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
78 #ifdef PERL_IN_XSUB_RE
84 #define RF_tainted 1 /* tainted information used? e.g. locale */
85 #define RF_warned 2 /* warned about big count? */
87 #define RF_utf8 8 /* Pattern contains multibyte chars? */
89 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
95 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
96 * call if there are no complications: i.e., if everything matchable is
97 * straight forward in the bitmap */
98 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
99 : ANYOF_BITMAP_TEST(p,*(c)))
105 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
106 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
108 #define HOPc(pos,off) \
109 (char *)(PL_reg_match_utf8 \
110 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
112 #define HOPBACKc(pos, off) \
113 (char*)(PL_reg_match_utf8\
114 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
115 : (pos - off >= PL_bostr) \
119 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
120 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
122 /* these are unrolled below in the CCC_TRY_XXX defined */
124 /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just
125 * skip the check on EBCDIC platforms */
126 # define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class)
128 # define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
129 if (!CAT2(PL_utf8_,class)) { \
131 ENTER; save_re_context(); \
132 ok=CAT2(is_utf8_,class)((const U8*)str); \
133 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
136 /* Doesn't do an assert to verify that is correct */
137 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
138 if (!CAT2(PL_utf8_,class)) { \
139 bool throw_away PERL_UNUSED_DECL; \
140 ENTER; save_re_context(); \
141 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
144 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
145 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
146 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
148 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
149 LOAD_UTF8_CHARCLASS(X_begin, " "); \
150 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
151 /* These are utf8 constants, and not utf-ebcdic constants, so the \
152 * assert should likely and hopefully fail on an EBCDIC machine */ \
153 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
155 /* No asserts are done for these, in case called on an early \
156 * Unicode version in which they map to nothing */ \
157 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
158 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
159 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
160 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
161 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
162 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
163 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
165 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
167 /* The actual code for CCC_TRY, which uses several variables from the routine
168 * it's callable from. It is designed to be the bulk of a case statement.
169 * FUNC is the macro or function to call on non-utf8 targets that indicate if
170 * nextchr matches the class.
171 * UTF8_TEST is the whole test string to use for utf8 targets
172 * LOAD is what to use to test, and if not present to load in the swash for the
174 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
176 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
177 * utf8 and a variant, load the swash if necessary and test using the utf8
178 * test. Advance to the next character if test is ok, otherwise fail; If not
179 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
180 * fails, or advance to the next character */
182 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
183 if (locinput >= PL_regeol) { \
186 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
187 LOAD_UTF8_CHARCLASS(CLASS, STR); \
188 if (POS_OR_NEG (UTF8_TEST)) { \
191 locinput += PL_utf8skip[nextchr]; \
192 nextchr = UCHARAT(locinput); \
195 if (POS_OR_NEG (FUNC(nextchr))) { \
198 nextchr = UCHARAT(++locinput); \
201 /* Handle the non-locale cases for a character class and its complement. It
202 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
203 * This is because that code fails when the test succeeds, so we want to have
204 * the test fail so that the code succeeds. The swash is stored in a
205 * predictable PL_ place */
206 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
209 _CCC_TRY_CODE( !, FUNC, \
210 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
211 (U8*)locinput, TRUE)), \
214 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
215 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
216 (U8*)locinput, TRUE)), \
219 /* Generate the case statements for both locale and non-locale character
220 * classes in regmatch for classes that don't have special unicode semantics.
221 * Locales don't use an immediate swash, but an intermediary special locale
222 * function that is called on the pointer to the current place in the input
223 * string. That function will resolve to needing the same swash. One might
224 * think that because we don't know what the locale will match, we shouldn't
225 * check with the swash loading function that it loaded properly; ie, that we
226 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
227 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
229 #define CCC_TRY(NAME, NNAME, FUNC, \
230 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
231 NAMEA, NNAMEA, FUNCA, \
234 PL_reg_flags |= RF_tainted; \
235 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
237 PL_reg_flags |= RF_tainted; \
238 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
241 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
244 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
245 nextchr = UCHARAT(++locinput); \
248 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
252 locinput += PL_utf8skip[nextchr]; \
253 nextchr = UCHARAT(locinput); \
256 nextchr = UCHARAT(++locinput); \
259 /* Generate the non-locale cases */ \
260 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
262 /* This is like CCC_TRY, but has an extra set of parameters for generating case
263 * statements to handle separate Unicode semantics nodes */
264 #define CCC_TRY_U(NAME, NNAME, FUNC, \
265 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
266 NAMEU, NNAMEU, FUNCU, \
267 NAMEA, NNAMEA, FUNCA, \
269 CCC_TRY(NAME, NNAME, FUNC, \
270 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
271 NAMEA, NNAMEA, FUNCA, \
273 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
275 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
277 /* for use after a quantifier and before an EXACT-like node -- japhy */
278 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
280 * NOTE that *nothing* that affects backtracking should be in here, specifically
281 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
282 * node that is in between two EXACT like nodes when ascertaining what the required
283 * "follow" character is. This should probably be moved to regex compile time
284 * although it may be done at run time beause of the REF possibility - more
285 * investigation required. -- demerphq
287 #define JUMPABLE(rn) ( \
289 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
291 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
292 OP(rn) == PLUS || OP(rn) == MINMOD || \
294 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
296 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
298 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
301 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
302 we don't need this definition. */
303 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
304 #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 )
305 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
308 /* ... so we use this as its faster. */
309 #define IS_TEXT(rn) ( OP(rn)==EXACT )
310 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
311 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
312 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
317 Search for mandatory following text node; for lookahead, the text must
318 follow but for lookbehind (rn->flags != 0) we skip to the next step.
320 #define FIND_NEXT_IMPT(rn) STMT_START { \
321 while (JUMPABLE(rn)) { \
322 const OPCODE type = OP(rn); \
323 if (type == SUSPEND || PL_regkind[type] == CURLY) \
324 rn = NEXTOPER(NEXTOPER(rn)); \
325 else if (type == PLUS) \
327 else if (type == IFMATCH) \
328 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
329 else rn += NEXT_OFF(rn); \
334 static void restore_pos(pTHX_ void *arg);
336 #define REGCP_PAREN_ELEMS 3
337 #define REGCP_OTHER_ELEMS 3
338 #define REGCP_FRAME_ELEMS 1
339 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
340 * are needed for the regexp context stack bookkeeping. */
343 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor)
346 const int retval = PL_savestack_ix;
347 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
348 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
349 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
351 GET_RE_DEBUG_FLAGS_DECL;
353 PERL_ARGS_ASSERT_REGCPPUSH;
355 if (paren_elems_to_push < 0)
356 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
357 paren_elems_to_push);
359 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
360 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
361 " out of range (%lu-%ld)",
362 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
364 SSGROW(total_elems + REGCP_FRAME_ELEMS);
367 if ((int)PL_regsize > (int)parenfloor)
368 PerlIO_printf(Perl_debug_log,
369 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
374 for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
375 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
376 SSPUSHINT(rex->offs[p].end);
377 SSPUSHINT(rex->offs[p].start);
378 SSPUSHINT(rex->offs[p].start_tmp);
379 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
380 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
382 (IV)rex->offs[p].start,
383 (IV)rex->offs[p].start_tmp,
387 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
388 SSPUSHINT(PL_regsize);
389 SSPUSHINT(rex->lastparen);
390 SSPUSHINT(rex->lastcloseparen);
391 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
396 /* These are needed since we do not localize EVAL nodes: */
397 #define REGCP_SET(cp) \
399 PerlIO_printf(Perl_debug_log, \
400 " Setting an EVAL scope, savestack=%"IVdf"\n", \
401 (IV)PL_savestack_ix)); \
404 #define REGCP_UNWIND(cp) \
406 if (cp != PL_savestack_ix) \
407 PerlIO_printf(Perl_debug_log, \
408 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
409 (IV)(cp), (IV)PL_savestack_ix)); \
412 #define UNWIND_PAREN(lp, lcp) \
413 for (n = rex->lastparen; n > lp; n--) \
414 rex->offs[n].end = -1; \
415 rex->lastparen = n; \
416 rex->lastcloseparen = lcp;
420 S_regcppop(pTHX_ regexp *rex)
425 GET_RE_DEBUG_FLAGS_DECL;
427 PERL_ARGS_ASSERT_REGCPPOP;
429 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
431 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
432 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
433 rex->lastcloseparen = SSPOPINT;
434 rex->lastparen = SSPOPINT;
435 PL_regsize = SSPOPINT;
437 i -= REGCP_OTHER_ELEMS;
438 /* Now restore the parentheses context. */
440 if (i || rex->lastparen + 1 <= rex->nparens)
441 PerlIO_printf(Perl_debug_log,
442 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
448 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
450 rex->offs[paren].start_tmp = SSPOPINT;
451 rex->offs[paren].start = SSPOPINT;
453 if (paren <= rex->lastparen)
454 rex->offs[paren].end = tmps;
455 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
456 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
458 (IV)rex->offs[paren].start,
459 (IV)rex->offs[paren].start_tmp,
460 (IV)rex->offs[paren].end,
461 (paren > rex->lastparen ? "(skipped)" : ""));
466 /* It would seem that the similar code in regtry()
467 * already takes care of this, and in fact it is in
468 * a better location to since this code can #if 0-ed out
469 * but the code in regtry() is needed or otherwise tests
470 * requiring null fields (pat.t#187 and split.t#{13,14}
471 * (as of patchlevel 7877) will fail. Then again,
472 * this code seems to be necessary or otherwise
473 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
474 * --jhi updated by dapm */
475 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
477 rex->offs[i].start = -1;
478 rex->offs[i].end = -1;
479 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
480 " \\%"UVuf": %s ..-1 undeffing\n",
482 (i > PL_regsize) ? "-1" : " "
488 /* restore the parens and associated vars at savestack position ix,
489 * but without popping the stack */
492 S_regcp_restore(pTHX_ regexp *rex, I32 ix)
494 I32 tmpix = PL_savestack_ix;
495 PL_savestack_ix = ix;
497 PL_savestack_ix = tmpix;
500 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
503 * pregexec and friends
506 #ifndef PERL_IN_XSUB_RE
508 - pregexec - match a regexp against a string
511 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
512 char *strbeg, I32 minend, SV *screamer, U32 nosave)
513 /* strend: pointer to null at end of string */
514 /* strbeg: real beginning of string */
515 /* minend: end of match must be >=minend after stringarg. */
516 /* nosave: For optimizations. */
518 PERL_ARGS_ASSERT_PREGEXEC;
521 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
522 nosave ? 0 : REXEC_COPY_STR);
527 * Need to implement the following flags for reg_anch:
529 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
531 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
532 * INTUIT_AUTORITATIVE_ML
533 * INTUIT_ONCE_NOML - Intuit can match in one location only.
536 * Another flag for this function: SECOND_TIME (so that float substrs
537 * with giant delta may be not rechecked).
540 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
542 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
543 Otherwise, only SvCUR(sv) is used to get strbeg. */
545 /* XXXX We assume that strpos is strbeg unless sv. */
547 /* XXXX Some places assume that there is a fixed substring.
548 An update may be needed if optimizer marks as "INTUITable"
549 RExen without fixed substrings. Similarly, it is assumed that
550 lengths of all the strings are no more than minlen, thus they
551 cannot come from lookahead.
552 (Or minlen should take into account lookahead.)
553 NOTE: Some of this comment is not correct. minlen does now take account
554 of lookahead/behind. Further research is required. -- demerphq
558 /* A failure to find a constant substring means that there is no need to make
559 an expensive call to REx engine, thus we celebrate a failure. Similarly,
560 finding a substring too deep into the string means that less calls to
561 regtry() should be needed.
563 REx compiler's optimizer found 4 possible hints:
564 a) Anchored substring;
566 c) Whether we are anchored (beginning-of-line or \G);
567 d) First node (of those at offset 0) which may distinguish positions;
568 We use a)b)d) and multiline-part of c), and try to find a position in the
569 string which does not contradict any of them.
572 /* Most of decisions we do here should have been done at compile time.
