIE_EXECUTE_r(
3580 PerlIO_printf( Perl_debug_log,
3581 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3582 charid, uvc, (UV)state, PL_colors[5] );
3588 /* calculate total number of accept states */
3593 w = trie->wordinfo[w].prev;
3596 ST.accepted = accepted;
3600 PerlIO_printf( Perl_debug_log,
3601 "%*s %sgot %"IVdf" possible matches%s\n",
3602 REPORT_CODE_OFF + depth * 2, "",
3603 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3605 goto trie_first_try; /* jump into the fail handler */
3607 assert(0); /* NOTREACHED */
3609 case TRIE_next_fail: /* we failed - try next alternative */
3613 REGCP_UNWIND(ST.cp);
3614 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3616 if (!--ST.accepted) {
3618 PerlIO_printf( Perl_debug_log,
3619 "%*s %sTRIE failed...%s\n",
3620 REPORT_CODE_OFF+depth*2, "",
3627 /* Find next-highest word to process. Note that this code
3628 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3631 U16 const nextword = ST.nextword;
3632 reg_trie_wordinfo * const wordinfo
3633 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3634 for (word=ST.topword; word; word=wordinfo[word].prev) {
3635 if (word > nextword && (!min || word < min))
3648 ST.lastparen = rex->lastparen;
3649 ST.lastcloseparen = rex->lastcloseparen;
3653 /* find start char of end of current word */
3655 U32 chars; /* how many chars to skip */
3656 reg_trie_data * const trie
3657 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3659 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3661 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3666 /* the hard option - fold each char in turn and find
3667 * its folded length (which may be different */
3668 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3676 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3684 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3689 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3705 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
3706 ? ST.jump[ST.nextword]
3710 PerlIO_printf( Perl_debug_log,
3711 "%*s %sTRIE matched word #%d, continuing%s\n",
3712 REPORT_CODE_OFF+depth*2, "",
3719 if (ST.accepted > 1 || has_cutgroup) {
3720 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
3721 assert(0); /* NOTREACHED */
3723 /* only one choice left - just continue */
3725 AV *const trie_words
3726 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3727 SV ** const tmp = av_fetch( trie_words,
3729 SV *sv= tmp ? sv_newmortal() : NULL;
3731 PerlIO_printf( Perl_debug_log,
3732 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3733 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3735 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3736 PL_colors[0], PL_colors[1],
3737 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3739 : "not compiled under -Dr",
3743 locinput = (char*)uc;
3744 nextchr = UCHARAT(locinput);
3745 continue; /* execute rest of RE */
3746 assert(0); /* NOTREACHED */
3751 char *s = STRING(scan);
3753 if (utf8_target != UTF_PATTERN) {
3754 /* The target and the pattern have differing utf8ness. */
3756 const char * const e = s + ln;
3759 /* The target is utf8, the pattern is not utf8.
3760 * Above-Latin1 code points can't match the pattern;
3761 * invariants match exactly, and the other Latin1 ones need
3762 * to be downgraded to a single byte in order to do the
3763 * comparison. (If we could be confident that the target
3764 * is not malformed, this could be refactored to have fewer
3765 * tests by just assuming that if the first bytes match, it
3766 * is an invariant, but there are tests in the test suite
3767 * dealing with (??{...}) which violate this) */
3771 if (UTF8_IS_ABOVE_LATIN1(* (U8*) l)) {
3774 if (UTF8_IS_INVARIANT(*(U8*)l)) {
3781 if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) {
3790 /* The target is not utf8, the pattern is utf8. */
3792 if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
3796 if (UTF8_IS_INVARIANT(*(U8*)s)) {
3803 if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) {
3812 nextchr = UCHARAT(locinput);
3815 /* The target and the pattern have the same utf8ness. */
3816 /* Inline the first character, for speed. */
3817 if (UCHARAT(s) != nextchr)
3819 if (PL_regeol - locinput < ln)
3821 if (ln > 1 && memNE(s, locinput, ln))
3824 nextchr = UCHARAT(locinput);
3829 const U8 * fold_array;
3831 U32 fold_utf8_flags;
3833 PL_reg_flags |= RF_tainted;
3834 folder = foldEQ_locale;
3835 fold_array = PL_fold_locale;
3836 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3840 case EXACTFU_TRICKYFOLD:
3842 folder = foldEQ_latin1;
3843 fold_array = PL_fold_latin1;
3844 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3848 folder = foldEQ_latin1;
3849 fold_array = PL_fold_latin1;
3850 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3855 fold_array = PL_fold;
3856 fold_utf8_flags = 0;
3862 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
3863 /* Either target or the pattern are utf8, or has the issue where
3864 * the fold lengths may differ. */
3865 const char * const l = locinput;
3866 char *e = PL_regeol;
3868 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3869 l, &e, 0, utf8_target, fold_utf8_flags))
3874 nextchr = UCHARAT(locinput);
3878 /* Neither the target nor the pattern are utf8 */
3879 if (UCHARAT(s) != nextchr &&
3880 UCHARAT(s) != fold_array[nextchr])
3884 if (PL_regeol - locinput < ln)
3886 if (ln > 1 && ! folder(s, locinput, ln))
3889 nextchr = UCHARAT(locinput);
3893 /* XXX Could improve efficiency by separating these all out using a
3894 * macro or in-line function. At that point regcomp.c would no longer
3895 * have to set the FLAGS fields of these */
3898 PL_reg_flags |= RF_tainted;
3906 /* was last char in word? */
3908 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3909 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3911 if (locinput == PL_bostr)
3914 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3916 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3918 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3919 ln = isALNUM_uni(ln);
3920 LOAD_UTF8_CHARCLASS_ALNUM();
3921 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3924 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3925 n = isALNUM_LC_utf8((U8*)locinput);
3930 /* Here the string isn't utf8, or is utf8 and only ascii
3931 * characters are to match \w. In the latter case looking at
3932 * the byte just prior to the current one may be just the final
3933 * byte of a multi-byte character. This is ok. There are two
3935 * 1) it is a single byte character, and then the test is doing
3936 * just what it's supposed to.
3937 * 2) it is a multi-byte character, in which case the final
3938 * byte is never mistakable for ASCII, and so the test
3939 * will say it is not a word character, which is the
3940 * correct answer. */
3941 ln = (locinput != PL_bostr) ?
3942 UCHARAT(locinput - 1) : '\n';
3943 switch (FLAGS(scan)) {
3944 case REGEX_UNICODE_CHARSET:
3945 ln = isWORDCHAR_L1(ln);
3946 n = isWORDCHAR_L1(nextchr);
3948 case REGEX_LOCALE_CHARSET:
3949 ln = isALNUM_LC(ln);
3950 n = isALNUM_LC(nextchr);
3952 case REGEX_DEPENDS_CHARSET:
3954 n = isALNUM(nextchr);
3956 case REGEX_ASCII_RESTRICTED_CHARSET:
3957 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3958 ln = isWORDCHAR_A(ln);
3959 n = isWORDCHAR_A(nextchr);
3962 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3966 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3968 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3973 if (utf8_target || state_num == ANYOFV) {
3974 STRLEN inclasslen = PL_regeol - locinput;
3975 if (locinput >= PL_regeol)
3978 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3980 locinput += inclasslen;
3981 nextchr = UCHARAT(locinput);
3986 nextchr = UCHARAT(locinput);
3987 if (!nextchr && locinput >= PL_regeol)
3989 if (!REGINCLASS(rex, scan, (U8*)locinput))
3991 nextchr = UCHARAT(++locinput);
3995 /* Special char classes - The defines start on line 129 or so */
3996 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3997 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3998 ALNUMU, NALNUMU, isWORDCHAR_L1,
3999 ALNUMA, NALNUMA, isWORDCHAR_A,
4002 CCC_TRY_U(SPACE, NSPACE, isSPACE,
4003 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
4004 SPACEU, NSPACEU, isSPACE_L1,
4005 SPACEA, NSPACEA, isSPACE_A,
4008 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
4009 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
4010 DIGITA, NDIGITA, isDIGIT_A,
4014 if (locinput >= PL_regeol || ! _generic_isCC_A(nextchr, FLAGS(scan))) {
4017 /* Matched a utf8-invariant, so don't have to worry about utf8 */
4018 nextchr = UCHARAT(++locinput);
4021 if (locinput >= PL_regeol || _generic_isCC_A(nextchr, FLAGS(scan))) {
4025 locinput += PL_utf8skip[nextchr];
4026 nextchr = UCHARAT(locinput);
4029 nextchr = UCHARAT(++locinput);
4033 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4034 a Unicode extended Grapheme Cluster */
4035 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4036 extended Grapheme Cluster is:
4039 | Prepend* Begin Extend*
4042 Begin is: ( Special_Begin | ! Control )
4043 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4044 Extend is: ( Grapheme_Extend | Spacing_Mark )
4045 Control is: [ GCB_Control CR LF ]
4046 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4048 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4051 Begin is ( Regular_Begin + Special Begin )
4053 It turns out that 98.4% of all Unicode code points match
4054 Regular_Begin. Doing it this way eliminates a table match in
4055 the previous implementation for almost all Unicode code points.
4057 There is a subtlety with Prepend* which showed up in testing.
4058 Note that the Begin, and only the Begin is required in:
4059 | Prepend* Begin Extend*
4060 Also, Begin contains '! Control'. A Prepend must be a
4061 '! Control', which means it must also be a Begin. What it
4062 comes down to is that if we match Prepend* and then find no
4063 suitable Begin afterwards, that if we backtrack the last
4064 Prepend, that one will be a suitable Begin.
4067 if (locinput >= PL_regeol)
4069 if (! utf8_target) {
4071 /* Match either CR LF or '.', as all the other possibilities
4073 locinput++; /* Match the . or CR */
4074 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4076 && locinput < PL_regeol
4077 && UCHARAT(locinput) == '\n') locinput++;
4081 /* Utf8: See if is ( CR LF ); already know that locinput <
4082 * PL_regeol, so locinput+1 is in bounds */
4083 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
4089 /* In case have to backtrack to beginning, then match '.' */
4090 char *starting = locinput;
4092 /* In case have to backtrack the last prepend */
4093 char *previous_prepend = 0;
4095 LOAD_UTF8_CHARCLASS_GCB();
4097 /* Match (prepend)* */
4098 while (locinput < PL_regeol
4099 && (len = is_GCB_Prepend_utf8(locinput)))
4101 previous_prepend = locinput;
4105 /* As noted above, if we matched a prepend character, but
4106 * the next thing won't match, back off the last prepend we
4107 * matched, as it is guaranteed to match the begin */
4108 if (previous_prepend
4109 && (locinput >= PL_regeol
4110 || (! swash_fetch(PL_utf8_X_regular_begin,
4111 (U8*)locinput, utf8_target)
4112 && ! is_GCB_SPECIAL_BEGIN_utf8(locinput)))
4115 locinput = previous_prepend;
4118 /* Note that here we know PL_regeol > locinput, as we
4119 * tested that upon input to this switch case, and if we
4120 * moved locinput forward, we tested the result just above
4121 * and it either passed, or we backed off so that it will
4123 if (swash_fetch(PL_utf8_X_regular_begin,
4124 (U8*)locinput, utf8_target)) {
4125 locinput += UTF8SKIP(locinput);
4127 else if (! is_GCB_SPECIAL_BEGIN_utf8(locinput)) {
4129 /* Here did not match the required 'Begin' in the
4130 * second term. So just match the very first
4131 * character, the '.' of the final term of the regex */
4132 locinput = starting + UTF8SKIP(starting);
4136 /* Here is a special begin. It can be composed of
4137 * several individual characters. One possibility is
4139 if ((len = is_GCB_RI_utf8(locinput))) {
4141 while (locinput < PL_regeol
4142 && (len = is_GCB_RI_utf8(locinput)))
4146 } else if ((len = is_GCB_T_utf8(locinput))) {
4147 /* Another possibility is T+ */
4149 while (locinput < PL_regeol
4150 && (len = is_GCB_T_utf8(locinput)))
4156 /* Here, neither RI+ nor T+; must be some other
4157 * Hangul. That means it is one of the others: L,
4158 * LV, LVT or V, and matches:
4159 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4162 while (locinput < PL_regeol
4163 && (len = is_GCB_L_utf8(locinput)))
4168 /* Here, have exhausted L*. If the next character
4169 * is not an LV, LVT nor V, it means we had to have
4170 * at least one L, so matches L+ in the original
4171 * equation, we have a complete hangul syllable.
4174 if (locinput < PL_regeol
4175 && is_GCB_LV_LVT_V_utf8(locinput))
4178 /* Otherwise keep going. Must be LV, LVT or V.
4180 if (is_utf8_X_LVT((U8*)locinput)) {
4181 locinput += UTF8SKIP(locinput);
4184 /* Must be V or LV. Take it, then match
4186 locinput += UTF8SKIP(locinput);
4187 while (locinput < PL_regeol
4188 && (len = is_GCB_V_utf8(locinput)))
4194 /* And any of LV, LVT, or V can be followed
4196 while (locinput < PL_regeol
4197 && (len = is_GCB_T_utf8(locinput)))
4205 /* Match any extender */
4206 while (locinput < PL_regeol
4207 && swash_fetch(PL_utf8_X_extend,
4208 (U8*)locinput, utf8_target))
4210 locinput += UTF8SKIP(locinput);
4214 if (locinput > PL_regeol) sayNO;
4216 nextchr = UCHARAT(locinput);
4220 { /* The capture buffer cases. The ones beginning with N for the
4221 named buffers just convert to the equivalent numbered and
4222 pretend they were called as the corresponding numbered buffer
4224 /* don't initialize these in the declaration, it makes C++
4229 const U8 *fold_array;
4232 PL_reg_flags |= RF_tainted;
4233 folder = foldEQ_locale;
4234 fold_array = PL_fold_locale;
4236 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4240 folder = foldEQ_latin1;
4241 fold_array = PL_fold_latin1;
4243 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4247 folder = foldEQ_latin1;
4248 fold_array = PL_fold_latin1;
4250 utf8_fold_flags = 0;
4255 fold_array = PL_fold;
4257 utf8_fold_flags = 0;
4264 utf8_fold_flags = 0;
4267 /* For the named back references, find the corresponding buffer
4269 n = reg_check_named_buff_matched(rex,scan);
4274 goto do_nref_ref_common;
4277 PL_reg_flags |= RF_tainted;
4278 folder = foldEQ_locale;
4279 fold_array = PL_fold_locale;
4280 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4284 folder = foldEQ_latin1;
4285 fold_array = PL_fold_latin1;
4286 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4290 folder = foldEQ_latin1;
4291 fold_array = PL_fold_latin1;
4292 utf8_fold_flags = 0;
4297 fold_array = PL_fold;
4298 utf8_fold_flags = 0;
4304 utf8_fold_flags = 0;
4308 n = ARG(scan); /* which paren pair */
4311 ln = rex->offs[n].start;
4312 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4313 if (rex->lastparen < n || ln == -1)
4314 sayNO; /* Do not match unless seen CLOSEn. */
4315 if (ln == rex->offs[n].end)
4319 if (type != REF /* REF can do byte comparison */
4320 && (utf8_target || type == REFFU))
4321 { /* XXX handle REFFL better */
4322 char * limit = PL_regeol;
4324 /* This call case insensitively compares the entire buffer
4325 * at s, with the current input starting at locinput, but
4326 * not going off the end given by PL_regeol, and returns in
4327 * limit upon success, how much of the current input was
4329 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4330 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4335 nextchr = UCHARAT(locinput);
4339 /* Not utf8: Inline the first character, for speed. */
4340 if (UCHARAT(s) != nextchr &&
4342 UCHARAT(s) != fold_array[nextchr]))
4344 ln = rex->offs[n].end - ln;
4345 if (locinput + ln > PL_regeol)
4347 if (ln > 1 && (type == REF
4348 ? memNE(s, locinput, ln)
4349 : ! folder(s, locinput, ln)))
4352 nextchr = UCHARAT(locinput);
4362 #define ST st->u.eval
4367 regexp_internal *rei;
4368 regnode *startpoint;
4371 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4372 if (cur_eval && cur_eval->locinput==locinput) {
4373 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4374 Perl_croak(aTHX_ "Infinite recursion in regex");
4375 if ( ++nochange_depth > max_nochange_depth )
4377 "Pattern subroutine nesting without pos change"
4378 " exceeded limit in regex");
4385 if (OP(scan)==GOSUB) {
4386 startpoint = scan + ARG2L(scan);
4387 ST.close_paren = ARG(scan);
4389 startpoint = rei->program+1;
4392 goto eval_recurse_doit;
4393 assert(0); /* NOTREACHED */
4394 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4395 if (cur_eval && cur_eval->locinput==locinput) {
4396 if ( ++nochange_depth > max_nochange_depth )
4397 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4402 /* execute the code in the {...} */
4406 OP * const oop = PL_op;
4407 COP * const ocurcop = PL_curcop;
4409 char *saved_regeol = PL_regeol;
4410 struct re_save_state saved_state;
4413 /* save *all* paren positions */
4415 REGCP_SET(runops_cp);
4417 /* To not corrupt the existing regex state while executing the
4418 * eval we would normally put it on the save stack, like with
4419 * save_re_context. However, re-evals have a weird scoping so we
4420 * can't just add ENTER/LEAVE here. With that, things like
4422 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4424 * would break, as they expect the localisation to be unwound
4425 * only when the re-engine backtracks through the bit that
4428 * What we do instead is just saving the state in a local c
4431 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4433 PL_reg_state.re_reparsing = FALSE;
4436 caller_cv = find_runcv(NULL);
4440 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4441 newcv = ((struct regexp *)SvANY(
4442 (REGEXP*)(rexi->data->data[n])
4445 nop = (OP*)rexi->data->data[n+1];
4447 else if (rexi->data->what[n] == 'l') { /* literal code */
4449 nop = (OP*)rexi->data->data[n];
4450 assert(CvDEPTH(newcv));
4453 /* literal with own CV */
4454 assert(rexi->data->what[n] == 'L');
4455 newcv = rex->qr_anoncv;
4456 nop = (OP*)rexi->data->data[n];
4459 /* normally if we're about to execute code from the same
4460 * CV that we used previously, we just use the existing
4461 * CX stack entry. However, its possible that in the
4462 * meantime we may have backtracked, popped from the save
4463 * stack, and undone the SAVECOMPPAD(s) associated with
4464 * PUSH_MULTICALL; in which case PL_comppad no longer
4465 * points to newcv's pad. */
4466 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4468 I32 depth = (newcv == caller_cv) ? 0 : 1;
4469 if (last_pushed_cv) {
4470 CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
4473 PUSH_MULTICALL_WITHDEPTH(newcv, depth);
4475 last_pushed_cv = newcv;
4477 last_pad = PL_comppad;
4479 /* the initial nextstate you would normally execute
4480 * at the start of an eval (which would cause error
4481 * messages to come from the eval), may be optimised
4482 * away from the execution path in the regex code blocks;
4483 * so manually set PL_curcop to it initially */
4485 OP *o = cUNOPx(nop)->op_first;
4486 assert(o->op_type == OP_NULL);
4487 if (o->op_targ == OP_SCOPE) {
4488 o = cUNOPo->op_first;
4491 assert(o->op_targ == OP_LEAVE);
4492 o = cUNOPo->op_first;
4493 assert(o->op_type == OP_ENTER);
4497 if (o->op_type != OP_STUB) {
4498 assert( o->op_type == OP_NEXTSTATE
4499 || o->op_type == OP_DBSTATE
4500 || (o->op_type == OP_NULL
4501 && ( o->op_targ == OP_NEXTSTATE
4502 || o->op_targ == OP_DBSTATE
4506 PL_curcop = (COP*)o;
4511 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4512 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4514 rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4517 SV *sv_mrk = get_sv("REGMARK", 1);