573 The nodes of the REx which we used for the search should have been
574 deleted from the finite automaton. */
577 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
578 char *strend, const U32 flags, re_scream_pos_data *data)
581 struct regexp *const prog = (struct regexp *)SvANY(rx);
582 register I32 start_shift = 0;
583 /* Should be nonnegative! */
584 register I32 end_shift = 0;
589 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
591 register char *other_last = NULL; /* other substr checked before this */
592 char *check_at = NULL; /* check substr found at this pos */
593 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
594 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
595 RXi_GET_DECL(prog,progi);
597 const char * const i_strpos = strpos;
599 GET_RE_DEBUG_FLAGS_DECL;
601 PERL_ARGS_ASSERT_RE_INTUIT_START;
602 PERL_UNUSED_ARG(flags);
603 PERL_UNUSED_ARG(data);
605 RX_MATCH_UTF8_set(rx,utf8_target);
608 PL_reg_flags |= RF_utf8;
611 debug_start_match(rx, utf8_target, strpos, strend,
612 sv ? "Guessing start of match in sv for"
613 : "Guessing start of match in string for");
616 /* CHR_DIST() would be more correct here but it makes things slow. */
617 if (prog->minlen > strend - strpos) {
618 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
619 "String too short... [re_intuit_start]\n"));
623 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
626 if (!prog->check_utf8 && prog->check_substr)
627 to_utf8_substr(prog);
628 check = prog->check_utf8;
630 if (!prog->check_substr && prog->check_utf8)
631 to_byte_substr(prog);
632 check = prog->check_substr;
634 if (check == &PL_sv_undef) {
635 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
636 "Non-utf8 string cannot match utf8 check string\n"));
639 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
640 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
641 || ( (prog->extflags & RXf_ANCH_BOL)
642 && !multiline ) ); /* Check after \n? */
645 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
646 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
647 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
649 && (strpos != strbeg)) {
650 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
653 if (prog->check_offset_min == prog->check_offset_max &&
654 !(prog->extflags & RXf_CANY_SEEN)) {
655 /* Substring at constant offset from beg-of-str... */
658 s = HOP3c(strpos, prog->check_offset_min, strend);
661 slen = SvCUR(check); /* >= 1 */
663 if ( strend - s > slen || strend - s < slen - 1
664 || (strend - s == slen && strend[-1] != '\n')) {
665 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
668 /* Now should match s[0..slen-2] */
670 if (slen && (*SvPVX_const(check) != *s
672 && memNE(SvPVX_const(check), s, slen)))) {
674 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
678 else if (*SvPVX_const(check) != *s
679 || ((slen = SvCUR(check)) > 1
680 && memNE(SvPVX_const(check), s, slen)))
683 goto success_at_start;
686 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
688 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
689 end_shift = prog->check_end_shift;
692 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
693 - (SvTAIL(check) != 0);
694 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
696 if (end_shift < eshift)
700 else { /* Can match at random position */
703 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
704 end_shift = prog->check_end_shift;
706 /* end shift should be non negative here */
709 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
711 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
712 (IV)end_shift, RX_PRECOMP(prog));
716 /* Find a possible match in the region s..strend by looking for
717 the "check" substring in the region corrected by start/end_shift. */
720 I32 srch_start_shift = start_shift;
721 I32 srch_end_shift = end_shift;
724 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
725 srch_end_shift -= ((strbeg - s) - srch_start_shift);
726 srch_start_shift = strbeg - s;
728 DEBUG_OPTIMISE_MORE_r({
729 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
730 (IV)prog->check_offset_min,
731 (IV)srch_start_shift,
733 (IV)prog->check_end_shift);
736 if (prog->extflags & RXf_CANY_SEEN) {
737 start_point= (U8*)(s + srch_start_shift);
738 end_point= (U8*)(strend - srch_end_shift);
740 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
741 end_point= HOP3(strend, -srch_end_shift, strbeg);
743 DEBUG_OPTIMISE_MORE_r({
744 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
745 (int)(end_point - start_point),
746 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
750 s = fbm_instr( start_point, end_point,
751 check, multiline ? FBMrf_MULTILINE : 0);
753 /* Update the count-of-usability, remove useless subpatterns,
757 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
758 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
759 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
760 (s ? "Found" : "Did not find"),
761 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
762 ? "anchored" : "floating"),
765 (s ? " at offset " : "...\n") );
770 /* Finish the diagnostic message */
771 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
773 /* XXX dmq: first branch is for positive lookbehind...
774 Our check string is offset from the beginning of the pattern.
775 So we need to do any stclass tests offset forward from that
784 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
785 Start with the other substr.
786 XXXX no SCREAM optimization yet - and a very coarse implementation
787 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
788 *always* match. Probably should be marked during compile...
789 Probably it is right to do no SCREAM here...
792 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
793 : (prog->float_substr && prog->anchored_substr))
795 /* Take into account the "other" substring. */
796 /* XXXX May be hopelessly wrong for UTF... */
799 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
802 char * const last = HOP3c(s, -start_shift, strbeg);
804 char * const saved_s = s;
807 t = s - prog->check_offset_max;
808 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
810 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
815 t = HOP3c(t, prog->anchored_offset, strend);
816 if (t < other_last) /* These positions already checked */
818 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
821 /* XXXX It is not documented what units *_offsets are in.
822 We assume bytes, but this is clearly wrong.
823 Meaning this code needs to be carefully reviewed for errors.
827 /* On end-of-str: see comment below. */
828 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
829 if (must == &PL_sv_undef) {
831 DEBUG_r(must = prog->anchored_utf8); /* for debug */
836 HOP3(HOP3(last1, prog->anchored_offset, strend)
837 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
839 multiline ? FBMrf_MULTILINE : 0
842 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
843 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
844 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
845 (s ? "Found" : "Contradicts"),
846 quoted, RE_SV_TAIL(must));
851 if (last1 >= last2) {
852 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
853 ", giving up...\n"));
856 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
857 ", trying floating at offset %ld...\n",
858 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
859 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
860 s = HOP3c(last, 1, strend);
864 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
865 (long)(s - i_strpos)));
866 t = HOP3c(s, -prog->anchored_offset, strbeg);
867 other_last = HOP3c(s, 1, strend);
875 else { /* Take into account the floating substring. */
877 char * const saved_s = s;
880 t = HOP3c(s, -start_shift, strbeg);
882 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
883 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
884 last = HOP3c(t, prog->float_max_offset, strend);
885 s = HOP3c(t, prog->float_min_offset, strend);
888 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
889 must = utf8_target ? prog->float_utf8 : prog->float_substr;
890 /* fbm_instr() takes into account exact value of end-of-str
891 if the check is SvTAIL(ed). Since false positives are OK,
892 and end-of-str is not later than strend we are OK. */
893 if (must == &PL_sv_undef) {
895 DEBUG_r(must = prog->float_utf8); /* for debug message */
898 s = fbm_instr((unsigned char*)s,
899 (unsigned char*)last + SvCUR(must)
901 must, multiline ? FBMrf_MULTILINE : 0);
903 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
904 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
905 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
906 (s ? "Found" : "Contradicts"),
907 quoted, RE_SV_TAIL(must));
911 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
912 ", giving up...\n"));
915 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
916 ", trying anchored starting at offset %ld...\n",
917 (long)(saved_s + 1 - i_strpos)));
919 s = HOP3c(t, 1, strend);
923 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
924 (long)(s - i_strpos)));
925 other_last = s; /* Fix this later. --Hugo */
935 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
937 DEBUG_OPTIMISE_MORE_r(
938 PerlIO_printf(Perl_debug_log,
939 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
940 (IV)prog->check_offset_min,
941 (IV)prog->check_offset_max,
949 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
951 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
954 /* Fixed substring is found far enough so that the match
955 cannot start at strpos. */
957 if (ml_anch && t[-1] != '\n') {
958 /* Eventually fbm_*() should handle this, but often
959 anchored_offset is not 0, so this check will not be wasted. */
960 /* XXXX In the code below we prefer to look for "^" even in
961 presence of anchored substrings. And we search even
962 beyond the found float position. These pessimizations
963 are historical artefacts only. */
965 while (t < strend - prog->minlen) {
967 if (t < check_at - prog->check_offset_min) {
968 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
969 /* Since we moved from the found position,
970 we definitely contradict the found anchored
971 substr. Due to the above check we do not
972 contradict "check" substr.
973 Thus we can arrive here only if check substr
974 is float. Redo checking for "other"=="fixed".
977 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
978 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
979 goto do_other_anchored;
981 /* We don't contradict the found floating substring. */
982 /* XXXX Why not check for STCLASS? */
984 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
985 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
988 /* Position contradicts check-string */
989 /* XXXX probably better to look for check-string
990 than for "\n", so one should lower the limit for t? */
991 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
992 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
993 other_last = strpos = s = t + 1;
998 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
999 PL_colors[0], PL_colors[1]));
1003 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1004 PL_colors[0], PL_colors[1]));
1008 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1011 /* The found string does not prohibit matching at strpos,
1012 - no optimization of calling REx engine can be performed,
1013 unless it was an MBOL and we are not after MBOL,
1014 or a future STCLASS check will fail this. */
1016 /* Even in this situation we may use MBOL flag if strpos is offset
1017 wrt the start of the string. */
1018 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1019 && (strpos != strbeg) && strpos[-1] != '\n'
1020 /* May be due to an implicit anchor of m{.*foo} */
1021 && !(prog->intflags & PREGf_IMPLICIT))
1026 DEBUG_EXECUTE_r( if (ml_anch)
1027 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1028 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1031 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1033 prog->check_utf8 /* Could be deleted already */
1034 && --BmUSEFUL(prog->check_utf8) < 0
1035 && (prog->check_utf8 == prog->float_utf8)
1037 prog->check_substr /* Could be deleted already */
1038 && --BmUSEFUL(prog->check_substr) < 0
1039 && (prog->check_substr == prog->float_substr)
1042 /* If flags & SOMETHING - do not do it many times on the same match */
1043 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1044 /* XXX Does the destruction order has to change with utf8_target? */
1045 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1046 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1047 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1048 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1049 check = NULL; /* abort */
1051 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1052 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1053 if (prog->intflags & PREGf_IMPLICIT)
1054 prog->extflags &= ~RXf_ANCH_MBOL;
1055 /* XXXX This is a remnant of the old implementation. It
1056 looks wasteful, since now INTUIT can use many
1057 other heuristics. */
1058 prog->extflags &= ~RXf_USE_INTUIT;
1059 /* XXXX What other flags might need to be cleared in this branch? */
1065 /* Last resort... */
1066 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1067 /* trie stclasses are too expensive to use here, we are better off to
1068 leave it to regmatch itself */
1069 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1070 /* minlen == 0 is possible if regstclass is \b or \B,
1071 and the fixed substr is ''$.
1072 Since minlen is already taken into account, s+1 is before strend;
1073 accidentally, minlen >= 1 guaranties no false positives at s + 1
1074 even for \b or \B. But (minlen? 1 : 0) below assumes that
1075 regstclass does not come from lookahead... */
1076 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1077 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1078 const U8* const str = (U8*)STRING(progi->regstclass);
1079 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1080 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1083 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1084 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1085 else if (prog->float_substr || prog->float_utf8)
1086 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1090 if (checked_upto < s)
1092 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1093 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1096 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1101 const char *what = NULL;
1103 if (endpos == strend) {
1104 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1105 "Could not match STCLASS...\n") );
1108 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1109 "This position contradicts STCLASS...\n") );
1110 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1112 checked_upto = HOPBACKc(endpos, start_shift);
1113 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1114 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1115 /* Contradict one of substrings */
1116 if (prog->anchored_substr || prog->anchored_utf8) {
1117 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1118 DEBUG_EXECUTE_r( what = "anchored" );
1120 s = HOP3c(t, 1, strend);
1121 if (s + start_shift + end_shift > strend) {
1122 /* XXXX Should be taken into account earlier? */
1123 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1124 "Could not match STCLASS...\n") );
1129 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1130 "Looking for %s substr starting at offset %ld...\n",
1131 what, (long)(s + start_shift - i_strpos)) );
1134 /* Have both, check_string is floating */
1135 if (t + start_shift >= check_at) /* Contradicts floating=check */
1136 goto retry_floating_check;
1137 /* Recheck anchored substring, but not floating... */
1141 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1142 "Looking for anchored substr starting at offset %ld...\n",
1143 (long)(other_last - i_strpos)) );
1144 goto do_other_anchored;
1146 /* Another way we could have checked stclass at the
1147 current position only: */
1152 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1153 "Looking for /%s^%s/m starting at offset %ld...\n",
1154 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1157 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1159 /* Check is floating substring. */
1160 retry_floating_check:
1161 t = check_at - start_shift;
1162 DEBUG_EXECUTE_r( what = "floating" );
1163 goto hop_and_restart;
1166 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1167 "By STCLASS: moving %ld --> %ld\n",
1168 (long)(t - i_strpos), (long)(s - i_strpos))
1172 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1173 "Does not contradict STCLASS...\n");
1178 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1179 PL_colors[4], (check ? "Guessed" : "Giving up"),
1180 PL_colors[5], (long)(s - i_strpos)) );
1183 fail_finish: /* Substring not found */
1184 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1185 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1187 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1188 PL_colors[4], PL_colors[5]));
1192 #define DECL_TRIE_TYPE(scan) \
1193 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1194 trie_type = ((scan->flags == EXACT) \
1195 ? (utf8_target ? trie_utf8 : trie_plain) \
1196 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1198 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1199 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1201 switch (trie_type) { \
1202 case trie_utf8_fold: \
1203 if ( foldlen>0 ) { \
1204 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1209 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1210 len = UTF8SKIP(uc); \
1211 skiplen = UNISKIP( uvc ); \
1212 foldlen -= skiplen; \