4518 sv_setsv(sv_mrk, sv_yes_mark);
4521 /* we don't use MULTICALL here as we want to call the
4522 * first op of the block of interest, rather than the
4523 * first op of the sub */
4526 CALLRUNOPS(aTHX); /* Scalar context. */
4529 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4535 /* before restoring everything, evaluate the returned
4536 * value, so that 'uninit' warnings don't use the wrong
4537 * PL_op or pad. Also need to process any magic vars
4538 * (e.g. $1) *before* parentheses are restored */
4543 if (logical == 0) /* (?{})/ */
4544 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
4545 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
4546 sw = cBOOL(SvTRUE(ret));
4549 else { /* /(??{}) */
4550 /* if its overloaded, let the regex compiler handle
4551 * it; otherwise extract regex, or stringify */
4552 if (!SvAMAGIC(ret)) {
4556 if (SvTYPE(sv) == SVt_REGEXP)
4557 re_sv = (REGEXP*) sv;
4558 else if (SvSMAGICAL(sv)) {
4559 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
4561 re_sv = (REGEXP *) mg->mg_obj;
4564 /* force any magic, undef warnings here */
4566 ret = sv_mortalcopy(ret);
4567 (void) SvPV_force_nolen(ret);
4573 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4575 /* *** Note that at this point we don't restore
4576 * PL_comppad, (or pop the CxSUB) on the assumption it may
4577 * be used again soon. This is safe as long as nothing
4578 * in the regexp code uses the pad ! */
4580 PL_curcop = ocurcop;
4581 PL_regeol = saved_regeol;
4582 S_regcp_restore(aTHX_ rex, runops_cp);
4588 /* only /(??{})/ from now on */
4591 /* extract RE object from returned value; compiling if
4595 re_sv = reg_temp_copy(NULL, re_sv);
4599 const I32 osize = PL_regsize;
4601 if (SvUTF8(ret) && IN_BYTES) {
4602 /* In use 'bytes': make a copy of the octet
4603 * sequence, but without the flag on */
4605 const char *const p = SvPV(ret, len);
4606 ret = newSVpvn_flags(p, len, SVs_TEMP);
4608 if (rex->intflags & PREGf_USE_RE_EVAL)
4609 pm_flags |= PMf_USE_RE_EVAL;
4611 /* if we got here, it should be an engine which
4612 * supports compiling code blocks and stuff */
4613 assert(rex->engine && rex->engine->op_comp);
4614 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
4615 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
4616 rex->engine, NULL, NULL,
4617 /* copy /msix etc to inner pattern */
4622 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4624 /* This isn't a first class regexp. Instead, it's
4625 caching a regexp onto an existing, Perl visible
4627 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
4630 /* safe to do now that any $1 etc has been
4631 * interpolated into the new pattern string and
4633 S_regcp_restore(aTHX_ rex, runops_cp);
4635 re = (struct regexp *)SvANY(re_sv);
4637 RXp_MATCH_COPIED_off(re);
4638 re->subbeg = rex->subbeg;
4639 re->sublen = rex->sublen;
4640 re->suboffset = rex->suboffset;
4641 re->subcoffset = rex->subcoffset;
4644 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4645 "Matching embedded");
4647 startpoint = rei->program + 1;
4648 ST.close_paren = 0; /* only used for GOSUB */
4650 eval_recurse_doit: /* Share code with GOSUB below this line */
4651 /* run the pattern returned from (??{...}) */
4652 ST.cp = regcppush(rex, 0); /* Save *all* the positions. */
4653 REGCP_SET(ST.lastcp);
4656 re->lastcloseparen = 0;
4660 /* XXXX This is too dramatic a measure... */
4663 ST.toggle_reg_flags = PL_reg_flags;
4665 PL_reg_flags |= RF_utf8;
4667 PL_reg_flags &= ~RF_utf8;
4668 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4670 ST.prev_rex = rex_sv;
4671 ST.prev_curlyx = cur_curlyx;
4673 SET_reg_curpm(rex_sv);
4678 ST.prev_eval = cur_eval;
4680 /* now continue from first node in postoned RE */
4681 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
4682 assert(0); /* NOTREACHED */
4685 case EVAL_AB: /* cleanup after a successful (??{A})B */
4686 /* note: this is called twice; first after popping B, then A */
4687 PL_reg_flags ^= ST.toggle_reg_flags;
4688 rex_sv = ST.prev_rex;
4689 SET_reg_curpm(rex_sv);
4690 rex = (struct regexp *)SvANY(rex_sv);
4691 rexi = RXi_GET(rex);
4693 cur_eval = ST.prev_eval;
4694 cur_curlyx = ST.prev_curlyx;
4696 /* XXXX This is too dramatic a measure... */
4698 if ( nochange_depth )
4703 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4704 /* note: this is called twice; first after popping B, then A */
4705 PL_reg_flags ^= ST.toggle_reg_flags;
4706 rex_sv = ST.prev_rex;
4707 SET_reg_curpm(rex_sv);
4708 rex = (struct regexp *)SvANY(rex_sv);
4709 rexi = RXi_GET(rex);
4711 REGCP_UNWIND(ST.lastcp);
4713 cur_eval = ST.prev_eval;
4714 cur_curlyx = ST.prev_curlyx;
4715 /* XXXX This is too dramatic a measure... */
4717 if ( nochange_depth )
4723 n = ARG(scan); /* which paren pair */
4724 rex->offs[n].start_tmp = locinput - PL_bostr;
4727 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
4728 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; regsize=%"UVuf"\n",
4732 (IV)rex->offs[n].start_tmp,
4738 /* XXX really need to log other places start/end are set too */
4739 #define CLOSE_CAPTURE \
4740 rex->offs[n].start = rex->offs[n].start_tmp; \
4741 rex->offs[n].end = locinput - PL_bostr; \
4742 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
4743 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
4745 PTR2UV(rex->offs), \
4747 (IV)rex->offs[n].start, \
4748 (IV)rex->offs[n].end \
4752 n = ARG(scan); /* which paren pair */
4754 /*if (n > PL_regsize)
4756 if (n > rex->lastparen)
4758 rex->lastcloseparen = n;
4759 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4767 cursor && OP(cursor)!=END;
4768 cursor=regnext(cursor))
4770 if ( OP(cursor)==CLOSE ){
4772 if ( n <= lastopen ) {
4774 /*if (n > PL_regsize)
4776 if (n > rex->lastparen)
4778 rex->lastcloseparen = n;
4779 if ( n == ARG(scan) || (cur_eval &&
4780 cur_eval->u.eval.close_paren == n))
4789 n = ARG(scan); /* which paren pair */
4790 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
4793 /* reg_check_named_buff_matched returns 0 for no match */
4794 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4798 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4804 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4806 next = NEXTOPER(NEXTOPER(scan));
4808 next = scan + ARG(scan);
4809 if (OP(next) == IFTHEN) /* Fake one. */
4810 next = NEXTOPER(NEXTOPER(next));
4814 logical = scan->flags;
4817 /*******************************************************************
4819 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4820 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4821 STAR/PLUS/CURLY/CURLYN are used instead.)
4823 A*B is compiled as <CURLYX><A><WHILEM><B>
4825 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4826 state, which contains the current count, initialised to -1. It also sets
4827 cur_curlyx to point to this state, with any previous value saved in the
4830 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4831 since the pattern may possibly match zero times (i.e. it's a while {} loop
4832 rather than a do {} while loop).
4834 Each entry to WHILEM represents a successful match of A. The count in the
4835 CURLYX block is incremented, another WHILEM state is pushed, and execution
4836 passes to A or B depending on greediness and the current count.
4838 For example, if matching against the string a1a2a3b (where the aN are
4839 substrings that match /A/), then the match progresses as follows: (the
4840 pushed states are interspersed with the bits of strings matched so far):
4843 <CURLYX cnt=0><WHILEM>
4844 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4845 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4846 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4847 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4849 (Contrast this with something like CURLYM, which maintains only a single
4853 a1 <CURLYM cnt=1> a2
4854 a1 a2 <CURLYM cnt=2> a3
4855 a1 a2 a3 <CURLYM cnt=3> b
4858 Each WHILEM state block marks a point to backtrack to upon partial failure
4859 of A or B, and also contains some minor state data related to that
4860 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4861 overall state, such as the count, and pointers to the A and B ops.
4863 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4864 must always point to the *current* CURLYX block, the rules are:
4866 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4867 and set cur_curlyx to point the new block.
4869 When popping the CURLYX block after a successful or unsuccessful match,
4870 restore the previous cur_curlyx.
4872 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4873 to the outer one saved in the CURLYX block.
4875 When popping the WHILEM block after a successful or unsuccessful B match,
4876 restore the previous cur_curlyx.
4878 Here's an example for the pattern (AI* BI)*BO
4879 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4882 curlyx backtrack stack
4883 ------ ---------------
4885 CO <CO prev=NULL> <WO>
4886 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4887 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4888 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4890 At this point the pattern succeeds, and we work back down the stack to
4891 clean up, restoring as we go:
4893 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4894 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4895 CO <CO prev=NULL> <WO>
4898 *******************************************************************/
4900 #define ST st->u.curlyx
4902 case CURLYX: /* start of /A*B/ (for complex A) */
4904 /* No need to save/restore up to this paren */
4905 I32 parenfloor = scan->flags;
4907 assert(next); /* keep Coverity happy */
4908 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4911 /* XXXX Probably it is better to teach regpush to support
4912 parenfloor > PL_regsize... */
4913 if (parenfloor > (I32)rex->lastparen)
4914 parenfloor = rex->lastparen; /* Pessimization... */
4916 ST.prev_curlyx= cur_curlyx;
4918 ST.cp = PL_savestack_ix;
4920 /* these fields contain the state of the current curly.