1213 uscan = foldbuf + skiplen; \
1216 case trie_latin_utf8_fold: \
1217 if ( foldlen>0 ) { \
1218 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1224 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1225 skiplen = UNISKIP( uvc ); \
1226 foldlen -= skiplen; \
1227 uscan = foldbuf + skiplen; \
1231 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1238 charid = trie->charmap[ uvc ]; \
1242 if (widecharmap) { \
1243 SV** const svpp = hv_fetch(widecharmap, \
1244 (char*)&uvc, sizeof(UV), 0); \
1246 charid = (U16)SvIV(*svpp); \
1251 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1255 && (ln == 1 || folder(s, pat_string, ln)) \
1256 && (!reginfo || regtry(reginfo, &s)) ) \
1262 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1264 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1270 #define REXEC_FBC_SCAN(CoDe) \
1272 while (s < strend) { \
1278 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1279 REXEC_FBC_UTF8_SCAN( \
1281 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1290 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1293 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1302 #define REXEC_FBC_TRYIT \
1303 if ((!reginfo || regtry(reginfo, &s))) \
1306 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1307 if (utf8_target) { \
1308 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1311 REXEC_FBC_CLASS_SCAN(CoNd); \
1314 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1315 if (utf8_target) { \
1317 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1320 REXEC_FBC_CLASS_SCAN(CoNd); \
1323 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1324 PL_reg_flags |= RF_tainted; \
1325 if (utf8_target) { \
1326 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1329 REXEC_FBC_CLASS_SCAN(CoNd); \
1332 #define DUMP_EXEC_POS(li,s,doutf8) \
1333 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1336 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1337 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1338 tmp = TEST_NON_UTF8(tmp); \
1339 REXEC_FBC_UTF8_SCAN( \
1340 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1349 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1350 if (s == PL_bostr) { \
1354 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1355 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1358 LOAD_UTF8_CHARCLASS_ALNUM(); \
1359 REXEC_FBC_UTF8_SCAN( \
1360 if (tmp == ! (TeSt2_UtF8)) { \
1369 /* The only difference between the BOUND and NBOUND cases is that
1370 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1371 * NBOUND. This is accomplished by passing it in either the if or else clause,
1372 * with the other one being empty */
1373 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1374 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1376 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1377 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1379 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1380 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1382 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1383 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1386 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1387 * be passed in completely with the variable name being tested, which isn't
1388 * such a clean interface, but this is easier to read than it was before. We
1389 * are looking for the boundary (or non-boundary between a word and non-word
1390 * character. The utf8 and non-utf8 cases have the same logic, but the details
1391 * must be different. Find the "wordness" of the character just prior to this
1392 * one, and compare it with the wordness of this one. If they differ, we have
1393 * a boundary. At the beginning of the string, pretend that the previous
1394 * character was a new-line */
1395 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1396 if (utf8_target) { \
1399 else { /* Not utf8 */ \
1400 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1401 tmp = TEST_NON_UTF8(tmp); \
1403 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1412 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1415 /* We know what class REx starts with. Try to find this position... */
1416 /* if reginfo is NULL, its a dryrun */
1417 /* annoyingly all the vars in this routine have different names from their counterparts
1418 in regmatch. /grrr */
1421 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1422 const char *strend, regmatch_info *reginfo)
1425 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1426 char *pat_string; /* The pattern's exactish string */
1427 char *pat_end; /* ptr to end char of pat_string */
1428 re_fold_t folder; /* Function for computing non-utf8 folds */
1429 const U8 *fold_array; /* array for folding ords < 256 */
1432 register STRLEN uskip;
1436 register I32 tmp = 1; /* Scratch variable? */
1437 register const bool utf8_target = PL_reg_match_utf8;
1438 UV utf8_fold_flags = 0;
1439 RXi_GET_DECL(prog,progi);
1441 PERL_ARGS_ASSERT_FIND_BYCLASS;
1443 /* We know what class it must start with. */
1447 if (utf8_target || OP(c) == ANYOFV) {
1448 STRLEN inclasslen = strend - s;
1449 REXEC_FBC_UTF8_CLASS_SCAN(
1450 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1453 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1458 if (tmp && (!reginfo || regtry(reginfo, &s)))
1466 if (UTF_PATTERN || utf8_target) {
1467 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1468 goto do_exactf_utf8;
1470 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1471 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1472 goto do_exactf_non_utf8; /* isn't dealt with by these */
1477 /* regcomp.c already folded this if pattern is in UTF-8 */
1478 utf8_fold_flags = 0;
1479 goto do_exactf_utf8;
1481 fold_array = PL_fold;
1483 goto do_exactf_non_utf8;
1486 if (UTF_PATTERN || utf8_target) {
1487 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1488 goto do_exactf_utf8;
1490 fold_array = PL_fold_locale;
1491 folder = foldEQ_locale;
1492 goto do_exactf_non_utf8;
1496 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1498 goto do_exactf_utf8;
1500 case EXACTFU_TRICKYFOLD:
1502 if (UTF_PATTERN || utf8_target) {
1503 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1504 goto do_exactf_utf8;
1507 /* Any 'ss' in the pattern should have been replaced by regcomp,
1508 * so we don't have to worry here about this single special case
1509 * in the Latin1 range */
1510 fold_array = PL_fold_latin1;
1511 folder = foldEQ_latin1;
1515 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1516 are no glitches with fold-length differences
1517 between the target string and pattern */
1519 /* The idea in the non-utf8 EXACTF* cases is to first find the
1520 * first character of the EXACTF* node and then, if necessary,
1521 * case-insensitively compare the full text of the node. c1 is the
1522 * first character. c2 is its fold. This logic will not work for
1523 * Unicode semantics and the german sharp ss, which hence should
1524 * not be compiled into a node that gets here. */
1525 pat_string = STRING(c);
1526 ln = STR_LEN(c); /* length to match in octets/bytes */
1528 /* We know that we have to match at least 'ln' bytes (which is the
1529 * same as characters, since not utf8). If we have to match 3
1530 * characters, and there are only 2 availabe, we know without
1531 * trying that it will fail; so don't start a match past the
1532 * required minimum number from the far end */
1533 e = HOP3c(strend, -((I32)ln), s);
1535 if (!reginfo && e < s) {
1536 e = s; /* Due to minlen logic of intuit() */
1540 c2 = fold_array[c1];
1541 if (c1 == c2) { /* If char and fold are the same */
1542 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1545 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1554 /* If one of the operands is in utf8, we can't use the simpler
1555 * folding above, due to the fact that many different characters
1556 * can have the same fold, or portion of a fold, or different-
1558 pat_string = STRING(c);
1559 ln = STR_LEN(c); /* length to match in octets/bytes */
1560 pat_end = pat_string + ln;
1561 lnc = (UTF_PATTERN) /* length to match in characters */
1562 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1565 /* We have 'lnc' characters to match in the pattern, but because of
1566 * multi-character folding, each character in the target can match
1567 * up to 3 characters (Unicode guarantees it will never exceed
1568 * this) if it is utf8-encoded; and up to 2 if not (based on the
1569 * fact that the Latin 1 folds are already determined, and the
1570 * only multi-char fold in that range is the sharp-s folding to
1571 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1572 * string character. Adjust lnc accordingly, rounding up, so that
1573 * if we need to match at least 4+1/3 chars, that really is 5. */
1574 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1575 lnc = (lnc + expansion - 1) / expansion;
1577 /* As in the non-UTF8 case, if we have to match 3 characters, and
1578 * only 2 are left, it's guaranteed to fail, so don't start a
1579 * match that would require us to go beyond the end of the string
1581 e = HOP3c(strend, -((I32)lnc), s);
1583 if (!reginfo && e < s) {
1584 e = s; /* Due to minlen logic of intuit() */
1587 /* XXX Note that we could recalculate e to stop the loop earlier,
1588 * as the worst case expansion above will rarely be met, and as we
1589 * go along we would usually find that e moves further to the left.
1590 * This would happen only after we reached the point in the loop
1591 * where if there were no expansion we should fail. Unclear if
1592 * worth the expense */
1595 char *my_strend= (char *)strend;
1596 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1597 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1598 && (!reginfo || regtry(reginfo, &s)) )
1602 s += (utf8_target) ? UTF8SKIP(s) : 1;
1607 PL_reg_flags |= RF_tainted;
1608 FBC_BOUND(isALNUM_LC,
1609 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1610 isALNUM_LC_utf8((U8*)s));
1613 PL_reg_flags |= RF_tainted;
1614 FBC_NBOUND(isALNUM_LC,
1615 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1616 isALNUM_LC_utf8((U8*)s));
1619 FBC_BOUND(isWORDCHAR,
1621 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1624 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1626 isWORDCHAR_A((U8*)s));
1629 FBC_NBOUND(isWORDCHAR,
1631 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1634 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1636 isWORDCHAR_A((U8*)s));
1639 FBC_BOUND(isWORDCHAR_L1,
1641 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1644 FBC_NBOUND(isWORDCHAR_L1,
1646 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1649 REXEC_FBC_CSCAN_TAINT(
1650 isALNUM_LC_utf8((U8*)s),
1655 REXEC_FBC_CSCAN_PRELOAD(
1656 LOAD_UTF8_CHARCLASS_ALNUM(),
1657 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1658 isWORDCHAR_L1((U8) *s)
1662 REXEC_FBC_CSCAN_PRELOAD(
1663 LOAD_UTF8_CHARCLASS_ALNUM(),
1664 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1669 /* Don't need to worry about utf8, as it can match only a single
1670 * byte invariant character */
1671 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1674 REXEC_FBC_CSCAN_PRELOAD(
1675 LOAD_UTF8_CHARCLASS_ALNUM(),
1676 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1677 ! isWORDCHAR_L1((U8) *s)
1681 REXEC_FBC_CSCAN_PRELOAD(
1682 LOAD_UTF8_CHARCLASS_ALNUM(),
1683 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1694 REXEC_FBC_CSCAN_TAINT(
1695 !isALNUM_LC_utf8((U8*)s),
1700 REXEC_FBC_CSCAN_PRELOAD(
1701 LOAD_UTF8_CHARCLASS_SPACE(),
1702 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1707 REXEC_FBC_CSCAN_PRELOAD(
1708 LOAD_UTF8_CHARCLASS_SPACE(),
1709 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1714 /* Don't need to worry about utf8, as it can match only a single
1715 * byte invariant character */
1716 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1719 REXEC_FBC_CSCAN_TAINT(
1720 isSPACE_LC_utf8((U8*)s),
1725 REXEC_FBC_CSCAN_PRELOAD(
1726 LOAD_UTF8_CHARCLASS_SPACE(),
1727 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1728 ! isSPACE_L1((U8) *s)
1732 REXEC_FBC_CSCAN_PRELOAD(
1733 LOAD_UTF8_CHARCLASS_SPACE(),
1734 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1745 REXEC_FBC_CSCAN_TAINT(
1746 !isSPACE_LC_utf8((U8*)s),
1751 REXEC_FBC_CSCAN_PRELOAD(
1752 LOAD_UTF8_CHARCLASS_DIGIT(),
1753 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1758 /* Don't need to worry about utf8, as it can match only a single
1759 * byte invariant character */
1760 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1763 REXEC_FBC_CSCAN_TAINT(
1764 isDIGIT_LC_utf8((U8*)s),
1769 REXEC_FBC_CSCAN_PRELOAD(
1770 LOAD_UTF8_CHARCLASS_DIGIT(),
1771 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1782 REXEC_FBC_CSCAN_TAINT(
1783 !isDIGIT_LC_utf8((U8*)s),
1790 is_LNBREAK_latin1(s)
1802 !is_VERTWS_latin1(s)
1808 is_HORIZWS_latin1(s)
1813 !is_HORIZWS_utf8(s),
1814 !is_HORIZWS_latin1(s)
1821 /* what trie are we using right now */
1823 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1825 = (reg_trie_data*)progi->data->data[ aho->trie ];
1826 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1828 const char *last_start = strend - trie->minlen;
1830 const char *real_start = s;
1832 STRLEN maxlen = trie->maxlen;
1834 U8 **points; /* map of where we were in the input string
1835 when reading a given char. For ASCII this
1836 is unnecessary overhead as the relationship
1837 is always 1:1, but for Unicode, especially
1838 case folded Unicode this is not true. */
1839 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1843 GET_RE_DEBUG_FLAGS_DECL;
1845 /* We can't just allocate points here. We need to wrap it in
1846 * an SV so it gets freed properly if there is a croak while
1847 * running the match */
1850 sv_points=newSV(maxlen * sizeof(U8 *));
1851 SvCUR_set(sv_points,
1852 maxlen * sizeof(U8 *));
1853 SvPOK_on(sv_points);
1854 sv_2mortal(sv_points);
1855 points=(U8**)SvPV_nolen(sv_points );
1856 if ( trie_type != trie_utf8_fold
1857 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1860 bitmap=(U8*)trie->bitmap;
1862 bitmap=(U8*)ANYOF_BITMAP(c);
1864 /* this is the Aho-Corasick algorithm modified a touch
1865 to include special handling for long "unknown char"
1866 sequences. The basic idea being that we use AC as long
1867 as we are dealing with a possible matching char, when
1868 we encounter an unknown char (and we have not encountered
1869 an accepting state) we scan forward until we find a legal
1871 AC matching is basically that of trie matching, except
1872 that when we encounter a failing transition, we fall back
1873 to the current states "fail state", and try the current char
1874 again, a process we repeat until we reach the root state,
1875 state 1, or a legal transition. If we fail on the root state
1876 then we can either terminate if we have reached an accepting
1877 state previously, or restart the entire process from the beginning
1881 while (s <= last_start) {
1882 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1890 U8 *uscan = (U8*)NULL;
1891 U8 *leftmost = NULL;
1893 U32 accepted_word= 0;
1897 while ( state && uc <= (U8*)strend ) {
1899 U32 word = aho->states[ state ].wordnum;
1903 DEBUG_TRIE_EXECUTE_r(
1904 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1905 dump_exec_pos( (char *)uc, c, strend, real_start,
1906 (char *)uc, utf8_target );
1907 PerlIO_printf( Perl_debug_log,
1908 " Scanning for legal start char...\n");
1912 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1916 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1922 if (uc >(U8*)last_start) break;
1926 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1927 if (!leftmost || lpos < leftmost) {
1928 DEBUG_r(accepted_word=word);
1934 points[pointpos++ % maxlen]= uc;
1935 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1936 uscan, len, uvc, charid, foldlen,
1938 DEBUG_TRIE_EXECUTE_r({
1939 dump_exec_pos( (char *)uc, c, strend, real_start,
1941 PerlIO_printf(Perl_debug_log,
1942 " Charid:%3u CP:%4"UVxf" ",
1948 word = aho->states[ state ].wordnum;
1950 base = aho->states[ state ].trans.base;
1952 DEBUG_TRIE_EXECUTE_r({
1954 dump_exec_pos( (char *)uc, c, strend, real_start,
1956 PerlIO_printf( Perl_debug_log,
1957 "%sState: %4"UVxf", word=%"UVxf,
1958 failed ? " Fail transition to " : "",
1959 (UV)state, (UV)word);
1965 ( ((offset = base + charid
1966 - 1 - trie->uniquecharcount)) >= 0)
1967 && ((U32)offset < trie->lasttrans)
1968 && trie->trans[offset].check == state
1969 && (tmp=trie->trans[offset].next))
1971 DEBUG_TRIE_EXECUTE_r(
1972 PerlIO_printf( Perl_debug_log," - legal\n"));
1977 DEBUG_TRIE_EXECUTE_r(
1978 PerlIO_printf( Perl_debug_log," - fail\n"));
1980 state = aho->fail[state];
1984 /* we must be accepting here */
1985 DEBUG_TRIE_EXECUTE_r(
1986 PerlIO_printf( Perl_debug_log," - accepting\n"));
1995 if (!state) state = 1;
1998 if ( aho->states[ state ].wordnum ) {
1999 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2000 if (!leftmost || lpos < leftmost) {
2001 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2006 s = (char*)leftmost;
2007 DEBUG_TRIE_EXECUTE_r({
2009 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2010 (UV)accepted_word, (IV)(s - real_start)
2013 if (!reginfo || regtry(reginfo, &s)) {
2019 DEBUG_TRIE_EXECUTE_r({
2020 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2023 DEBUG_TRIE_EXECUTE_r(
2024 PerlIO_printf( Perl_debug_log,"No match.\n"));
2033 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2043 - regexec_flags - match a regexp against a string
2046 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2047 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2048 /* strend: pointer to null at end of string */
2049 /* strbeg: real beginning of string */
2050 /* minend: end of match must be >=minend after stringarg. */
2051 /* data: May be used for some additional optimizations.
2052 Currently its only used, with a U32 cast, for transmitting
2053 the ganch offset when doing a /g match. This will change */
2054 /* nosave: For optimizations. */
2057 struct regexp *const prog = (struct regexp *)SvANY(rx);
2058 /*register*/ char *s;
2059 register regnode *c;
2060 /*register*/ char *startpos = stringarg;
2061 I32 minlen; /* must match at least this many chars */
2062 I32 dontbother = 0; /* how many characters not to try at end */
2063 I32 end_shift = 0; /* Same for the end. */ /* CC */
2064 I32 scream_pos = -1; /* Internal iterator of scream. */
2065 char *scream_olds = NULL;
2066 const bool utf8_target = cBOOL(DO_UTF8(sv));
2068 RXi_GET_DECL(prog,progi);
2069 regmatch_info reginfo; /* create some info to pass to regtry etc */
2070 regexp_paren_pair *swap = NULL;
2071 GET_RE_DEBUG_FLAGS_DECL;
2073 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2074 PERL_UNUSED_ARG(data);
2076 /* Be paranoid... */
2077 if (prog == NULL || startpos == NULL) {
2078 Perl_croak(aTHX_ "NULL regexp parameter");
2082 multiline = prog->extflags & RXf_PMf_MULTILINE;
2083 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2085 RX_MATCH_UTF8_set(rx, utf8_target);
2087 debug_start_match(rx, utf8_target, startpos, strend,
2091 minlen = prog->minlen;
2093 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2094 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2095 "String too short [regexec_flags]...\n"));
2100 /* Check validity of program. */
2101 if (UCHARAT(progi->program) != REG_MAGIC) {
2102 Perl_croak(aTHX_ "corrupted regexp program");
2106 PL_reg_state.re_state_eval_setup_done = FALSE;
2110 PL_reg_flags |= RF_utf8;
2112 /* Mark beginning of line for ^ and lookbehind. */
2113 reginfo.bol = startpos; /* XXX not used ??? */
2117 /* Mark end of line for $ (and such) */
2120 /* see how far we have to get to not match where we matched before */
2121 reginfo.till = startpos+minend;
2123 /* If there is a "must appear" string, look for it. */
2126 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2128 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2129 reginfo.ganch = startpos + prog->gofs;
2130 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2131 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2132 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2134 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2135 && mg->mg_len >= 0) {
2136 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2137 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2138 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2140 if (prog->extflags & RXf_ANCH_GPOS) {
2141 if (s > reginfo.ganch)
2143 s = reginfo.ganch - prog->gofs;
2144 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2145 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2151 reginfo.ganch = strbeg + PTR2UV(data);
2152 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2153 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2155 } else { /* pos() not defined */
2156 reginfo.ganch = strbeg;
2157 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2158 "GPOS: reginfo.ganch = strbeg\n"));
2161 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2162 /* We have to be careful. If the previous successful match
2163 was from this regex we don't want a subsequent partially
2164 successful match to clobber the old results.
2165 So when we detect this possibility we add a swap buffer
2166 to the re, and switch the buffer each match. If we fail
2167 we switch it back, otherwise we leave it swapped.
2170 /* do we need a save destructor here for eval dies? */
2171 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2172 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2173 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2179 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2180 re_scream_pos_data d;
2182 d.scream_olds = &scream_olds;
2183 d.scream_pos = &scream_pos;
2184 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2186 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2187 goto phooey; /* not present */
2193 /* Simplest case: anchored match need be tried only once. */
2194 /* [unless only anchor is BOL and multiline is set] */
2195 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2196 if (s == startpos && regtry(®info, &startpos))
2198 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2199 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2204 dontbother = minlen - 1;
2205 end = HOP3c(strend, -dontbother, strbeg) - 1;
2206 /* for multiline we only have to try after newlines */
2207 if (prog->check_substr || prog->check_utf8) {
2208 /* because of the goto we can not easily reuse the macros for bifurcating the
2209 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2212 goto after_try_utf8;
2214 if (regtry(®info, &s)) {
2221 if (prog->extflags & RXf_USE_INTUIT) {
2222 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2231 } /* end search for check string in unicode */
2233 if (s == startpos) {
2234 goto after_try_latin;
2237 if (regtry(®info, &s)) {
2244 if (prog->extflags & RXf_USE_INTUIT) {
2245 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2254 } /* end search for check string in latin*/
2255 } /* end search for check string */
2256 else { /* search for newline */
2258 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2261 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2262 while (s <= end) { /* note it could be possible to match at the end of the string */
2263 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2264 if (regtry(®info, &s))
2268 } /* end search for newline */
2269 } /* end anchored/multiline check string search */
2271 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2273 /* the warning about reginfo.ganch being used without initialization
2274 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2275 and we only enter this block when the same bit is set. */
2276 char *tmp_s = reginfo.ganch - prog->gofs;
2278 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2283 /* Messy cases: unanchored match. */
2284 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2285 /* we have /x+whatever/ */
2286 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2291 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2292 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2293 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2298 DEBUG_EXECUTE_r( did_match = 1 );
2299 if (regtry(®info, &s)) goto got_it;
2301 while (s < strend && *s == ch)
2309 DEBUG_EXECUTE_r( did_match = 1 );
2310 if (regtry(®info, &s)) goto got_it;
2312 while (s < strend && *s == ch)
2317 DEBUG_EXECUTE_r(if (!did_match)
2318 PerlIO_printf(Perl_debug_log,
2319 "Did not find anchored character...\n")
2322 else if (prog->anchored_substr != NULL
2323 || prog->anchored_utf8 != NULL
2324 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2325 && prog->float_max_offset < strend - s)) {
2330 char *last1; /* Last position checked before */
2334 if (prog->anchored_substr || prog->anchored_utf8) {
2335 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2336 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2337 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2338 back_max = back_min = prog->anchored_offset;
2340 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2341 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2342 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2343 back_max = prog->float_max_offset;
2344 back_min = prog->float_min_offset;
2348 if (must == &PL_sv_undef)
2349 /* could not downgrade utf8 check substring, so must fail */
2355 last = HOP3c(strend, /* Cannot start after this */
2356 -(I32)(CHR_SVLEN(must)
2357 - (SvTAIL(must) != 0) + back_min), strbeg);
2360 last1 = HOPc(s, -1);
2362 last1 = s - 1; /* bogus */
2364 /* XXXX check_substr already used to find "s", can optimize if
2365 check_substr==must. */
2367 dontbother = end_shift;
2368 strend = HOPc(strend, -dontbother);
2369 while ( (s <= last) &&
2370 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2371 (unsigned char*)strend, must,
2372 multiline ? FBMrf_MULTILINE : 0)) ) {
2373 DEBUG_EXECUTE_r( did_match = 1 );
2374 if (HOPc(s, -back_max) > last1) {
2375 last1 = HOPc(s, -back_min);
2376 s = HOPc(s, -back_max);
2379 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2381 last1 = HOPc(s, -back_min);
2385 while (s <= last1) {
2386 if (regtry(®info, &s))
2392 while (s <= last1) {
2393 if (regtry(®info, &s))
2399 DEBUG_EXECUTE_r(if (!did_match) {
2400 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2401 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2402 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2403 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2404 ? "anchored" : "floating"),
2405 quoted, RE_SV_TAIL(must));
2409 else if ( (c = progi->regstclass) ) {
2411 const OPCODE op = OP(progi->regstclass);
2412 /* don't bother with what can't match */
2413 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2414 strend = HOPc(strend, -(minlen - 1));
2417 SV * const prop = sv_newmortal();
2418 regprop(prog, prop, c);
2420 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2422 PerlIO_printf(Perl_debug_log,
2423 "Matching stclass %.*s against %s (%d bytes)\n",
2424 (int)SvCUR(prop), SvPVX_const(prop),
2425 quoted, (int)(strend - s));
2428 if (find_byclass(prog, c, s, strend, ®info))
2430 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2434 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2441 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2442 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2443 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2445 little = SvPV_const(float_real, len);
2446 if (SvTAIL(float_real)) {
2447 /* This means that float_real contains an artificial \n on the end
2448 * due to the presence of something like this: /foo$/
2449 * where we can match both "foo" and "foo\n" at the end of the string.
2450 * So we have to compare the end of the string first against the float_real
2451 * without the \n and then against the full float_real with the string.
2452 * We have to watch out for cases where the string might be smaller
2453 * than the float_real or the float_real without the \n.
2455 char *checkpos= strend - len;
2457 PerlIO_printf(Perl_debug_log,
2458 "%sChecking for float_real.%s\n",
2459 PL_colors[4], PL_colors[5]));
2460 if (checkpos + 1 < strbeg) {
2461 /* can't match, even if we remove the trailing \n string is too short to match */
2463 PerlIO_printf(Perl_debug_log,
2464 "%sString shorter than required trailing substring, cannot match.%s\n",
2465 PL_colors[4], PL_colors[5]));
2467 } else if (memEQ(checkpos + 1, little, len - 1)) {
2468 /* can match, the end of the string matches without the "\n" */
2469 last = checkpos + 1;
2470 } else if (checkpos < strbeg) {
2471 /* cant match, string is too short when the "\n" is included */
2473 PerlIO_printf(Perl_debug_log,
2474 "%sString does not contain required trailing substring, cannot match.%s\n",
2475 PL_colors[4], PL_colors[5]));
2477 } else if (!multiline) {
2478 /* non multiline match, so compare with the "\n" at the end of the string */
2479 if (memEQ(checkpos, little, len)) {
2483 PerlIO_printf(Perl_debug_log,
2484 "%sString does not contain required trailing substring, cannot match.%s\n",
2485 PL_colors[4], PL_colors[5]));
2489 /* multiline match, so we have to search for a place where the full string is located */
2495 last = rninstr(s, strend, little, little + len);
2497 last = strend; /* matching "$" */
2500 /* at one point this block contained a comment which was probably
2501 * incorrect, which said that this was a "should not happen" case.
2502 * Even if it was true when it was written I am pretty sure it is
2503 * not anymore, so I have removed the comment and replaced it with
2506 PerlIO_printf(Perl_debug_log,
2507 "String does not contain required substring, cannot match.\n"
2511 dontbother = strend - last + prog->float_min_offset;
2513 if (minlen && (dontbother < minlen))
2514 dontbother = minlen - 1;
2515 strend -= dontbother; /* this one's always in bytes! */
2516 /* We don't know much -- general case. */
2519 if (regtry(®info, &s))
2528 if (regtry(®info, &s))
2530 } while (s++ < strend);
2540 PerlIO_printf(Perl_debug_log,
2541 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2547 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2549 if (PL_reg_state.re_state_eval_setup_done)
2550 restore_pos(aTHX_ prog);
2551 if (RXp_PAREN_NAMES(prog))
2552 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2554 /* make sure $`, $&, $', and $digit will work later */
2555 if ( !(flags & REXEC_NOT_FIRST) ) {
2556 RX_MATCH_COPY_FREE(rx);
2557 if (flags & REXEC_COPY_STR) {
2558 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2559 #ifdef PERL_OLD_COPY_ON_WRITE
2561 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2563 PerlIO_printf(Perl_debug_log,
2564 "Copy on write: regexp capture, type %d\n",
2567 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2568 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2569 assert (SvPOKp(prog->saved_copy));
2573 RX_MATCH_COPIED_on(rx);
2574 s = savepvn(strbeg, i);
2580 prog->subbeg = strbeg;
2581 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2588 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2589 PL_colors[4], PL_colors[5]));
2590 if (PL_reg_state.re_state_eval_setup_done)
2591 restore_pos(aTHX_ prog);
2593 /* we failed :-( roll it back */
2594 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2595 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2600 Safefree(prog->offs);
2608 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2609 * Do inc before dec, in case old and new rex are the same */
2610 #define SET_reg_curpm(Re2) \
2611 if (PL_reg_state.re_state_eval_setup_done) { \
2612 (void)ReREFCNT_inc(Re2); \
2613 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2614 PM_SETRE((PL_reg_curpm), (Re2)); \
2619 - regtry - try match at specific point
2621 STATIC I32 /* 0 failure, 1 success */
2622 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2626 REGEXP *const rx = reginfo->prog;
2627 regexp *const prog = (struct regexp *)SvANY(rx);
2628 RXi_GET_DECL(prog,progi);
2629 GET_RE_DEBUG_FLAGS_DECL;
2631 PERL_ARGS_ASSERT_REGTRY;
2633 reginfo->cutpoint=NULL;
2635 if ((prog->extflags & RXf_EVAL_SEEN)
2636 && !PL_reg_state.re_state_eval_setup_done)
2640 PL_reg_state.re_state_eval_setup_done = TRUE;
2642 /* Make $_ available to executed code. */
2643 if (reginfo->sv != DEFSV) {
2645 DEFSV_set(reginfo->sv);
2648 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2649 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2650 /* prepare for quick setting of pos */
2651 #ifdef PERL_OLD_COPY_ON_WRITE
2652 if (SvIsCOW(reginfo->sv))
2653 sv_force_normal_flags(reginfo->sv, 0);
2655 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2656 &PL_vtbl_mglob, NULL, 0);
2660 PL_reg_oldpos = mg->mg_len;
2661 SAVEDESTRUCTOR_X(restore_pos, prog);
2663 if (!PL_reg_curpm) {
2664 Newxz(PL_reg_curpm, 1, PMOP);
2667 SV* const repointer = &PL_sv_undef;
2668 /* this regexp is also owned by the new PL_reg_curpm, which
2669 will try to free it. */
2670 av_push(PL_regex_padav, repointer);
2671 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2672 PL_regex_pad = AvARRAY(PL_regex_padav);
2677 PL_reg_oldcurpm = PL_curpm;
2678 PL_curpm = PL_reg_curpm;
2679 if (RXp_MATCH_COPIED(prog)) {
2680 /* Here is a serious problem: we cannot rewrite subbeg,
2681 since it may be needed if this match fails. Thus
2682 $` inside (?{}) could fail... */
2683 PL_reg_oldsaved = prog->subbeg;
2684 PL_reg_oldsavedlen = prog->sublen;
2685 #ifdef PERL_OLD_COPY_ON_WRITE
2686 PL_nrs = prog->saved_copy;
2688 RXp_MATCH_COPIED_off(prog);
2691 PL_reg_oldsaved = NULL;
2692 prog->subbeg = PL_bostr;
2693 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2696 PL_reg_starttry = *startpos;
2698 prog->offs[0].start = *startpos - PL_bostr;
2699 PL_reginput = *startpos;
2700 prog->lastparen = 0;
2701 prog->lastcloseparen = 0;
2704 /* XXXX What this code is doing here?!!! There should be no need
2705 to do this again and again, prog->lastparen should take care of
2708 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2709 * Actually, the code in regcppop() (which Ilya may be meaning by
2710 * prog->lastparen), is not needed at all by the test suite
2711 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2712 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2713 * Meanwhile, this code *is* needed for the
2714 * above-mentioned test suite tests to succeed. The common theme
2715 * on those tests seems to be returning null fields from matches.