4921 * they are accessed by subsequent WHILEMs */
4922 ST.parenfloor = parenfloor;
4927 ST.count = -1; /* this will be updated by WHILEM */
4928 ST.lastloc = NULL; /* this will be updated by WHILEM */
4930 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
4931 assert(0); /* NOTREACHED */
4934 case CURLYX_end: /* just finished matching all of A*B */
4935 cur_curlyx = ST.prev_curlyx;
4937 assert(0); /* NOTREACHED */
4939 case CURLYX_end_fail: /* just failed to match all of A*B */
4941 cur_curlyx = ST.prev_curlyx;
4943 assert(0); /* NOTREACHED */
4947 #define ST st->u.whilem
4949 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4951 /* see the discussion above about CURLYX/WHILEM */
4953 int min = ARG1(cur_curlyx->u.curlyx.me);
4954 int max = ARG2(cur_curlyx->u.curlyx.me);
4955 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4957 assert(cur_curlyx); /* keep Coverity happy */
4958 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4959 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4960 ST.cache_offset = 0;
4964 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4965 "%*s whilem: matched %ld out of %d..%d\n",
4966 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4969 /* First just match a string of min A's. */
4972 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
4973 cur_curlyx->u.curlyx.lastloc = locinput;
4974 REGCP_SET(ST.lastcp);
4976 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
4977 assert(0); /* NOTREACHED */
4980 /* If degenerate A matches "", assume A done. */
4982 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4983 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4984 "%*s whilem: empty match detected, trying continuation...\n",
4985 REPORT_CODE_OFF+depth*2, "")
4987 goto do_whilem_B_max;
4990 /* super-linear cache processing */
4994 if (!PL_reg_maxiter) {
4995 /* start the countdown: Postpone detection until we
4996 * know the match is not *that* much linear. */
4997 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4998 /* possible overflow for long strings and many CURLYX's */
4999 if (PL_reg_maxiter < 0)
5000 PL_reg_maxiter = I32_MAX;
5001 PL_reg_leftiter = PL_reg_maxiter;
5004 if (PL_reg_leftiter-- == 0) {
5005 /* initialise cache */
5006 const I32 size = (PL_reg_maxiter + 7)/8;
5007 if (PL_reg_poscache) {
5008 if ((I32)PL_reg_poscache_size < size) {
5009 Renew(PL_reg_poscache, size, char);
5010 PL_reg_poscache_size = size;
5012 Zero(PL_reg_poscache, size, char);
5015 PL_reg_poscache_size = size;
5016 Newxz(PL_reg_poscache, size, char);
5018 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5019 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5020 PL_colors[4], PL_colors[5])
5024 if (PL_reg_leftiter < 0) {
5025 /* have we already failed at this position? */
5027 offset = (scan->flags & 0xf) - 1
5028 + (locinput - PL_bostr) * (scan->flags>>4);
5029 mask = 1 << (offset % 8);
5031 if (PL_reg_poscache[offset] & mask) {
5032 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5033 "%*s whilem: (cache) already tried at this position...\n",
5034 REPORT_CODE_OFF+depth*2, "")
5036 sayNO; /* cache records failure */
5038 ST.cache_offset = offset;
5039 ST.cache_mask = mask;
5043 /* Prefer B over A for minimal matching. */
5045 if (cur_curlyx->u.curlyx.minmod) {
5046 ST.save_curlyx = cur_curlyx;
5047 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5048 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor);
5049 REGCP_SET(ST.lastcp);
5050 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5052 assert(0); /* NOTREACHED */
5055 /* Prefer A over B for maximal matching. */
5057 if (n < max) { /* More greed allowed? */
5058 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5059 cur_curlyx->u.curlyx.lastloc = locinput;
5060 REGCP_SET(ST.lastcp);
5061 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5062 assert(0); /* NOTREACHED */
5064 goto do_whilem_B_max;
5066 assert(0); /* NOTREACHED */
5068 case WHILEM_B_min: /* just matched B in a minimal match */
5069 case WHILEM_B_max: /* just matched B in a maximal match */
5070 cur_curlyx = ST.save_curlyx;
5072 assert(0); /* NOTREACHED */
5074 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5075 cur_curlyx = ST.save_curlyx;
5076 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5077 cur_curlyx->u.curlyx.count--;
5079 assert(0); /* NOTREACHED */
5081 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5083 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5084 REGCP_UNWIND(ST.lastcp);
5086 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5087 cur_curlyx->u.curlyx.count--;
5089 assert(0); /* NOTREACHED */
5091 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5092 REGCP_UNWIND(ST.lastcp);
5093 regcppop(rex); /* Restore some previous $<digit>s? */
5094 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5095 "%*s whilem: failed, trying continuation...\n",
5096 REPORT_CODE_OFF+depth*2, "")
5099 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5100 && ckWARN(WARN_REGEXP)
5101 && !(PL_reg_flags & RF_warned))
5103 PL_reg_flags |= RF_warned;
5104 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5105 "Complex regular subexpression recursion limit (%d) "
5111 ST.save_curlyx = cur_curlyx;
5112 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5113 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5115 assert(0); /* NOTREACHED */
5117 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5118 cur_curlyx = ST.save_curlyx;
5119 REGCP_UNWIND(ST.lastcp);
5122 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5123 /* Maximum greed exceeded */
5124 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5125 && ckWARN(WARN_REGEXP)
5126 && !(PL_reg_flags & RF_warned))
5128 PL_reg_flags |= RF_warned;
5129 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5130 "Complex regular subexpression recursion "
5131 "limit (%d) exceeded",
5134 cur_curlyx->u.curlyx.count--;
5138 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5139 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5141 /* Try grabbing another A and see if it helps. */
5142 cur_curlyx->u.curlyx.lastloc = locinput;
5143 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5144 REGCP_SET(ST.lastcp);
5145 PUSH_STATE_GOTO(WHILEM_A_min,
5146 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5148 assert(0); /* NOTREACHED */
5151 #define ST st->u.branch
5153 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5154 next = scan + ARG(scan);
5157 scan = NEXTOPER(scan);
5160 case BRANCH: /* /(...|A|...)/ */
5161 scan = NEXTOPER(scan); /* scan now points to inner node */
5162 ST.lastparen = rex->lastparen;
5163 ST.lastcloseparen = rex->lastcloseparen;
5164 ST.next_branch = next;
5167 /* Now go into the branch */
5169 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5171 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5173 assert(0); /* NOTREACHED */
5175 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5176 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5177 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
5178 assert(0); /* NOTREACHED */
5179 case CUTGROUP_next_fail:
5182 if (st->u.mark.mark_name)
5183 sv_commit = st->u.mark.mark_name;
5185 assert(0); /* NOTREACHED */
5188 assert(0); /* NOTREACHED */
5189 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5194 REGCP_UNWIND(ST.cp);
5195 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5196 scan = ST.next_branch;
5197 /* no more branches? */
5198 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5200 PerlIO_printf( Perl_debug_log,
5201 "%*s %sBRANCH failed...%s\n",
5202 REPORT_CODE_OFF+depth*2, "",
5208 continue; /* execute next BRANCH[J] op */
5209 assert(0); /* NOTREACHED */
5216 #define ST st->u.curlym
5218 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5220 /* This is an optimisation of CURLYX that enables us to push
5221 * only a single backtracking state, no matter how many matches
5222 * there are in {m,n}. It relies on the pattern being constant
5223 * length, with no parens to influence future backrefs
5227 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5229 ST.lastparen = rex->lastparen;
5230 ST.lastcloseparen = rex->lastcloseparen;
5232 /* if paren positive, emulate an OPEN/CLOSE around A */
5234 U32 paren = ST.me->flags;
5235 if (paren > PL_regsize)
5237 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5245 ST.c1 = CHRTEST_UNINIT;
5248 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5251 curlym_do_A: /* execute the A in /A{m,n}B/ */
5252 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
5253 assert(0); /* NOTREACHED */
5255 case CURLYM_A: /* we've just matched an A */
5257 /* after first match, determine A's length: u.curlym.alen */
5258 if (ST.count == 1) {
5259 if (PL_reg_match_utf8) {
5260 char *s = st->locinput;
5261 while (s < locinput) {
5267 ST.alen = locinput - st->locinput;
5270 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5273 PerlIO_printf(Perl_debug_log,
5274 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5275 (int)(REPORT_CODE_OFF+(depth*2)), "",
5276 (IV) ST.count, (IV)ST.alen)
5279 if (cur_eval && cur_eval->u.eval.close_paren &&
5280 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5284 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5285 if ( max == REG_INFTY || ST.count < max )
5286 goto curlym_do_A; /* try to match another A */
5288 goto curlym_do_B; /* try to match B */
5290 case CURLYM_A_fail: /* just failed to match an A */
5291 REGCP_UNWIND(ST.cp);
5293 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5294 || (cur_eval && cur_eval->u.eval.close_paren &&
5295 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5298 curlym_do_B: /* execute the B in /A{m,n}B/ */
5299 if (ST.c1 == CHRTEST_UNINIT) {
5300 /* calculate c1 and c2 for possible match of 1st char
5301 * following curly */
5302 ST.c1 = ST.c2 = CHRTEST_VOID;
5303 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5304 regnode *text_node = ST.B;
5305 if (! HAS_TEXT(text_node))
5306 FIND_NEXT_IMPT(text_node);
5309 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5311 But the former is redundant in light of the latter.
5313 if this changes back then the macro for
5314 IS_TEXT and friends need to change.
5316 if (PL_regkind[OP(text_node)] == EXACT)
5319 ST.c1 = (U8)*STRING(text_node);
5320 switch (OP(text_node)) {
5321 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5324 case EXACTFU_TRICKYFOLD:
5325 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5326 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5327 default: ST.c2 = ST.c1;
5334 PerlIO_printf(Perl_debug_log,
5335 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5336 (int)(REPORT_CODE_OFF+(depth*2)),
5339 if (ST.c1 != CHRTEST_VOID
5341 && nextchr != ST.c2)
5343 /* simulate B failing */
5345 PerlIO_printf(Perl_debug_log,
5346 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5347 (int)(REPORT_CODE_OFF+(depth*2)),"",
5350 state_num = CURLYM_B_fail;
5351 goto reenter_switch;
5355 /* emulate CLOSE: mark current A as captured */
5356 I32 paren = ST.me->flags;
5358 rex->offs[paren].start
5359 = HOPc(locinput, -ST.alen) - PL_bostr;
5360 rex->offs[paren].end = locinput - PL_bostr;
5361 if ((U32)paren > rex->lastparen)
5362 rex->lastparen = paren;
5363 rex->lastcloseparen = paren;
5366 rex->offs[paren].end = -1;
5367 if (cur_eval && cur_eval->u.eval.close_paren &&
5368 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5377 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
5378 assert(0); /* NOTREACHED */
5380 case CURLYM_B_fail: /* just failed to match a B */
5381 REGCP_UNWIND(ST.cp);
5382 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5384 I32 max = ARG2(ST.me);
5385 if (max != REG_INFTY && ST.count == max)
5387 goto curlym_do_A; /* try to match a further A */
5389 /* backtrack one A */
5390 if (ST.count == ARG1(ST.me) /* min */)
5393 locinput = HOPc(locinput, -ST.alen);
5394 nextchr = UCHARAT(locinput);
5395 goto curlym_do_B; /* try to match B */
5398 #define ST st->u.curly
5400 #define CURLY_SETPAREN(paren, success) \
5403 rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5404 rex->offs[paren].end = locinput - PL_bostr; \
5405 if (paren > rex->lastparen) \
5406 rex->lastparen = paren; \
5407 rex->lastcloseparen = paren; \
5410 rex->offs[paren].end = -1; \
5411 rex->lastparen = ST.lastparen; \
5412 rex->lastcloseparen = ST.lastcloseparen; \
5416 case STAR: /* /A*B/ where A is width 1 */
5420 scan = NEXTOPER(scan);
5422 case PLUS: /* /A+B/ where A is width 1 */
5426 scan = NEXTOPER(scan);
5428 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5429 ST.paren = scan->flags; /* Which paren to set */
5430 ST.lastparen = rex->lastparen;
5431 ST.lastcloseparen = rex->lastcloseparen;
5432 if (ST.paren > PL_regsize)
5433 PL_regsize = ST.paren;
5434 ST.min = ARG1(scan); /* min to match */
5435 ST.max = ARG2(scan); /* max to match */
5436 if (cur_eval && cur_eval->u.eval.close_paren &&
5437 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5441 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5443 case CURLY: /* /A{m,n}B/ where A is width 1 */
5445 ST.min = ARG1(scan); /* min to match */
5446 ST.max = ARG2(scan); /* max to match */
5447 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5450 * Lookahead to avoid useless match attempts
5451 * when we know what character comes next.