2716 * --jhi updated by dapm */
2718 if (prog->nparens) {
2719 regexp_paren_pair *pp = prog->offs;
2721 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2729 if (regmatch(reginfo, progi->program + 1)) {
2730 prog->offs[0].end = PL_reginput - PL_bostr;
2733 if (reginfo->cutpoint)
2734 *startpos= reginfo->cutpoint;
2735 REGCP_UNWIND(lastcp);
2740 #define sayYES goto yes
2741 #define sayNO goto no
2742 #define sayNO_SILENT goto no_silent
2744 /* we dont use STMT_START/END here because it leads to
2745 "unreachable code" warnings, which are bogus, but distracting. */
2746 #define CACHEsayNO \
2747 if (ST.cache_mask) \
2748 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2751 /* this is used to determine how far from the left messages like
2752 'failed...' are printed. It should be set such that messages
2753 are inline with the regop output that created them.
2755 #define REPORT_CODE_OFF 32
2758 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2759 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2761 #define SLAB_FIRST(s) (&(s)->states[0])
2762 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2764 /* grab a new slab and return the first slot in it */
2766 STATIC regmatch_state *
2769 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2772 regmatch_slab *s = PL_regmatch_slab->next;
2774 Newx(s, 1, regmatch_slab);
2775 s->prev = PL_regmatch_slab;
2777 PL_regmatch_slab->next = s;
2779 PL_regmatch_slab = s;
2780 return SLAB_FIRST(s);
2784 /* push a new state then goto it */
2786 #define PUSH_STATE_GOTO(state, node) \
2788 st->resume_state = state; \
2791 /* push a new state with success backtracking, then goto it */
2793 #define PUSH_YES_STATE_GOTO(state, node) \
2795 st->resume_state = state; \
2796 goto push_yes_state;
2802 regmatch() - main matching routine
2804 This is basically one big switch statement in a loop. We execute an op,
2805 set 'next' to point the next op, and continue. If we come to a point which
2806 we may need to backtrack to on failure such as (A|B|C), we push a
2807 backtrack state onto the backtrack stack. On failure, we pop the top
2808 state, and re-enter the loop at the state indicated. If there are no more
2809 states to pop, we return failure.
2811 Sometimes we also need to backtrack on success; for example /A+/, where
2812 after successfully matching one A, we need to go back and try to
2813 match another one; similarly for lookahead assertions: if the assertion
2814 completes successfully, we backtrack to the state just before the assertion
2815 and then carry on. In these cases, the pushed state is marked as
2816 'backtrack on success too'. This marking is in fact done by a chain of
2817 pointers, each pointing to the previous 'yes' state. On success, we pop to
2818 the nearest yes state, discarding any intermediate failure-only states.
2819 Sometimes a yes state is pushed just to force some cleanup code to be
2820 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2821 it to free the inner regex.
2823 Note that failure backtracking rewinds the cursor position, while
2824 success backtracking leaves it alone.
2826 A pattern is complete when the END op is executed, while a subpattern
2827 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2828 ops trigger the "pop to last yes state if any, otherwise return true"
2831 A common convention in this function is to use A and B to refer to the two
2832 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2833 the subpattern to be matched possibly multiple times, while B is the entire
2834 rest of the pattern. Variable and state names reflect this convention.
2836 The states in the main switch are the union of ops and failure/success of
2837 substates associated with with that op. For example, IFMATCH is the op
2838 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2839 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2840 successfully matched A and IFMATCH_A_fail is a state saying that we have
2841 just failed to match A. Resume states always come in pairs. The backtrack
2842 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2843 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2844 on success or failure.
2846 The struct that holds a backtracking state is actually a big union, with
2847 one variant for each major type of op. The variable st points to the
2848 top-most backtrack struct. To make the code clearer, within each
2849 block of code we #define ST to alias the relevant union.
2851 Here's a concrete example of a (vastly oversimplified) IFMATCH
2857 #define ST st->u.ifmatch
2859 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2860 ST.foo = ...; // some state we wish to save
2862 // push a yes backtrack state with a resume value of
2863 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2865 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2868 case IFMATCH_A: // we have successfully executed A; now continue with B
2870 bar = ST.foo; // do something with the preserved value
2873 case IFMATCH_A_fail: // A failed, so the assertion failed
2874 ...; // do some housekeeping, then ...
2875 sayNO; // propagate the failure
2882 For any old-timers reading this who are familiar with the old recursive
2883 approach, the code above is equivalent to:
2885 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2894 ...; // do some housekeeping, then ...
2895 sayNO; // propagate the failure
2898 The topmost backtrack state, pointed to by st, is usually free. If you
2899 want to claim it, populate any ST.foo fields in it with values you wish to
2900 save, then do one of
2902 PUSH_STATE_GOTO(resume_state, node);
2903 PUSH_YES_STATE_GOTO(resume_state, node);
2905 which sets that backtrack state's resume value to 'resume_state', pushes a
2906 new free entry to the top of the backtrack stack, then goes to 'node'.
2907 On backtracking, the free slot is popped, and the saved state becomes the
2908 new free state. An ST.foo field in this new top state can be temporarily
2909 accessed to retrieve values, but once the main loop is re-entered, it
2910 becomes available for reuse.
2912 Note that the depth of the backtrack stack constantly increases during the
2913 left-to-right execution of the pattern, rather than going up and down with
2914 the pattern nesting. For example the stack is at its maximum at Z at the
2915 end of the pattern, rather than at X in the following:
2917 /(((X)+)+)+....(Y)+....Z/
2919 The only exceptions to this are lookahead/behind assertions and the cut,
2920 (?>A), which pop all the backtrack states associated with A before
2923 Backtrack state structs are allocated in slabs of about 4K in size.
2924 PL_regmatch_state and st always point to the currently active state,
2925 and PL_regmatch_slab points to the slab currently containing
2926 PL_regmatch_state. The first time regmatch() is called, the first slab is
2927 allocated, and is never freed until interpreter destruction. When the slab
2928 is full, a new one is allocated and chained to the end. At exit from
2929 regmatch(), slabs allocated since entry are freed.
2934 #define DEBUG_STATE_pp(pp) \
2936 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2937 PerlIO_printf(Perl_debug_log, \
2938 " %*s"pp" %s%s%s%s%s\n", \
2940 PL_reg_name[st->resume_state], \
2941 ((st==yes_state||st==mark_state) ? "[" : ""), \
2942 ((st==yes_state) ? "Y" : ""), \
2943 ((st==mark_state) ? "M" : ""), \
2944 ((st==yes_state||st==mark_state) ? "]" : "") \
2949 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2954 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2955 const char *start, const char *end, const char *blurb)
2957 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2959 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2964 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2965 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2967 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2968 start, end - start, 60);
2970 PerlIO_printf(Perl_debug_log,
2971 "%s%s REx%s %s against %s\n",
2972 PL_colors[4], blurb, PL_colors[5], s0, s1);
2974 if (utf8_target||utf8_pat)
2975 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2976 utf8_pat ? "pattern" : "",
2977 utf8_pat && utf8_target ? " and " : "",
2978 utf8_target ? "string" : ""
2984 S_dump_exec_pos(pTHX_ const char *locinput,
2985 const regnode *scan,
2986 const char *loc_regeol,
2987 const char *loc_bostr,
2988 const char *loc_reg_starttry,
2989 const bool utf8_target)
2991 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2992 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2993 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2994 /* The part of the string before starttry has one color
2995 (pref0_len chars), between starttry and current
2996 position another one (pref_len - pref0_len chars),
2997 after the current position the third one.
2998 We assume that pref0_len <= pref_len, otherwise we
2999 decrease pref0_len. */
3000 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3001 ? (5 + taill) - l : locinput - loc_bostr;
3004 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3006 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3008 pref0_len = pref_len - (locinput - loc_reg_starttry);
3009 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3010 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3011 ? (5 + taill) - pref_len : loc_regeol - locinput);
3012 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3016 if (pref0_len > pref_len)
3017 pref0_len = pref_len;
3019 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3021 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3022 (locinput - pref_len),pref0_len, 60, 4, 5);
3024 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3025 (locinput - pref_len + pref0_len),
3026 pref_len - pref0_len, 60, 2, 3);
3028 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3029 locinput, loc_regeol - locinput, 10, 0, 1);
3031 const STRLEN tlen=len0+len1+len2;
3032 PerlIO_printf(Perl_debug_log,
3033 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3034 (IV)(locinput - loc_bostr),
3037 (docolor ? "" : "> <"),
3039 (int)(tlen > 19 ? 0 : 19 - tlen),
3046 /* reg_check_named_buff_matched()
3047 * Checks to see if a named buffer has matched. The data array of
3048 * buffer numbers corresponding to the buffer is expected to reside
3049 * in the regexp->data->data array in the slot stored in the ARG() of
3050 * node involved. Note that this routine doesn't actually care about the
3051 * name, that information is not preserved from compilation to execution.
3052 * Returns the index of the leftmost defined buffer with the given name
3053 * or 0 if non of the buffers matched.
3056 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3059 RXi_GET_DECL(rex,rexi);
3060 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3061 I32 *nums=(I32*)SvPVX(sv_dat);
3063 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3065 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3066 if ((I32)rex->lastparen >= nums[n] &&
3067 rex->offs[nums[n]].end != -1)
3076 /* free all slabs above current one - called during LEAVE_SCOPE */
3079 S_clear_backtrack_stack(pTHX_ void *p)
3081 regmatch_slab *s = PL_regmatch_slab->next;
3086 PL_regmatch_slab->next = NULL;
3088 regmatch_slab * const osl = s;
3095 STATIC I32 /* 0 failure, 1 success */
3096 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3098 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3102 register const bool utf8_target = PL_reg_match_utf8;
3103 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3104 REGEXP *rex_sv = reginfo->prog;
3105 regexp *rex = (struct regexp *)SvANY(rex_sv);
3106 RXi_GET_DECL(rex,rexi);
3108 /* the current state. This is a cached copy of PL_regmatch_state */
3109 register regmatch_state *st;
3110 /* cache heavy used fields of st in registers */
3111 register regnode *scan;
3112 register regnode *next;
3113 register U32 n = 0; /* general value; init to avoid compiler warning */
3114 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3115 register char *locinput = PL_reginput;
3116 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3118 bool result = 0; /* return value of S_regmatch */
3119 int depth = 0; /* depth of backtrack stack */
3120 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3121 const U32 max_nochange_depth =
3122 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3123 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3124 regmatch_state *yes_state = NULL; /* state to pop to on success of
3126 /* mark_state piggy backs on the yes_state logic so that when we unwind
3127 the stack on success we can update the mark_state as we go */
3128 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3129 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3130 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3132 bool no_final = 0; /* prevent failure from backtracking? */
3133 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3134 char *startpoint = PL_reginput;
3135 SV *popmark = NULL; /* are we looking for a mark? */
3136 SV *sv_commit = NULL; /* last mark name seen in failure */
3137 SV *sv_yes_mark = NULL; /* last mark name we have seen
3138 during a successful match */
3139 U32 lastopen = 0; /* last open we saw */
3140 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3141 SV* const oreplsv = GvSV(PL_replgv);
3142 /* these three flags are set by various ops to signal information to
3143 * the very next op. They have a useful lifetime of exactly one loop
3144 * iteration, and are not preserved or restored by state pushes/pops
3146 bool sw = 0; /* the condition value in (?(cond)a|b) */
3147 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3148 int logical = 0; /* the following EVAL is:
3152 or the following IFMATCH/UNLESSM is:
3153 false: plain (?=foo)
3154 true: used as a condition: (?(?=foo))
3156 PAD* last_pad = NULL;
3158 I32 gimme = G_SCALAR;
3159 CV *caller_cv = NULL; /* who called us */
3160 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3161 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3164 GET_RE_DEBUG_FLAGS_DECL;
3167 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3168 multicall_oldcatch = 0;
3169 multicall_cv = NULL;
3173 PERL_ARGS_ASSERT_REGMATCH;
3175 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3176 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3178 /* on first ever call to regmatch, allocate first slab */
3179 if (!PL_regmatch_slab) {
3180 Newx(PL_regmatch_slab, 1, regmatch_slab);
3181 PL_regmatch_slab->prev = NULL;
3182 PL_regmatch_slab->next = NULL;
3183 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3186 oldsave = PL_savestack_ix;
3187 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3188 SAVEVPTR(PL_regmatch_slab);
3189 SAVEVPTR(PL_regmatch_state);
3191 /* grab next free state slot */
3192 st = ++PL_regmatch_state;
3193 if (st > SLAB_LAST(PL_regmatch_slab))
3194 st = PL_regmatch_state = S_push_slab(aTHX);
3196 /* Note that nextchr is a byte even in UTF */
3197 nextchr = UCHARAT(locinput);
3199 while (scan != NULL) {
3202 SV * const prop = sv_newmortal();
3203 regnode *rnext=regnext(scan);
3204 DUMP_EXEC_POS( locinput, scan, utf8_target );
3205 regprop(rex, prop, scan);
3207 PerlIO_printf(Perl_debug_log,
3208 "%3"IVdf":%*s%s(%"IVdf")\n",
3209 (IV)(scan - rexi->program), depth*2, "",
3211 (PL_regkind[OP(scan)] == END || !rnext) ?