5453 * Used to only do .*x and .*?x, but now it allows
5454 * for )'s, ('s and (?{ ... })'s to be in the way
5455 * of the quantifier and the EXACT-like node. -- japhy
5458 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5460 if (HAS_TEXT(next) || JUMPABLE(next)) {
5462 regnode *text_node = next;
5464 if (! HAS_TEXT(text_node))
5465 FIND_NEXT_IMPT(text_node);
5467 if (! HAS_TEXT(text_node))
5468 ST.c1 = ST.c2 = CHRTEST_VOID;
5470 if ( PL_regkind[OP(text_node)] != EXACT ) {
5471 ST.c1 = ST.c2 = CHRTEST_VOID;
5472 goto assume_ok_easy;
5475 s = (U8*)STRING(text_node);
5477 /* Currently we only get here when
5479 PL_rekind[OP(text_node)] == EXACT
5481 if this changes back then the macro for IS_TEXT and
5482 friends need to change. */
5485 switch (OP(text_node)) {
5486 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5489 case EXACTFU_TRICKYFOLD:
5490 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5491 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5492 default: ST.c2 = ST.c1; break;
5495 else { /* UTF_PATTERN */
5496 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5498 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5500 to_utf8_fold((U8*)s, tmpbuf, &ulen);
5501 ST.c1 = ST.c2 = utf8n_to_uvchr(tmpbuf, UTF8_MAXLEN, 0,
5505 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5512 ST.c1 = ST.c2 = CHRTEST_VOID;
5518 char *li = locinput;
5520 if (ST.min && regrepeat(rex, &li, ST.A, ST.min, depth) < ST.min)
5523 nextchr = UCHARAT(locinput);
5526 if (ST.c1 == CHRTEST_VOID)
5527 goto curly_try_B_min;
5529 ST.oldloc = locinput;
5531 /* set ST.maxpos to the furthest point along the
5532 * string that could possibly match */
5533 if (ST.max == REG_INFTY) {
5534 ST.maxpos = PL_regeol - 1;
5536 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5539 else if (utf8_target) {
5540 int m = ST.max - ST.min;
5541 for (ST.maxpos = locinput;
5542 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5543 ST.maxpos += UTF8SKIP(ST.maxpos);
5546 ST.maxpos = locinput + ST.max - ST.min;
5547 if (ST.maxpos >= PL_regeol)
5548 ST.maxpos = PL_regeol - 1;
5550 goto curly_try_B_min_known;
5554 /* avoid taking address of locinput, so it can remain
5556 char *li = locinput;
5557 ST.count = regrepeat(rex, &li, ST.A, ST.max, depth);
5558 if (ST.count < ST.min)
5561 nextchr = UCHARAT(locinput);
5562 if ((ST.count > ST.min)
5563 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5565 /* A{m,n} must come at the end of the string, there's
5566 * no point in backing off ... */
5568 /* ...except that $ and \Z can match before *and* after
5569 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5570 We may back off by one in this case. */
5571 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
5575 goto curly_try_B_max;
5577 assert(0); /* NOTREACHED */
5580 case CURLY_B_min_known_fail:
5581 /* failed to find B in a non-greedy match where c1,c2 valid */
5583 REGCP_UNWIND(ST.cp);
5585 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5587 /* Couldn't or didn't -- move forward. */
5588 ST.oldloc = locinput;
5590 locinput += UTF8SKIP(locinput);
5594 curly_try_B_min_known:
5595 /* find the next place where 'B' could work, then call B */
5599 n = (ST.oldloc == locinput) ? 0 : 1;
5600 if (ST.c1 == ST.c2) {
5602 /* set n to utf8_distance(oldloc, locinput) */
5603 while (locinput <= ST.maxpos &&
5604 utf8n_to_uvchr((U8*)locinput,
5605 UTF8_MAXBYTES, &len,
5606 uniflags) != (UV)ST.c1) {
5612 /* set n to utf8_distance(oldloc, locinput) */
5613 while (locinput <= ST.maxpos) {
5615 const UV c = utf8n_to_uvchr((U8*)locinput,
5616 UTF8_MAXBYTES, &len,
5618 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5626 if (ST.c1 == ST.c2) {
5627 while (locinput <= ST.maxpos &&
5628 UCHARAT(locinput) != ST.c1)
5632 while (locinput <= ST.maxpos
5633 && UCHARAT(locinput) != ST.c1
5634 && UCHARAT(locinput) != ST.c2)
5637 n = locinput - ST.oldloc;
5639 if (locinput > ST.maxpos)
5642 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
5643 * at b; check that everything between oldloc and
5644 * locinput matches */
5645 char *li = ST.oldloc;
5647 if (regrepeat(rex, &li, ST.A, n, depth) < n)
5649 assert(n == REG_INFTY || locinput == li);
5651 CURLY_SETPAREN(ST.paren, ST.count);
5652 if (cur_eval && cur_eval->u.eval.close_paren &&
5653 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5656 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
5658 assert(0); /* NOTREACHED */
5661 case CURLY_B_min_fail:
5662 /* failed to find B in a non-greedy match where c1,c2 invalid */
5664 REGCP_UNWIND(ST.cp);
5666 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5668 /* failed -- move forward one */
5670 char *li = locinput;
5671 if (!regrepeat(rex, &li, ST.A, 1, depth)) {
5678 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5679 ST.count > 0)) /* count overflow ? */
5682 CURLY_SETPAREN(ST.paren, ST.count);
5683 if (cur_eval && cur_eval->u.eval.close_paren &&
5684 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5687 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
5690 assert(0); /* NOTREACHED */
5694 /* a successful greedy match: now try to match B */
5695 if (cur_eval && cur_eval->u.eval.close_paren &&
5696 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5701 if (ST.c1 != CHRTEST_VOID)
5702 c = utf8_target ? utf8n_to_uvchr((U8*)locinput,
5703 UTF8_MAXBYTES, 0, uniflags)
5704 : (UV) UCHARAT(locinput);
5705 /* If it could work, try it. */
5706 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5707 CURLY_SETPAREN(ST.paren, ST.count);
5708 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
5709 assert(0); /* NOTREACHED */
5713 case CURLY_B_max_fail:
5714 /* failed to find B in a greedy match */
5716 REGCP_UNWIND(ST.cp);
5718 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5721 if (--ST.count < ST.min)
5723 locinput = HOPc(locinput, -1);
5724 goto curly_try_B_max;
5731 /* we've just finished A in /(??{A})B/; now continue with B */
5732 st->u.eval.toggle_reg_flags
5733 = cur_eval->u.eval.toggle_reg_flags;
5734 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5736 st->u.eval.prev_rex = rex_sv; /* inner */
5737 st->u.eval.cp = regcppush(rex, 0); /* Save *all* the positions. */
5738 rex_sv = cur_eval->u.eval.prev_rex;
5739 SET_reg_curpm(rex_sv);
5740 rex = (struct regexp *)SvANY(rex_sv);
5741 rexi = RXi_GET(rex);
5742 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5744 REGCP_SET(st->u.eval.lastcp);
5746 /* Restore parens of the outer rex without popping the
5748 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp);
5750 st->u.eval.prev_eval = cur_eval;
5751 cur_eval = cur_eval->u.eval.prev_eval;
5753 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5754 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5755 if ( nochange_depth )
5758 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
5759 locinput); /* match B */
5762 if (locinput < reginfo->till) {
5763 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5764 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5766 (long)(locinput - PL_reg_starttry),
5767 (long)(reginfo->till - PL_reg_starttry),
5770 sayNO_SILENT; /* Cannot match: too short. */
5772 sayYES; /* Success! */
5774 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5776 PerlIO_printf(Perl_debug_log,
5777 "%*s %ssubpattern success...%s\n",
5778 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5779 sayYES; /* Success! */
5782 #define ST st->u.ifmatch
5787 case SUSPEND: /* (?>A) */
5789 newstart = locinput;
5792 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5794 goto ifmatch_trivial_fail_test;
5796 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5798 ifmatch_trivial_fail_test:
5800 char * const s = HOPBACKc(locinput, scan->flags);
5805 sw = 1 - cBOOL(ST.wanted);
5809 next = scan + ARG(scan);
5817 newstart = locinput;
5821 ST.logical = logical;
5822 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5824 /* execute body of (?...A) */
5825 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
5826 assert(0); /* NOTREACHED */
5829 case IFMATCH_A_fail: /* body of (?...A) failed */
5830 ST.wanted = !ST.wanted;
5833 case IFMATCH_A: /* body of (?...A) succeeded */
5835 sw = cBOOL(ST.wanted);
5837 else if (!ST.wanted)
5840 if (OP(ST.me) != SUSPEND) {
5841 /* restore old position except for (?>...) */
5842 locinput = st->locinput;
5843 nextchr = UCHARAT(locinput);
5845 scan = ST.me + ARG(ST.me);
5848 continue; /* execute B */
5853 next = scan + ARG(scan);
5858 reginfo->cutpoint = PL_regeol;
5862 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5863 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
5864 assert(0); /* NOTREACHED */
5865 case COMMIT_next_fail:
5870 assert(0); /* NOTREACHED */
5872 #define ST st->u.mark
5874 ST.prev_mark = mark_state;
5875 ST.mark_name = sv_commit = sv_yes_mark
5876 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5878 ST.mark_loc = locinput;
5879 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
5880 assert(0); /* NOTREACHED */
5881 case MARKPOINT_next:
5882 mark_state = ST.prev_mark;
5884 assert(0); /* NOTREACHED */
5885 case MARKPOINT_next_fail:
5886 if (popmark && sv_eq(ST.mark_name,popmark))
5888 if (ST.mark_loc > startpoint)
5889 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5890 popmark = NULL; /* we found our mark */
5891 sv_commit = ST.mark_name;
5894 PerlIO_printf(Perl_debug_log,
5895 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5896 REPORT_CODE_OFF+depth*2, "",
5897 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5900 mark_state = ST.prev_mark;
5901 sv_yes_mark = mark_state ?