3212 0 : (IV)(rnext - rexi->program));
3215 next = scan + NEXT_OFF(scan);
3218 state_num = OP(scan);
3220 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3223 switch (state_num) {
3225 if (locinput == PL_bostr)
3227 /* reginfo->till = reginfo->bol; */
3232 if (locinput == PL_bostr ||
3233 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3239 if (locinput == PL_bostr)
3243 if (locinput == reginfo->ganch)
3248 /* update the startpoint */
3249 st->u.keeper.val = rex->offs[0].start;
3250 PL_reginput = locinput;
3251 rex->offs[0].start = locinput - PL_bostr;
3252 PUSH_STATE_GOTO(KEEPS_next, next);
3254 case KEEPS_next_fail:
3255 /* rollback the start point change */
3256 rex->offs[0].start = st->u.keeper.val;
3262 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3267 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3269 if (PL_regeol - locinput > 1)
3273 if (PL_regeol != locinput)
3277 if (!nextchr && locinput >= PL_regeol)
3280 locinput += PL_utf8skip[nextchr];
3281 if (locinput > PL_regeol)
3283 nextchr = UCHARAT(locinput);
3286 nextchr = UCHARAT(++locinput);
3289 if (!nextchr && locinput >= PL_regeol)
3291 nextchr = UCHARAT(++locinput);
3294 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3297 locinput += PL_utf8skip[nextchr];
3298 if (locinput > PL_regeol)
3300 nextchr = UCHARAT(locinput);
3303 nextchr = UCHARAT(++locinput);
3307 #define ST st->u.trie
3309 /* In this case the charclass data is available inline so
3310 we can fail fast without a lot of extra overhead.
3312 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3314 PerlIO_printf(Perl_debug_log,
3315 "%*s %sfailed to match trie start class...%s\n",
3316 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3319 assert(0); /* NOTREACHED */
3323 /* the basic plan of execution of the trie is:
3324 * At the beginning, run though all the states, and
3325 * find the longest-matching word. Also remember the position
3326 * of the shortest matching word. For example, this pattern:
3329 * when matched against the string "abcde", will generate
3330 * accept states for all words except 3, with the longest
3331 * matching word being 4, and the shortest being 1 (with
3332 * the position being after char 1 of the string).
3334 * Then for each matching word, in word order (i.e. 1,2,4,5),
3335 * we run the remainder of the pattern; on each try setting
3336 * the current position to the character following the word,
3337 * returning to try the next word on failure.
3339 * We avoid having to build a list of words at runtime by
3340 * using a compile-time structure, wordinfo[].prev, which
3341 * gives, for each word, the previous accepting word (if any).
3342 * In the case above it would contain the mappings 1->2, 2->0,
3343 * 3->0, 4->5, 5->1. We can use this table to generate, from
3344 * the longest word (4 above), a list of all words, by
3345 * following the list of prev pointers; this gives us the
3346 * unordered list 4,5,1,2. Then given the current word we have
3347 * just tried, we can go through the list and find the
3348 * next-biggest word to try (so if we just failed on word 2,
3349 * the next in the list is 4).
3351 * Since at runtime we don't record the matching position in
3352 * the string for each word, we have to work that out for
3353 * each word we're about to process. The wordinfo table holds
3354 * the character length of each word; given that we recorded
3355 * at the start: the position of the shortest word and its
3356 * length in chars, we just need to move the pointer the
3357 * difference between the two char lengths. Depending on
3358 * Unicode status and folding, that's cheap or expensive.
3360 * This algorithm is optimised for the case where are only a
3361 * small number of accept states, i.e. 0,1, or maybe 2.
3362 * With lots of accepts states, and having to try all of them,
3363 * it becomes quadratic on number of accept states to find all
3368 /* what type of TRIE am I? (utf8 makes this contextual) */
3369 DECL_TRIE_TYPE(scan);
3371 /* what trie are we using right now */
3372 reg_trie_data * const trie
3373 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3374 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3375 U32 state = trie->startstate;
3377 if (trie->bitmap && !TRIE_BITMAP_TEST(trie,*locinput) ) {
3378 if (trie->states[ state ].wordnum) {
3380 PerlIO_printf(Perl_debug_log,
3381 "%*s %smatched empty string...%s\n",
3382 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3388 PerlIO_printf(Perl_debug_log,
3389 "%*s %sfailed to match trie start class...%s\n",
3390 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3397 U8 *uc = ( U8* )locinput;
3401 U8 *uscan = (U8*)NULL;
3402 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3403 U32 charcount = 0; /* how many input chars we have matched */
3404 U32 accepted = 0; /* have we seen any accepting states? */
3406 ST.jump = trie->jump;
3409 ST.longfold = FALSE; /* char longer if folded => it's harder */
3412 /* fully traverse the TRIE; note the position of the
3413 shortest accept state and the wordnum of the longest
3416 while ( state && uc <= (U8*)PL_regeol ) {
3417 U32 base = trie->states[ state ].trans.base;
3421 wordnum = trie->states[ state ].wordnum;
3423 if (wordnum) { /* it's an accept state */
3426 /* record first match position */
3428 ST.firstpos = (U8*)locinput;
3433 ST.firstchars = charcount;
3436 if (!ST.nextword || wordnum < ST.nextword)
3437 ST.nextword = wordnum;
3438 ST.topword = wordnum;
3441 DEBUG_TRIE_EXECUTE_r({
3442 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3443 PerlIO_printf( Perl_debug_log,
3444 "%*s %sState: %4"UVxf" Accepted: %c ",
3445 2+depth * 2, "", PL_colors[4],
3446 (UV)state, (accepted ? 'Y' : 'N'));
3449 /* read a char and goto next state */
3452 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3453 uscan, len, uvc, charid, foldlen,
3460 base + charid - 1 - trie->uniquecharcount)) >= 0)
3462 && ((U32)offset < trie->lasttrans)
3463 && trie->trans[offset].check == state)
3465 state = trie->trans[offset].next;
3476 DEBUG_TRIE_EXECUTE_r(
3477 PerlIO_printf( Perl_debug_log,
3478 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3479 charid, uvc, (UV)state, PL_colors[5] );
3485 /* calculate total number of accept states */
3490 w = trie->wordinfo[w].prev;
3493 ST.accepted = accepted;
3497 PerlIO_printf( Perl_debug_log,
3498 "%*s %sgot %"IVdf" possible matches%s\n",
3499 REPORT_CODE_OFF + depth * 2, "",
3500 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3502 goto trie_first_try; /* jump into the fail handler */
3504 assert(0); /* NOTREACHED */
3506 case TRIE_next_fail: /* we failed - try next alternative */
3508 REGCP_UNWIND(ST.cp);
3509 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3511 if (!--ST.accepted) {
3513 PerlIO_printf( Perl_debug_log,
3514 "%*s %sTRIE failed...%s\n",
3515 REPORT_CODE_OFF+depth*2, "",
3522 /* Find next-highest word to process. Note that this code
3523 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3524 register U16 min = 0;
3526 register U16 const nextword = ST.nextword;
3527 register reg_trie_wordinfo * const wordinfo
3528 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3529 for (word=ST.topword; word; word=wordinfo[word].prev) {
3530 if (word > nextword && (!min || word < min))
3543 ST.lastparen = rex->lastparen;
3544 ST.lastcloseparen = rex->lastcloseparen;
3548 /* find start char of end of current word */
3550 U32 chars; /* how many chars to skip */
3551 U8 *uc = ST.firstpos;
3552 reg_trie_data * const trie
3553 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3555 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3557 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3561 /* the hard option - fold each char in turn and find
3562 * its folded length (which may be different */
3563 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3571 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3579 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3584 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3598 PL_reginput = (char *)uc;
3601 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
3602 ? ST.jump[ST.nextword]
3606 PerlIO_printf( Perl_debug_log,
3607 "%*s %sTRIE matched word #%d, continuing%s\n",
3608 REPORT_CODE_OFF+depth*2, "",
3615 if (ST.accepted > 1 || has_cutgroup) {
3616 PUSH_STATE_GOTO(TRIE_next, scan);
3617 assert(0); /* NOTREACHED */
3619 /* only one choice left - just continue */
3621 AV *const trie_words
3622 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3623 SV ** const tmp = av_fetch( trie_words,
3625 SV *sv= tmp ? sv_newmortal() : NULL;
3627 PerlIO_printf( Perl_debug_log,
3628 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3629 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3631 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3632 PL_colors[0], PL_colors[1],
3633 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3635 : "not compiled under -Dr",
3639 locinput = PL_reginput;
3640 nextchr = UCHARAT(locinput);
3641 continue; /* execute rest of RE */
3642 assert(0); /* NOTREACHED */
3646 char *s = STRING(scan);
3648 if (utf8_target != UTF_PATTERN) {
3649 /* The target and the pattern have differing utf8ness. */
3651 const char * const e = s + ln;
3654 /* The target is utf8, the pattern is not utf8. */
3659 if (NATIVE_TO_UNI(*(U8*)s) !=
3660 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3668 /* The target is not utf8, the pattern is utf8. */
3673 if (NATIVE_TO_UNI(*((U8*)l)) !=
3674 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3682 nextchr = UCHARAT(locinput);
3685 /* The target and the pattern have the same utf8ness. */
3686 /* Inline the first character, for speed. */
3687 if (UCHARAT(s) != nextchr)
3689 if (PL_regeol - locinput < ln)
3691 if (ln > 1 && memNE(s, locinput, ln))
3694 nextchr = UCHARAT(locinput);
3699 const U8 * fold_array;
3701 U32 fold_utf8_flags;
3703 PL_reg_flags |= RF_tainted;
3704 folder = foldEQ_locale;
3705 fold_array = PL_fold_locale;
3706 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3710 case EXACTFU_TRICKYFOLD:
3712 folder = foldEQ_latin1;
3713 fold_array = PL_fold_latin1;
3714 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3718 folder = foldEQ_latin1;
3719 fold_array = PL_fold_latin1;
3720 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3725 fold_array = PL_fold;
3726 fold_utf8_flags = 0;
3732 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
3733 /* Either target or the pattern are utf8, or has the issue where
3734 * the fold lengths may differ. */
3735 const char * const l = locinput;
3736 char *e = PL_regeol;
3738 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3739 l, &e, 0, utf8_target, fold_utf8_flags))
3744 nextchr = UCHARAT(locinput);
3748 /* Neither the target nor the pattern are utf8 */
3749 if (UCHARAT(s) != nextchr &&
3750 UCHARAT(s) != fold_array[nextchr])
3754 if (PL_regeol - locinput < ln)
3756 if (ln > 1 && ! folder(s, locinput, ln))
3759 nextchr = UCHARAT(locinput);
3763 /* XXX Could improve efficiency by separating these all out using a
3764 * macro or in-line function. At that point regcomp.c would no longer
3765 * have to set the FLAGS fields of these */
3768 PL_reg_flags |= RF_tainted;
3776 /* was last char in word? */
3778 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3779 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3781 if (locinput == PL_bostr)
3784 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3786 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3788 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3789 ln = isALNUM_uni(ln);
3790 LOAD_UTF8_CHARCLASS_ALNUM();
3791 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3794 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3795 n = isALNUM_LC_utf8((U8*)locinput);
3800 /* Here the string isn't utf8, or is utf8 and only ascii
3801 * characters are to match \w. In the latter case looking at
3802 * the byte just prior to the current one may be just the final
3803 * byte of a multi-byte character. This is ok. There are two
3805 * 1) it is a single byte character, and then the test is doing
3806 * just what it's supposed to.
3807 * 2) it is a multi-byte character, in which case the final
3808 * byte is never mistakable for ASCII, and so the test
3809 * will say it is not a word character, which is the
3810 * correct answer. */
3811 ln = (locinput != PL_bostr) ?
3812 UCHARAT(locinput - 1) : '\n';
3813 switch (FLAGS(scan)) {
3814 case REGEX_UNICODE_CHARSET:
3815 ln = isWORDCHAR_L1(ln);
3816 n = isWORDCHAR_L1(nextchr);
3818 case REGEX_LOCALE_CHARSET:
3819 ln = isALNUM_LC(ln);
3820 n = isALNUM_LC(nextchr);
3822 case REGEX_DEPENDS_CHARSET:
3824 n = isALNUM(nextchr);
3826 case REGEX_ASCII_RESTRICTED_CHARSET:
3827 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3828 ln = isWORDCHAR_A(ln);
3829 n = isWORDCHAR_A(nextchr);
3832 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3836 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3838 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3843 if (utf8_target || state_num == ANYOFV) {
3844 STRLEN inclasslen = PL_regeol - locinput;
3845 if (locinput >= PL_regeol)
3848 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3850 locinput += inclasslen;
3851 nextchr = UCHARAT(locinput);
3856 nextchr = UCHARAT(locinput);
3857 if (!nextchr && locinput >= PL_regeol)
3859 if (!REGINCLASS(rex, scan, (U8*)locinput))
3861 nextchr = UCHARAT(++locinput);
3865 /* Special char classes - The defines start on line 129 or so */
3866 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3867 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3868 ALNUMU, NALNUMU, isWORDCHAR_L1,
3869 ALNUMA, NALNUMA, isWORDCHAR_A,
3872 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3873 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3874 SPACEU, NSPACEU, isSPACE_L1,
3875 SPACEA, NSPACEA, isSPACE_A,
3878 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3879 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3880 DIGITA, NDIGITA, isDIGIT_A,
3883 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3884 a Unicode extended Grapheme Cluster */
3885 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3886 extended Grapheme Cluster is:
3889 | Prepend* Begin Extend*
3892 Begin is (Hangul-syllable | ! Control)
3893 Extend is (Grapheme_Extend | Spacing_Mark)
3894 Control is [ GCB_Control CR LF ]
3896 The discussion below shows how the code for CLUMP is derived
3897 from this regex. Note that most of these concepts are from
3898 property values of the Grapheme Cluster Boundary (GCB) property.
3899 No code point can have multiple property values for a given
3900 property. Thus a code point in Prepend can't be in Control, but
3901 it must be in !Control. This is why Control above includes
3902 GCB_Control plus CR plus LF. The latter two are used in the GCB
3903 property separately, and so can't be in GCB_Control, even though
3904 they logically are controls. Control is not the same as gc=cc,
3905 but includes format and other characters as well.