5902 mark_state->u.mark.mark_name : NULL;
5904 assert(0); /* NOTREACHED */
5907 /* (*SKIP) : if we fail we cut here*/
5908 ST.mark_name = NULL;
5909 ST.mark_loc = locinput;
5910 PUSH_STATE_GOTO(SKIP_next,next, locinput);
5912 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5913 otherwise do nothing. Meaning we need to scan
5915 regmatch_state *cur = mark_state;
5916 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5919 if ( sv_eq( cur->u.mark.mark_name,
5922 ST.mark_name = find;
5923 PUSH_STATE_GOTO( SKIP_next, next, locinput);
5925 cur = cur->u.mark.prev_mark;
5928 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5930 case SKIP_next_fail:
5932 /* (*CUT:NAME) - Set up to search for the name as we
5933 collapse the stack*/
5934 popmark = ST.mark_name;
5936 /* (*CUT) - No name, we cut here.*/
5937 if (ST.mark_loc > startpoint)
5938 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5939 /* but we set sv_commit to latest mark_name if there
5940 is one so they can test to see how things lead to this
5943 sv_commit=mark_state->u.mark.mark_name;
5947 assert(0); /* NOTREACHED */
5950 if ((n=is_LNBREAK(locinput,utf8_target))) {
5952 nextchr = UCHARAT(locinput);
5957 #define CASE_CLASS(nAmE) \
5959 if (locinput >= PL_regeol) \
5961 if ((n=is_##nAmE(locinput,utf8_target))) { \
5963 nextchr = UCHARAT(locinput); \
5968 if (locinput >= PL_regeol) \
5970 if ((n=is_##nAmE(locinput,utf8_target))) { \
5973 locinput += UTF8SKIP(locinput); \
5974 nextchr = UCHARAT(locinput); \
5979 CASE_CLASS(HORIZWS);
5983 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5984 PTR2UV(scan), OP(scan));
5985 Perl_croak(aTHX_ "regexp memory corruption");
5989 /* switch break jumps here */
5990 scan = next; /* prepare to execute the next op and ... */
5991 continue; /* ... jump back to the top, reusing st */
5992 assert(0); /* NOTREACHED */
5995 /* push a state that backtracks on success */
5996 st->u.yes.prev_yes_state = yes_state;
6000 /* push a new regex state, then continue at scan */
6002 regmatch_state *newst;
6005 regmatch_state *cur = st;
6006 regmatch_state *curyes = yes_state;
6008 regmatch_slab *slab = PL_regmatch_slab;
6009 for (;curd > -1;cur--,curd--) {
6010 if (cur < SLAB_FIRST(slab)) {
6012 cur = SLAB_LAST(slab);
6014 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6015 REPORT_CODE_OFF + 2 + depth * 2,"",
6016 curd, PL_reg_name[cur->resume_state],
6017 (curyes == cur) ? "yes" : ""
6020 curyes = cur->u.yes.prev_yes_state;
6023 DEBUG_STATE_pp("push")
6026 st->locinput = locinput;
6028 if (newst > SLAB_LAST(PL_regmatch_slab))
6029 newst = S_push_slab(aTHX);
6030 PL_regmatch_state = newst;
6032 locinput = pushinput;
6033 nextchr = UCHARAT(locinput);
6036 assert(0); /* NOTREACHED */
6041 * We get here only if there's trouble -- normally "case END" is
6042 * the terminating point.
6044 Perl_croak(aTHX_ "corrupted regexp pointers");
6050 /* we have successfully completed a subexpression, but we must now
6051 * pop to the state marked by yes_state and continue from there */
6052 assert(st != yes_state);
6054 while (st != yes_state) {
6056 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6057 PL_regmatch_slab = PL_regmatch_slab->prev;
6058 st = SLAB_LAST(PL_regmatch_slab);
6062 DEBUG_STATE_pp("pop (no final)");
6064 DEBUG_STATE_pp("pop (yes)");
6070 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6071 || yes_state > SLAB_LAST(PL_regmatch_slab))
6073 /* not in this slab, pop slab */
6074 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6075 PL_regmatch_slab = PL_regmatch_slab->prev;
6076 st = SLAB_LAST(PL_regmatch_slab);
6078 depth -= (st - yes_state);
6081 yes_state = st->u.yes.prev_yes_state;
6082 PL_regmatch_state = st;
6085 locinput= st->locinput;
6086 nextchr = UCHARAT(locinput);
6088 state_num = st->resume_state + no_final;
6089 goto reenter_switch;
6092 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6093 PL_colors[4], PL_colors[5]));
6095 if (PL_reg_state.re_state_eval_setup_done) {
6096 /* each successfully executed (?{...}) block does the equivalent of
6097 * local $^R = do {...}
6098 * When popping the save stack, all these locals would be undone;
6099 * bypass this by setting the outermost saved $^R to the latest
6101 if (oreplsv != GvSV(PL_replgv))
6102 sv_setsv(oreplsv, GvSV(PL_replgv));
6109 PerlIO_printf(Perl_debug_log,
6110 "%*s %sfailed...%s\n",
6111 REPORT_CODE_OFF+depth*2, "",
6112 PL_colors[4], PL_colors[5])
6124 /* there's a previous state to backtrack to */
6126 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6127 PL_regmatch_slab = PL_regmatch_slab->prev;
6128 st = SLAB_LAST(PL_regmatch_slab);
6130 PL_regmatch_state = st;
6131 locinput= st->locinput;
6132 nextchr = UCHARAT(locinput);
6134 DEBUG_STATE_pp("pop");
6136 if (yes_state == st)
6137 yes_state = st->u.yes.prev_yes_state;
6139 state_num = st->resume_state + 1; /* failure = success + 1 */
6140 goto reenter_switch;
6145 if (rex->intflags & PREGf_VERBARG_SEEN) {
6146 SV *sv_err = get_sv("REGERROR", 1);
6147 SV *sv_mrk = get_sv("REGMARK", 1);
6149 sv_commit = &PL_sv_no;
6151 sv_yes_mark = &PL_sv_yes;
6154 sv_commit = &PL_sv_yes;
6155 sv_yes_mark = &PL_sv_no;
6157 sv_setsv(sv_err, sv_commit);
6158 sv_setsv(sv_mrk, sv_yes_mark);
6162 if (last_pushed_cv) {
6165 PERL_UNUSED_VAR(SP);
6168 /* clean up; in particular, free all slabs above current one */
6169 LEAVE_SCOPE(oldsave);
6171 assert(!result || locinput - PL_bostr >= 0);
6172 return result ? locinput - PL_bostr : -1;
6176 - regrepeat - repeatedly match something simple, report how many
6178 * startposp - pointer a pointer to the start position. This is updated
6179 * to point to the byte following the highest successful
6181 * p - the regnode to be repeatedly matched against.
6182 * max - maximum number of characters to match.
6183 * depth - (for debugging) backtracking depth.
6186 S_regrepeat(pTHX_ const regexp *prog, char **startposp, const regnode *p, I32 max, int depth)
6191 char *loceol = PL_regeol;
6193 bool utf8_target = PL_reg_match_utf8;
6196 PERL_UNUSED_ARG(depth);
6199 PERL_ARGS_ASSERT_REGREPEAT;
6202 if (max == REG_INFTY)
6204 else if (max < loceol - scan)
6205 loceol = scan + max;
6210 while (scan < loceol && hardcount < max && *scan != '\n') {
6211 scan += UTF8SKIP(scan);
6215 while (scan < loceol && *scan != '\n')
6222 while (scan < loceol && hardcount < max) {
6223 scan += UTF8SKIP(scan);
6234 /* To get here, EXACTish nodes must have *byte* length == 1. That
6235 * means they match only characters in the string that can be expressed
6236 * as a single byte. For non-utf8 strings, that means a simple match.