3907 The Unicode definition of Hangul-syllable is:
3909 | (L* ( ( V | LV ) V* | LVT ) T*)
3912 Each of these is a value for the GCB property, and hence must be
3913 disjoint, so the order they are tested is immaterial, so the
3914 above can safely be changed to
3917 | (L* ( LVT | ( V | LV ) V*) T*)
3919 The last two terms can be combined like this:
3921 | (( LVT | ( V | LV ) V*) T*))
3923 And refactored into this:
3924 L* (L | LVT T* | V V* T* | LV V* T*)
3926 That means that if we have seen any L's at all we can quit
3927 there, but if the next character is an LVT, a V, or an LV we
3930 There is a subtlety with Prepend* which showed up in testing.
3931 Note that the Begin, and only the Begin is required in:
3932 | Prepend* Begin Extend*
3933 Also, Begin contains '! Control'. A Prepend must be a
3934 '! Control', which means it must also be a Begin. What it
3935 comes down to is that if we match Prepend* and then find no
3936 suitable Begin afterwards, that if we backtrack the last
3937 Prepend, that one will be a suitable Begin.
3940 if (locinput >= PL_regeol)
3942 if (! utf8_target) {
3944 /* Match either CR LF or '.', as all the other possibilities
3946 locinput++; /* Match the . or CR */
3947 if (nextchr == '\r' /* And if it was CR, and the next is LF,
3949 && locinput < PL_regeol
3950 && UCHARAT(locinput) == '\n') locinput++;
3954 /* Utf8: See if is ( CR LF ); already know that locinput <
3955 * PL_regeol, so locinput+1 is in bounds */
3956 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3960 /* In case have to backtrack to beginning, then match '.' */
3961 char *starting = locinput;
3963 /* In case have to backtrack the last prepend */
3964 char *previous_prepend = 0;
3966 LOAD_UTF8_CHARCLASS_GCB();
3968 /* Match (prepend)* */
3969 while (locinput < PL_regeol
3970 && swash_fetch(PL_utf8_X_prepend,
3971 (U8*)locinput, utf8_target))
3973 previous_prepend = locinput;
3974 locinput += UTF8SKIP(locinput);
3977 /* As noted above, if we matched a prepend character, but
3978 * the next thing won't match, back off the last prepend we
3979 * matched, as it is guaranteed to match the begin */
3980 if (previous_prepend
3981 && (locinput >= PL_regeol
3982 || ! swash_fetch(PL_utf8_X_begin,
3983 (U8*)locinput, utf8_target)))
3985 locinput = previous_prepend;
3988 /* Note that here we know PL_regeol > locinput, as we
3989 * tested that upon input to this switch case, and if we
3990 * moved locinput forward, we tested the result just above
3991 * and it either passed, or we backed off so that it will
3993 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3995 /* Here did not match the required 'Begin' in the
3996 * second term. So just match the very first
3997 * character, the '.' of the final term of the regex */
3998 locinput = starting + UTF8SKIP(starting);
4001 /* Here is the beginning of a character that can have
4002 * an extender. It is either a hangul syllable, or a
4004 if (swash_fetch(PL_utf8_X_non_hangul,
4005 (U8*)locinput, utf8_target))
4008 /* Here not a Hangul syllable, must be a
4009 * ('! * Control') */
4010 locinput += UTF8SKIP(locinput);
4013 /* Here is a Hangul syllable. It can be composed
4014 * of several individual characters. One
4015 * possibility is T+ */
4016 if (swash_fetch(PL_utf8_X_T,
4017 (U8*)locinput, utf8_target))
4019 while (locinput < PL_regeol
4020 && swash_fetch(PL_utf8_X_T,
4021 (U8*)locinput, utf8_target))
4023 locinput += UTF8SKIP(locinput);
4027 /* Here, not T+, but is a Hangul. That means
4028 * it is one of the others: L, LV, LVT or V,
4030 * L* (L | LVT T* | V V* T* | LV V* T*) */
4033 while (locinput < PL_regeol
4034 && swash_fetch(PL_utf8_X_L,
4035 (U8*)locinput, utf8_target))
4037 locinput += UTF8SKIP(locinput);
4040 /* Here, have exhausted L*. If the next
4041 * character is not an LV, LVT nor V, it means
4042 * we had to have at least one L, so matches L+
4043 * in the original equation, we have a complete
4044 * hangul syllable. Are done. */
4046 if (locinput < PL_regeol
4047 && swash_fetch(PL_utf8_X_LV_LVT_V,
4048 (U8*)locinput, utf8_target))
4051 /* Otherwise keep going. Must be LV, LVT
4052 * or V. See if LVT */
4053 if (swash_fetch(PL_utf8_X_LVT,
4054 (U8*)locinput, utf8_target))
4056 locinput += UTF8SKIP(locinput);
4059 /* Must be V or LV. Take it, then
4061 locinput += UTF8SKIP(locinput);
4062 while (locinput < PL_regeol
4063 && swash_fetch(PL_utf8_X_V,
4064 (U8*)locinput, utf8_target))
4066 locinput += UTF8SKIP(locinput);
4070 /* And any of LV, LVT, or V can be followed
4072 while (locinput < PL_regeol
4073 && swash_fetch(PL_utf8_X_T,
4077 locinput += UTF8SKIP(locinput);
4083 /* Match any extender */
4084 while (locinput < PL_regeol
4085 && swash_fetch(PL_utf8_X_extend,
4086 (U8*)locinput, utf8_target))
4088 locinput += UTF8SKIP(locinput);
4092 if (locinput > PL_regeol) sayNO;
4094 nextchr = UCHARAT(locinput);
4098 { /* The capture buffer cases. The ones beginning with N for the
4099 named buffers just convert to the equivalent numbered and
4100 pretend they were called as the corresponding numbered buffer
4102 /* don't initialize these in the declaration, it makes C++
4107 const U8 *fold_array;
4110 PL_reg_flags |= RF_tainted;
4111 folder = foldEQ_locale;
4112 fold_array = PL_fold_locale;
4114 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4118 folder = foldEQ_latin1;
4119 fold_array = PL_fold_latin1;
4121 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4125 folder = foldEQ_latin1;
4126 fold_array = PL_fold_latin1;
4128 utf8_fold_flags = 0;
4133 fold_array = PL_fold;
4135 utf8_fold_flags = 0;
4142 utf8_fold_flags = 0;
4145 /* For the named back references, find the corresponding buffer
4147 n = reg_check_named_buff_matched(rex,scan);
4152 goto do_nref_ref_common;
4155 PL_reg_flags |= RF_tainted;
4156 folder = foldEQ_locale;
4157 fold_array = PL_fold_locale;
4158 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4162 folder = foldEQ_latin1;
4163 fold_array = PL_fold_latin1;
4164 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4168 folder = foldEQ_latin1;
4169 fold_array = PL_fold_latin1;
4170 utf8_fold_flags = 0;
4175 fold_array = PL_fold;
4176 utf8_fold_flags = 0;
4182 utf8_fold_flags = 0;
4186 n = ARG(scan); /* which paren pair */
4189 ln = rex->offs[n].start;
4190 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4191 if (rex->lastparen < n || ln == -1)
4192 sayNO; /* Do not match unless seen CLOSEn. */
4193 if (ln == rex->offs[n].end)
4197 if (type != REF /* REF can do byte comparison */
4198 && (utf8_target || type == REFFU))
4199 { /* XXX handle REFFL better */
4200 char * limit = PL_regeol;
4202 /* This call case insensitively compares the entire buffer
4203 * at s, with the current input starting at locinput, but
4204 * not going off the end given by PL_regeol, and returns in
4205 * limit upon success, how much of the current input was
4207 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4208 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4213 nextchr = UCHARAT(locinput);
4217 /* Not utf8: Inline the first character, for speed. */
4218 if (UCHARAT(s) != nextchr &&
4220 UCHARAT(s) != fold_array[nextchr]))
4222 ln = rex->offs[n].end - ln;
4223 if (locinput + ln > PL_regeol)
4225 if (ln > 1 && (type == REF
4226 ? memNE(s, locinput, ln)
4227 : ! folder(s, locinput, ln)))
4230 nextchr = UCHARAT(locinput);
4240 #define ST st->u.eval
4245 regexp_internal *rei;
4246 regnode *startpoint;
4249 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4250 if (cur_eval && cur_eval->locinput==locinput) {
4251 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4252 Perl_croak(aTHX_ "Infinite recursion in regex");
4253 if ( ++nochange_depth > max_nochange_depth )
4255 "Pattern subroutine nesting without pos change"
4256 " exceeded limit in regex");
4263 if (OP(scan)==GOSUB) {
4264 startpoint = scan + ARG2L(scan);
4265 ST.close_paren = ARG(scan);
4267 startpoint = rei->program+1;
4270 goto eval_recurse_doit;
4271 assert(0); /* NOTREACHED */
4272 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4273 if (cur_eval && cur_eval->locinput==locinput) {
4274 if ( ++nochange_depth > max_nochange_depth )
4275 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4280 /* execute the code in the {...} */
4284 OP * const oop = PL_op;
4285 COP * const ocurcop = PL_curcop;
4287 char *saved_regeol = PL_regeol;
4288 struct re_save_state saved_state;
4291 /* save *all* paren positions */
4293 REGCP_SET(runops_cp);
4295 /* To not corrupt the existing regex state while executing the
4296 * eval we would normally put it on the save stack, like with
4297 * save_re_context. However, re-evals have a weird scoping so we
4298 * can't just add ENTER/LEAVE here. With that, things like
4300 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4302 * would break, as they expect the localisation to be unwound
4303 * only when the re-engine backtracks through the bit that
4306 * What we do instead is just saving the state in a local c
4309 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4311 PL_reg_state.re_reparsing = FALSE;
4314 caller_cv = find_runcv(NULL);
4318 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4319 newcv = ((struct regexp *)SvANY(
4320 (REGEXP*)(rexi->data->data[n])
4323 nop = (OP*)rexi->data->data[n+1];
4325 else if (rexi->data->what[n] == 'l') { /* literal code */
4327 nop = (OP*)rexi->data->data[n];
4328 assert(CvDEPTH(newcv));
4331 /* literal with own CV */
4332 assert(rexi->data->what[n] == 'L');
4333 newcv = rex->qr_anoncv;
4334 nop = (OP*)rexi->data->data[n];
4337 /* normally if we're about to execute code from the same
4338 * CV that we used previously, we just use the existing
4339 * CX stack entry. However, its possible that in the
4340 * meantime we may have backtracked, popped from the save
4341 * stack, and undone the SAVECOMPPAD(s) associated with
4342 * PUSH_MULTICALL; in which case PL_comppad no longer
4343 * points to newcv's pad. */
4344 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4346 I32 depth = (newcv == caller_cv) ? 0 : 1;
4347 if (last_pushed_cv) {
4348 CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
4351 PUSH_MULTICALL_WITHDEPTH(newcv, depth);
4353 last_pushed_cv = newcv;
4355 last_pad = PL_comppad;
4357 /* the initial nextstate you would normally execute
4358 * at the start of an eval (which would cause error
4359 * messages to come from the eval), may be optimised
4360 * away from the execution path in the regex code blocks;
4361 * so manually set PL_curcop to it initially */
4363 OP *o = cUNOPx(nop)->op_first;
4364 assert(o->op_type == OP_NULL);
4365 if (o->op_targ == OP_SCOPE) {
4366 o = cUNOPo->op_first;
4369 assert(o->op_targ == OP_LEAVE);
4370 o = cUNOPo->op_first;
4371 assert(o->op_type == OP_ENTER);
4375 if (o->op_type != OP_STUB) {
4376 assert( o->op_type == OP_NEXTSTATE
4377 || o->op_type == OP_DBSTATE
4378 || (o->op_type == OP_NULL
4379 && ( o->op_targ == OP_NEXTSTATE
4380 || o->op_targ == OP_DBSTATE
4384 PL_curcop = (COP*)o;
4389 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4390 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4392 rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4395 SV *sv_mrk = get_sv("REGMARK", 1);
4396 sv_setsv(sv_mrk, sv_yes_mark);
4399 /* we don't use MULTICALL here as we want to call the
4400 * first op of the block of interest, rather than the
4401 * first op of the sub */
4404 CALLRUNOPS(aTHX); /* Scalar context. */
4407 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4413 /* before restoring everything, evaluate the returned
4414 * value, so that 'uninit' warnings don't use the wrong
4415 * PL_op or pad. Also need to process any magic vars
4416 * (e.g. $1) *before* parentheses are restored */
4421 if (logical == 0) /* (?{})/ */
4422 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
4423 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
4424 sw = cBOOL(SvTRUE(ret));
4427 else { /* /(??{}) */
4428 /* if its overloaded, let the regex compiler handle
4429 * it; otherwise extract regex, or stringify */
4430 if (!SvAMAGIC(ret)) {
4434 if (SvTYPE(sv) == SVt_REGEXP)
4435 re_sv = (REGEXP*) sv;
4436 else if (SvSMAGICAL(sv)) {
4437 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
4439 re_sv = (REGEXP *) mg->mg_obj;
4442 /* force any magic, undef warnings here */
4444 ret = sv_mortalcopy(ret);
4445 (void) SvPV_force_nolen(ret);
4451 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4453 /* *** Note that at this point we don't restore
4454 * PL_comppad, (or pop the CxSUB) on the assumption it may
4455 * be used again soon. This is safe as long as nothing
4456 * in the regexp code uses the pad ! */
4458 PL_curcop = ocurcop;
4459 PL_regeol = saved_regeol;
4460 S_regcp_restore(aTHX_ rex, runops_cp);
4466 /* only /(??{})/ from now on */
4469 /* extract RE object from returned value; compiling if
4473 re_sv = reg_temp_copy(NULL, re_sv);
4477 const I32 osize = PL_regsize;
4479 if (SvUTF8(ret) && IN_BYTES) {
4480 /* In use 'bytes': make a copy of the octet
4481 * sequence, but without the flag on */
4483 const char *const p = SvPV(ret, len);
4484 ret = newSVpvn_flags(p, len, SVs_TEMP);
4486 if (rex->intflags & PREGf_USE_RE_EVAL)
4487 pm_flags |= PMf_USE_RE_EVAL;
4489 /* if we got here, it should be an engine which
4490 * supports compiling code blocks and stuff */
4491 assert(rex->engine && rex->engine->op_comp);
4492 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
4493 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
4494 rex->engine, NULL, NULL,
4495 /* copy /msix etc to inner pattern */
4500 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4502 /* This isn't a first class regexp. Instead, it's
4503 caching a regexp onto an existing, Perl visible
4505 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
4508 /* safe to do now that any $1 etc has been
4509 * interpolated into the new pattern string and
4511 S_regcp_restore(aTHX_ rex, runops_cp);
4513 re = (struct regexp *)SvANY(re_sv);
4515 RXp_MATCH_COPIED_off(re);
4516 re->subbeg = rex->subbeg;
4517 re->sublen = rex->sublen;
4520 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4521 "Matching embedded");
4523 startpoint = rei->program + 1;
4524 ST.close_paren = 0; /* only used for GOSUB */
4526 eval_recurse_doit: /* Share code with GOSUB below this line */
4527 /* run the pattern returned from (??{...}) */
4528 ST.cp = regcppush(rex, 0); /* Save *all* the positions. */
4529 REGCP_SET(ST.lastcp);
4532 re->lastcloseparen = 0;
4534 PL_reginput = locinput;
4537 /* XXXX This is too dramatic a measure... */
4540 ST.toggle_reg_flags = PL_reg_flags;
4542 PL_reg_flags |= RF_utf8;
4544 PL_reg_flags &= ~RF_utf8;
4545 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4547 ST.prev_rex = rex_sv;
4548 ST.prev_curlyx = cur_curlyx;
4550 SET_reg_curpm(rex_sv);
4555 ST.prev_eval = cur_eval;
4557 /* now continue from first node in postoned RE */
4558 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4559 assert(0); /* NOTREACHED */
4562 case EVAL_AB: /* cleanup after a successful (??{A})B */
4563 /* note: this is called twice; first after popping B, then A */
4564 PL_reg_flags ^= ST.toggle_reg_flags;
4565 rex_sv = ST.prev_rex;
4566 SET_reg_curpm(rex_sv);
4567 rex = (struct regexp *)SvANY(rex_sv);
4568 rexi = RXi_GET(rex);
4570 cur_eval = ST.prev_eval;
4571 cur_curlyx = ST.prev_curlyx;
4573 /* XXXX This is too dramatic a measure... */
4575 if ( nochange_depth )
4580 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4581 /* note: this is called twice; first after popping B, then A */
4582 PL_reg_flags ^= ST.toggle_reg_flags;
4583 rex_sv = ST.prev_rex;
4584 SET_reg_curpm(rex_sv);
4585 rex = (struct regexp *)SvANY(rex_sv);
4586 rexi = RXi_GET(rex);
4588 PL_reginput = locinput;
4589 REGCP_UNWIND(ST.lastcp);
4591 cur_eval = ST.prev_eval;
4592 cur_curlyx = ST.prev_curlyx;
4593 /* XXXX This is too dramatic a measure... */
4595 if ( nochange_depth )
4601 n = ARG(scan); /* which paren pair */
4602 rex->offs[n].start_tmp = locinput - PL_bostr;
4605 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
4606 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; regsize=%"UVuf"\n",
4610 (IV)rex->offs[n].start_tmp,
4616 /* XXX really need to log other places start/end are set too */
4617 #define CLOSE_CAPTURE \
4618 rex->offs[n].start = rex->offs[n].start_tmp; \
4619 rex->offs[n].end = locinput - PL_bostr; \
4620 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
4621 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
4623 PTR2UV(rex->offs), \
4625 (IV)rex->offs[n].start, \
4626 (IV)rex->offs[n].end \
4630 n = ARG(scan); /* which paren pair */
4632 /*if (n > PL_regsize)
4634 if (n > rex->lastparen)
4636 rex->lastcloseparen = n;
4637 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4645 cursor && OP(cursor)!=END;
4646 cursor=regnext(cursor))
4648 if ( OP(cursor)==CLOSE ){
4650 if ( n <= lastopen ) {
4652 /*if (n > PL_regsize)
4654 if (n > rex->lastparen)
4656 rex->lastcloseparen = n;
4657 if ( n == ARG(scan) || (cur_eval &&
4658 cur_eval->u.eval.close_paren == n))
4667 n = ARG(scan); /* which paren pair */
4668 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
4671 /* reg_check_named_buff_matched returns 0 for no match */
4672 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4676 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4682 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4684 next = NEXTOPER(NEXTOPER(scan));
4686 next = scan + ARG(scan);
4687 if (OP(next) == IFTHEN) /* Fake one. */
4688 next = NEXTOPER(NEXTOPER(next));
4692 logical = scan->flags;
4695 /*******************************************************************
4697 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4698 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4699 STAR/PLUS/CURLY/CURLYN are used instead.)
4701 A*B is compiled as <CURLYX><A><WHILEM><B>
4703 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4704 state, which contains the current count, initialised to -1. It also sets
4705 cur_curlyx to point to this state, with any previous value saved in the
4708 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4709 since the pattern may possibly match zero times (i.e. it's a while {} loop
4710 rather than a do {} while loop).
4712 Each entry to WHILEM represents a successful match of A. The count in the
4713 CURLYX block is incremented, another WHILEM state is pushed, and execution
4714 passes to A or B depending on greediness and the current count.
4716 For example, if matching against the string a1a2a3b (where the aN are
4717 substrings that match /A/), then the match progresses as follows: (the
4718 pushed states are interspersed with the bits of strings matched so far):
4721 <CURLYX cnt=0><WHILEM>
4722 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4723 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4724 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4725 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4727 (Contrast this with something like CURLYM, which maintains only a single
4731 a1 <CURLYM cnt=1> a2
4732 a1 a2 <CURLYM cnt=2> a3
4733 a1 a2 a3 <CURLYM cnt=3> b
4736 Each WHILEM state block marks a point to backtrack to upon partial failure
4737 of A or B, and also contains some minor state data related to that
4738 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4739 overall state, such as the count, and pointers to the A and B ops.
4741 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4742 must always point to the *current* CURLYX block, the rules are:
4744 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4745 and set cur_curlyx to point the new block.
4747 When popping the CURLYX block after a successful or unsuccessful match,
4748 restore the previous cur_curlyx.
4750 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4751 to the outer one saved in the CURLYX block.
4753 When popping the WHILEM block after a successful or unsuccessful B match,
4754 restore the previous cur_curlyx.
4756 Here's an example for the pattern (AI* BI)*BO
4757 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4760 curlyx backtrack stack
4761 ------ ---------------
4763 CO <CO prev=NULL> <WO>
4764 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4765 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4766 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4768 At this point the pattern succeeds, and we work back down the stack to
4769 clean up, restoring as we go:
4771 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4772 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4773 CO <CO prev=NULL> <WO>
4776 *******************************************************************/
4778 #define ST st->u.curlyx
4780 case CURLYX: /* start of /A*B/ (for complex A) */
4782 /* No need to save/restore up to this paren */
4783 I32 parenfloor = scan->flags;
4785 assert(next); /* keep Coverity happy */
4786 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4789 /* XXXX Probably it is better to teach regpush to support
4790 parenfloor > PL_regsize... */
4791 if (parenfloor > (I32)rex->lastparen)
4792 parenfloor = rex->lastparen; /* Pessimization... */
4794 ST.prev_curlyx= cur_curlyx;
4796 ST.cp = PL_savestack_ix;
4798 /* these fields contain the state of the current curly.
4799 * they are accessed by subsequent WHILEMs */
4800 ST.parenfloor = parenfloor;
4805 ST.count = -1; /* this will be updated by WHILEM */
4806 ST.lastloc = NULL; /* this will be updated by WHILEM */
4808 PL_reginput = locinput;
4809 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4810 assert(0); /* NOTREACHED */
4813 case CURLYX_end: /* just finished matching all of A*B */
4814 cur_curlyx = ST.prev_curlyx;
4816 assert(0); /* NOTREACHED */
4818 case CURLYX_end_fail: /* just failed to match all of A*B */
4820 cur_curlyx = ST.prev_curlyx;
4822 assert(0); /* NOTREACHED */
4826 #define ST st->u.whilem
4828 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4830 /* see the discussion above about CURLYX/WHILEM */
4832 int min = ARG1(cur_curlyx->u.curlyx.me);
4833 int max = ARG2(cur_curlyx->u.curlyx.me);
4834 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4836 assert(cur_curlyx); /* keep Coverity happy */
4837 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4838 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4839 ST.cache_offset = 0;
4842 PL_reginput = locinput;
4844 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4845 "%*s whilem: matched %ld out of %d..%d\n",
4846 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4849 /* First just match a string of min A's. */
4852 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
4853 cur_curlyx->u.curlyx.lastloc = locinput;
4854 REGCP_SET(ST.lastcp);
4856 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4857 assert(0); /* NOTREACHED */
4860 /* If degenerate A matches "", assume A done. */
4862 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4863 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4864 "%*s whilem: empty match detected, trying continuation...\n",
4865 REPORT_CODE_OFF+depth*2, "")
4867 goto do_whilem_B_max;
4870 /* super-linear cache processing */
4874 if (!PL_reg_maxiter) {
4875 /* start the countdown: Postpone detection until we
4876 * know the match is not *that* much linear. */
4877 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4878 /* possible overflow for long strings and many CURLYX's */
4879 if (PL_reg_maxiter < 0)
4880 PL_reg_maxiter = I32_MAX;
4881 PL_reg_leftiter = PL_reg_maxiter;
4884 if (PL_reg_leftiter-- == 0) {
4885 /* initialise cache */
4886 const I32 size = (PL_reg_maxiter + 7)/8;
4887 if (PL_reg_poscache) {
4888 if ((I32)PL_reg_poscache_size < size) {
4889 Renew(PL_reg_poscache, size, char);
4890 PL_reg_poscache_size = size;
4892 Zero(PL_reg_poscache, size, char);
4895 PL_reg_poscache_size = size;
4896 Newxz(PL_reg_poscache, size, char);
4898 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4899 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4900 PL_colors[4], PL_colors[5])
4904 if (PL_reg_leftiter < 0) {
4905 /* have we already failed at this position? */
4907 offset = (scan->flags & 0xf) - 1
4908 + (locinput - PL_bostr) * (scan->flags>>4);
4909 mask = 1 << (offset % 8);
4911 if (PL_reg_poscache[offset] & mask) {
4912 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4913 "%*s whilem: (cache) already tried at this position...\n",
4914 REPORT_CODE_OFF+depth*2, "")
4916 sayNO; /* cache records failure */
4918 ST.cache_offset = offset;
4919 ST.cache_mask = mask;
4923 /* Prefer B over A for minimal matching. */
4925 if (cur_curlyx->u.curlyx.minmod) {
4926 ST.save_curlyx = cur_curlyx;
4927 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4928 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor);
4929 REGCP_SET(ST.lastcp);
4930 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4931 assert(0); /* NOTREACHED */
4934 /* Prefer A over B for maximal matching. */
4936 if (n < max) { /* More greed allowed? */
4937 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
4938 cur_curlyx->u.curlyx.lastloc = locinput;
4939 REGCP_SET(ST.lastcp);
4940 PUSH_STATE_GOTO(WHILEM_A_max, A);
4941 assert(0); /* NOTREACHED */
4943 goto do_whilem_B_max;
4945 assert(0); /* NOTREACHED */
4947 case WHILEM_B_min: /* just matched B in a minimal match */
4948 case WHILEM_B_max: /* just matched B in a maximal match */
4949 cur_curlyx = ST.save_curlyx;
4951 assert(0); /* NOTREACHED */
4953 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4954 cur_curlyx = ST.save_curlyx;
4955 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4956 cur_curlyx->u.curlyx.count--;
4958 assert(0); /* NOTREACHED */
4960 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4962 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4963 REGCP_UNWIND(ST.lastcp);
4965 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4966 cur_curlyx->u.curlyx.count--;
4968 assert(0); /* NOTREACHED */
4970 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4971 REGCP_UNWIND(ST.lastcp);
4972 regcppop(rex); /* Restore some previous $<digit>s? */
4973 PL_reginput = locinput;
4974 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4975 "%*s whilem: failed, trying continuation...\n",
4976 REPORT_CODE_OFF+depth*2, "")
4979 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4980 && ckWARN(WARN_REGEXP)
4981 && !(PL_reg_flags & RF_warned))
4983 PL_reg_flags |= RF_warned;
4984 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4985 "Complex regular subexpression recursion limit (%d) "
4991 ST.save_curlyx = cur_curlyx;
4992 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4993 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4994 assert(0); /* NOTREACHED */
4996 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4997 cur_curlyx = ST.save_curlyx;
4998 REGCP_UNWIND(ST.lastcp);
5001 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5002 /* Maximum greed exceeded */
5003 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5004 && ckWARN(WARN_REGEXP)
5005 && !(PL_reg_flags & RF_warned))
5007 PL_reg_flags |= RF_warned;
5008 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5009 "Complex regular subexpression recursion "
5010 "limit (%d) exceeded",
5013 cur_curlyx->u.curlyx.count--;
5017 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5018 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5020 /* Try grabbing another A and see if it helps. */
5021 PL_reginput = locinput;
5022 cur_curlyx->u.curlyx.lastloc = locinput;
5023 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5024 REGCP_SET(ST.lastcp);
5025 PUSH_STATE_GOTO(WHILEM_A_min,
5026 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
5027 assert(0); /* NOTREACHED */
5030 #define ST st->u.branch
5032 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5033 next = scan + ARG(scan);
5036 scan = NEXTOPER(scan);
5039 case BRANCH: /* /(...|A|...)/ */
5040 scan = NEXTOPER(scan); /* scan now points to inner node */
5041 ST.lastparen = rex->lastparen;
5042 ST.lastcloseparen = rex->lastcloseparen;
5043 ST.next_branch = next;
5045 PL