6237 * For utf8 strings, the character matched must be an invariant, or
6238 * downgradable to a single byte. The pattern's utf8ness is
6239 * irrelevant, as since it's a single byte, it either isn't utf8, or if
6240 * it is, it's an invariant */
6243 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6245 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6246 while (scan < loceol && UCHARAT(scan) == c) {
6252 /* Here, the string is utf8, and the pattern char is different
6253 * in utf8 than not, so can't compare them directly. Outside the
6254 * loop, find the two utf8 bytes that represent c, and then
6255 * look for those in sequence in the utf8 string */
6256 U8 high = UTF8_TWO_BYTE_HI(c);
6257 U8 low = UTF8_TWO_BYTE_LO(c);
6260 while (hardcount < max
6261 && scan + 1 < loceol
6262 && UCHARAT(scan) == high
6263 && UCHARAT(scan + 1) == low)
6271 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6275 PL_reg_flags |= RF_tainted;
6276 utf8_flags = FOLDEQ_UTF8_LOCALE;
6284 case EXACTFU_TRICKYFOLD:
6286 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6288 /* The comments for the EXACT case above apply as well to these fold
6293 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6295 if (utf8_target || OP(p) == EXACTFU_SS) { /* Use full Unicode fold matching */
6296 char *tmpeol = loceol;
6297 while (hardcount < max
6298 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6299 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6306 /* XXX Note that the above handles properly the German sharp s in
6307 * the pattern matching ss in the string. But it doesn't handle
6308 * properly cases where the string contains say 'LIGATURE ff' and
6309 * the pattern is 'f+'. This would require, say, a new function or
6310 * revised interface to foldEQ_utf8(), in which the maximum number
6311 * of characters to match could be passed and it would return how
6312 * many actually did. This is just one of many cases where
6313 * multi-char folds don't work properly, and so the fix is being
6319 /* Here, the string isn't utf8 and c is a single byte; and either
6320 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6321 * doesn't affect c. Can just do simple comparisons for exact or
6324 case EXACTF: folded = PL_fold[c]; break;
6326 case EXACTFU_TRICKYFOLD:
6327 case EXACTFU: folded = PL_fold_latin1[c]; break;
6328 case EXACTFL: folded = PL_fold_locale[c]; break;
6329 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6331 while (scan < loceol &&
6332 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6340 if (utf8_target || OP(p) == ANYOFV) {
6343 inclasslen = loceol - scan;
6344 while (hardcount < max
6345 && ((inclasslen = loceol - scan) > 0)
6346 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6352 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6360 LOAD_UTF8_CHARCLASS_ALNUM();
6361 while (hardcount < max && scan < loceol &&
6362 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6364 scan += UTF8SKIP(scan);
6368 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6376 while (scan < loceol && isALNUM((U8) *scan)) {
6381 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6386 PL_reg_flags |= RF_tainted;
6389 while (hardcount < max && scan < loceol &&
6390 isALNUM_LC_utf8((U8*)scan)) {
6391 scan += UTF8SKIP(scan);
6395 while (scan < loceol && isALNUM_LC(*scan))
6405 LOAD_UTF8_CHARCLASS_ALNUM();
6406 while (hardcount < max && scan < loceol &&
6407 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6409 scan += UTF8SKIP(scan);
6413 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6420 goto utf8_Nwordchar;
6421 while (scan < loceol && ! isALNUM((U8) *scan)) {
6427 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
6433 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6434 scan += UTF8SKIP(scan);
6438 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6445 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6446 scan += UTF8SKIP(scan);
6450 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6456 PL_reg_flags |= RF_tainted;
6459 while (hardcount < max && scan < loceol &&
6460 !isALNUM_LC_utf8((U8*)scan)) {
6461 scan += UTF8SKIP(scan);
6465 while (scan < loceol && !isALNUM_LC(*scan))
6475 LOAD_UTF8_CHARCLASS_SPACE();
6476 while (hardcount < max && scan < loceol &&
6478 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6480 scan += UTF8SKIP(scan);
6486 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6495 while (scan < loceol && isSPACE((U8) *scan)) {
6500 while (scan < loceol && isSPACE_A((U8) *scan)) {
6505 PL_reg_flags |= RF_tainted;
6508 while (hardcount < max && scan < loceol &&
6509 isSPACE_LC_utf8((U8*)scan)) {
6510 scan += UTF8SKIP(scan);
6514 while (scan < loceol && isSPACE_LC(*scan))
6524 LOAD_UTF8_CHARCLASS_SPACE();
6525 while (hardcount < max && scan < loceol &&
6527 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6529 scan += UTF8SKIP(scan);
6535 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6544 while (scan < loceol && ! isSPACE((U8) *scan)) {
6550 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6551 scan += UTF8SKIP(scan);
6555 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6561 PL_reg_flags |= RF_tainted;
6564 while (hardcount < max && scan < loceol &&
6565 !isSPACE_LC_utf8((U8*)scan)) {
6566 scan += UTF8SKIP(scan);
6570 while (scan < loceol && !isSPACE_LC(*scan))
6577 LOAD_UTF8_CHARCLASS_DIGIT();
6578 while (hardcount < max && scan < loceol &&
6579 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6580 scan += UTF8SKIP(scan);
6584 while (scan < loceol && isDIGIT(*scan))
6589 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6594 PL_reg_flags |= RF_tainted;
6597 while (hardcount < max && scan < loceol &&
6598 isDIGIT_LC_utf8((U8*)scan)) {
6599 scan += UTF8SKIP(scan);
6603 while (scan < loceol && isDIGIT_LC(*scan))
6610 LOAD_UTF8_CHARCLASS_DIGIT();
6611 while (hardcount < max && scan < loceol &&
6612 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6613 scan += UTF8SKIP(scan);
6617 while (scan < loceol && !isDIGIT(*scan))
6623 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6624 scan += UTF8SKIP(scan);
6628 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6634 PL_reg_flags |= RF_tainted;
6637 while (hardcount < max && scan < loceol &&
6638 !isDIGIT_LC_utf8((U8*)scan)) {
6639 scan += UTF8SKIP(scan);
6643 while (scan < loceol && !isDIGIT_LC(*scan))
6650 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6656 LNBREAK can match two latin chars, which is ok,
6657 because we have a null terminated string, but we
6658 have to use hardcount in this situation
6660 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6669 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6674 while (scan < loceol && is_HORIZWS_latin1(scan))
6681 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6682 scan += UTF8SKIP(scan);
6686 while (scan < loceol && !is_HORIZWS_latin1(scan))
6694 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6699 while (scan < loceol && is_VERTWS_latin1(scan))
6707 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6708 scan += UTF8SKIP(scan);
6712 while (scan < loceol && !is_VERTWS_latin1(scan))
6718 default: /* Called on something of 0 width. */
6719 break; /* So match right here or not at all. */
6725 c = scan - *startposp;
6729 GET_RE_DEBUG_FLAGS_DECL;
6731 SV * const prop = sv_newmortal();
6732 regprop(prog, prop, p);
6733 PerlIO_printf(Perl_debug_log,
6734 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6735 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6743 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6745 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
6746 create a copy so that changes the caller makes won't change the shared one
6749 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6751 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6752 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp, altsvp));
6757 S_core_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6759 /* Returns the swash for the input 'node' in the regex 'prog'.
6760 * If <doinit> is true, will attempt to create the swash if not already
6762 * If <listsvp> is non-null, will return the swash initialization string in
6764 * If <altsvp> is non-null, will return the alternates to the regular swash
6766 * Tied intimately to how regcomp.c sets up the data structure */
6774 RXi_GET_DECL(prog,progi);
6775 const struct reg_data * const data = prog ? progi->data : NULL;
6777 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
6779 assert(ANYOF_NONBITMAP(node));
6781 if (data && data->count) {
6782 const U32 n = ARG(node);
6784 if (data->what[n] == 's') {
6785 SV * const rv = MUTABLE_SV(data->data[n]);
6786 AV * const av = MUTABLE_AV(SvRV(rv));
6787 SV **const ary = AvARRAY(av);
6788 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
6790 si = *ary; /* ary[0] = the string to initialize the swash with */
6792 /* Elements 3 and 4 are either both present or both absent. [3] is
6793 * any inversion list generated at compile time; [4] indicates if
6794 * that inversion list has any user-defined properties in it. */
6795 if (av_len(av) >= 3) {
6798 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
6805 /* Element [1] is reserved for the set-up swash. If already there,
6806 * return it; if not, create it and store it there */
6807 if (SvROK(ary[1])) {
6810 else if (si && doinit) {
6812 sw = _core_swash_init("utf8", /* the utf8 package */
6816 0, /* not from tr/// */
6819 (void)av_store(av, 1, sw);
6822 /* Element [2] is for any multi-char folds. Note that is a
6823 * fundamentally flawed design, because can't backtrack and try
6824 * again. See [perl #89774] */
6825 if (SvTYPE(ary[2]) == SVt_PVAV) {
6832 SV* matches_string = newSVpvn("", 0);
6834 /* Use the swash, if any, which has to have incorporated into it all
6836 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
6837 && (si && si != &PL_sv_undef))
6840 /* If no swash, use the input initialization string, if available */
6841 sv_catsv(matches_string, si);
6844 /* Add the inversion list to whatever we have. This may have come from
6845 * the swash, or from an input parameter */
6847 sv_catsv(matches_string, _invlist_contents(invlist));
6849 *listsvp = matches_string;
6859 - reginclass - determine if a character falls into a character class
6861 n is the ANYOF regnode
6862 p is the target string
6863 lenp is pointer to the maximum number of bytes of how far to go in p
6864 (This is assumed wthout checking to always be at least the current
6866 utf8_target tells whether p is in UTF-8.
6868 Returns true if matched; false otherwise. If lenp is not NULL, on return
6869 from a successful match, the value it points to will be updated to how many
6870 bytes in p were matched. If there was no match, the value is undefined,
6871 possibly changed from the input.
6873 Note that this can be a synthetic start class, a combination of various
6874 nodes, so things you think might be mutually exclusive, such as locale,
6875 aren't. It can match both locale and non-locale
6880 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6883 const char flags = ANYOF_FLAGS(n);
6889 PERL_ARGS_ASSERT_REGINCLASS;
6891 /* If c is not already the code point, get it */
6892 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6893 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6894 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6895 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6896 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6897 * UTF8_ALLOW_FFFF */
6898 if (c_len == (STRLEN)-1)
6899 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6905 /* Use passed in max length, or one character if none passed in or less
6906 * than one character. And assume will match just one character. This is
6907 * overwritten later if matched more. */
6909 maxlen = (*lenp > c_len) ? *lenp : c_len;
6917 /* If this character is potentially in the bitmap, check it */
6919 if (ANYOF_BITMAP_TEST(n, c))
6921 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6928 else if (flags & ANYOF_LOCALE) {
6929 PL_reg_flags |= RF_tainted;
6931 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6932 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6936 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6937 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6938 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6939 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6940 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6941 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6942 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6943 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6944 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6945 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6946 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6947 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII_LC(c)) ||
6948 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII_LC(c)) ||
6949 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6950 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6951 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6952 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6953 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6954 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6955 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6956 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6957 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6958 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6959 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6960 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6961 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6962 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6963 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6964 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6965 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK_LC(c)) ||
6966 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK_LC(c))
6967 ) /* How's that for a conditional? */
6974 /* If the bitmap didn't (or couldn't) match, and something outside the
6975 * bitmap could match, try that. Locale nodes specifiy completely the
6976 * behavior of code points in the bit map (otherwise, a utf8 target would
6977 * cause them to be treated as Unicode and not locale), except in
6978 * the very unlikely event when this node is a synthetic start class, which
6979 * could be a combination of locale and non-locale nodes. So allow locale
6980 * to match for the synthetic start class, which will give a false
6981 * positive that will be resolved when the match is done again as not part
6982 * of the synthetic start class */
6984 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6985 match = TRUE; /* Everything above 255 matches */
6987 else if (ANYOF_NONBITMAP(n)
6988 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6991 || (! (flags & ANYOF_LOCALE))
6992 || (flags & ANYOF_IS_SYNTHETIC)))))
6995 SV * const sw = core_regclass_swash(prog, n, TRUE, 0, (SV**)&av);
7003 /* Not utf8. Convert as much of the string as available up
7004 * to the limit of how far the (single) character in the
7005 * pattern can possibly match (no need to go further). If
7006 * the node is a straight ANYOF or not folding, it can't
7007 * match more than one. Otherwise, It can match up to how
7008 * far a single char can fold to. Since not utf8, each
7009 * character is a single byte, so the max it can be in
7010 * bytes is the same as the max it can be in characters */
7011 STRLEN len = (OP(n) == ANYOF
7012 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
7014 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
7016 : UTF8_MAX_FOLD_CHAR_EXPAND;
7017 utf8_p = bytes_to_utf8(p, &len);
7020 if (swash_fetch(sw, utf8_p, TRUE))
7022 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
7024 /* Here, we need to test if the fold of the target string
7025 * matches. The non-multi char folds have all been moved to
7026 * the compilation phase, and the multi-char folds have
7027 * been stored by regcomp into 'av'; we linearly check to
7028 * see if any match the target string (folded). We know
7029 * that the originals were each one character, but we don't
7030 * currently know how many characters/bytes each folded to,
7031 * except we do know that there are small limits imposed by
7032 * Unicode. XXX A performance enhancement would be to have
7033 * regcomp.c store the max number of chars/bytes that are
7034 * in an av entry, as, say the 0th element. Even better
7035 * would be to have a hash of the few characters that can
7036 * start a multi-char fold to the max number of chars of
7039 * If there is a match, we will need to advance (if lenp is
7040 * specified) the match pointer in the target string. But
7041 * what we are comparing here isn't that string directly,
7042 * but its fold, whose length may differ from the original.
7043 * As we go along in constructing the fold, therefore, we
7044 * create a map so that we know how many bytes in the
7045 * source to advance given that we have matched a certain
7046 * number of bytes in the fold. This map is stored in
7047 * 'map_fold_len_back'. Let n mean the number of bytes in
7048 * the fold of the first character that we are folding.
7049 * Then map_fold_len_back[n] is set to the number of bytes
7050 * in that first character. Similarly let m be the
7051 * corresponding number for the second character to be
7052 * folded. Then map_fold_len_back[n+m] is set to the
7053 * number of bytes occupied by the first two source
7054 * characters. ... */
7055 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
7056 U8 folded[UTF8_MAXBYTES_CASE+1];
7057 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
7058 STRLEN total_foldlen = 0; /* num bytes in fold of all
7061 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
7063 /* Here, only need to fold the first char of the target
7064 * string. It the source wasn't utf8, is 1 byte long */
7065 to_utf8_fold(utf8_p, folded, &foldlen);
7066 total_foldlen = foldlen;
7067 map_fold_len_back[foldlen] = (utf8_target)
7073 /* Here, need to fold more than the first char. Do so
7074 * up to the limits */
7075 U8* source_ptr = utf8_p; /* The source for the fold
7078 U8* folded_ptr = folded;
7079 U8* e = utf8_p + maxlen; /* Can't go beyond last
7080 available byte in the
7084 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
7088 /* Fold the next character */
7089 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
7090 STRLEN this_char_foldlen;
7091 to_utf8_fold(source_ptr,
7093 &this_char_foldlen);
7095 /* Bail if it would exceed the byte limit for
7096 * folding a single char. */
7097 if (this_char_foldlen + folded_ptr - folded >
7103 /* Add the fold of this character */
7104 Copy(this_char_folded,
7108 source_ptr += UTF8SKIP(source_ptr);
7109 folded_ptr += this_char_foldlen;
7110 total_foldlen = folded_ptr - folded;
7112 /* Create map from the number of bytes in the fold
7113 * back to the number of bytes in the source. If
7114 * the source isn't utf8, the byte count is just
7115 * the number of characters so far */
7116 map_fold_len_back[total_foldlen]
7118 ? source_ptr - utf8_p
7125 /* Do the linear search to see if the fold is in the list
7126 * of multi-char folds. */
7129 for (i = 0; i <= av_len(av); i++) {
7130 SV* const sv = *av_fetch(av, i, FALSE);
7132 const char * const s = SvPV_const(sv, len);
7134 if (len <= total_foldlen
7135 && memEQ(s, (char*)folded, len)
7137 /* If 0, means matched a partial char. See
7139 && map_fold_len_back[len])
7142 /* Advance the target string ptr to account for
7143 * this fold, but have to translate from the
7144 * folded length to the corresponding source
7147 *lenp = map_fold_len_back[len];
7156 /* If we allocated a string above, free it */
7157 if (! utf8_target) Safefree(utf8_p);
7161 if (UNICODE_IS_SUPER(c)
7162 && (flags & ANYOF_WARN_SUPER)
7163 && ckWARN_d(WARN_NON_UNICODE))
7165 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7166 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7170 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7171 return cBOOL(flags & ANYOF_INVERT) ^ match;
7175 S_reghop3(U8 *s, I32 off, const U8* lim)
7177 /* return the position 'off' UTF-8 characters away from 's', forward if
7178 * 'off' >= 0, backwards if negative. But don't go outside of position
7179 * 'lim', which better be < s if off < 0 */
7183 PERL_ARGS_ASSERT_REGHOP3;
7186 while (off-- && s < lim) {
7187 /* XXX could check well-formedness here */
7192 while (off++ && s > lim) {
7194 if (UTF8_IS_CONTINUED(*s)) {
7195 while (s > lim && UTF8_IS_CONTINUATION(*s))
7198 /* XXX could check well-formedness here */
7205 /* there are a bunch of places where we use two reghop3's that should
7206 be replaced with this routine. but since thats not done yet
7207 we ifdef it out - dmq
7210 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7214 PERL_ARGS_ASSERT_REGHOP4;
7217 while (off-- && s < rlim) {
7218 /* XXX could check well-formedness here */
7223 while (off++ && s > llim) {
7225 if (UTF8_IS_CONTINUED(*s)) {
7226 while (s > llim && UTF8_IS_CONTINUATION(*s))
7229 /* XXX could check well-formedness here */
7237 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7241 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7244 while (off-- && s < lim) {
7245 /* XXX could check well-formedness here */
7252 while (off++ && s > lim) {
7254 if (UTF8_IS_CONTINUED(*s)) {
7255 while (s > lim && UTF8_IS_CONTINUATION(*s))
7258 /* XXX could check well-formedness here */
7267 restore_pos(pTHX_ void *arg)
7270 regexp * const rex = (regexp *)arg;
7271 if (PL_reg_state.re_state_eval_setup_done) {
7272 if (PL_reg_oldsaved) {
7273 rex->subbeg = PL_reg_oldsaved;
7274 rex->sublen = PL_reg_oldsavedlen;
7275 rex->suboffset = PL_reg_oldsavedoffset;
7276 rex->subcoffset = PL_reg_oldsavedcoffset;
7277 #ifdef PERL_OLD_COPY_ON_WRITE
7278 rex->saved_copy = PL_nrs;
7280 RXp_MATCH_COPIED_on(rex);
7282 PL_reg_magic->mg_len = PL_reg_oldpos;
7283 PL_reg_state.re_state_eval_setup_done = FALSE;
7284 PL_curpm = PL_reg_oldcurpm;
7289 S_to_utf8_substr(pTHX_ register regexp *prog)
7293 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7296 if (prog->substrs->data[i].substr
7297 && !prog->substrs->data[i].utf8_substr) {
7298 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7299 prog->substrs->data[i].utf8_substr = sv;
7300 sv_utf8_upgrade(sv);
7301 if (SvVALID(prog->substrs->data[i].substr)) {
7302 if (SvTAIL(prog->substrs->data[i].substr)) {
7303 /* Trim the trailing \n that fbm_compile added last
7305 SvCUR_set(sv, SvCUR(sv) - 1);
7306 /* Whilst this makes the SV technically "invalid" (as its
7307 buffer is no longer followed by "\0") when fbm_compile()
7308 adds the "\n" back, a "\0" is restored. */
7309 fbm_compile(sv, FBMcf_TAIL);
7313 if (prog->substrs->data[i].substr == prog->check_substr)
7314 prog->check_utf8 = sv;
7320 S_to_byte_substr(pTHX_ register regexp *prog)
7325 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7328 if (prog->substrs->data[i].utf8_substr
7329 && !prog->substrs->data[i].substr) {
7330 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7331 if (sv_utf8_downgrade(sv, TRUE)) {
7332 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7333 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7334 /* Trim the trailing \n that fbm_compile added last
7336 SvCUR_set(sv, SvCUR(sv) - 1);
7337 fbm_compile(sv, FBMcf_TAIL);
7345 prog->substrs->data[i].substr = sv;
7346 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7347 prog->check_substr = sv;
7352 /* These constants are for finding GCB=LV and GCB=LVT. These are for the
7353 * pre-composed Hangul syllables, which are all in a contiguous block and
7354 * arranged there in such a way so as to facilitate alorithmic determination of
7355 * their characteristics. As such, they don't need a swash, but can be
7356 * determined by simple arithmetic. Almost all are GCB=LVT, but every 28th one
7358 #define SBASE 0xAC00 /* Start of block */
7359 #define SCount 11172 /* Length of block */
7362 #if 0 /* This routine is not currently used */
7363 PERL_STATIC_INLINE bool
7364 S_is_utf8_X_LV(pTHX_ const U8 *p)
7366 /* Unlike most other similarly named routines here, this does not create a
7367 * swash, so swash_fetch() cannot be used on PL_utf8_X_LV. */
7371 UV cp = valid_utf8_to_uvchr(p, NULL);
7373 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
7375 /* The earliest Unicode releases did not have these precomposed Hangul
7376 * syllables. Set to point to undef in that case, so will return false on
7378 if (! PL_utf8_X_LV) { /* Set up if this is the first time called */
7379 PL_utf8_X_LV = swash_init("utf8", "_X_GCB_LV", &PL_sv_undef, 1, 0);
7380 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LV)) == 0) {
7381 SvREFCNT_dec(PL_utf8_X_LV);
7382 PL_utf8_X_LV = &PL_sv_undef;
7386 return (PL_utf8_X_LV != &PL_sv_undef
7387 && cp >= SBASE && cp < SBASE + SCount
7388 && (cp - SBASE) % TCount == 0); /* Only every TCount one is LV */
7392 PERL_STATIC_INLINE bool
7393 S_is_utf8_X_LVT(pTHX_ const U8 *p)
7395 /* Unlike most other similarly named routines here, this does not create a
7396 * swash, so swash_fetch() cannot be used on PL_utf8_X_LVT. */
7400 UV cp = valid_utf8_to_uvchr(p, NULL);
7402 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
7404 /* The earliest Unicode releases did not have these precomposed Hangul
7405 * syllables. Set to point to undef in that case, so will return false on
7407 if (! PL_utf8_X_LVT) { /* Set up if this is the first time called */
7408 PL_utf8_X_LVT = swash_init("utf8", "_X_GCB_LVT", &PL_sv_undef, 1, 0);
7409 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LVT)) == 0) {
7410 SvREFCNT_dec(PL_utf8_X_LVT);
7411 PL_utf8_X_LVT = &PL_sv_undef;
7415 return (PL_utf8_X_LVT != &PL_sv_undef
7416 && cp >= SBASE && cp < SBASE + SCount
7417 && (cp - SBASE) % TCount != 0); /* All but every TCount one is LV */
7422 * c-indentation-style: bsd
7424 * indent-tabs-mode: nil
7427 * ex: set ts=8 sts=4 sw=4 et: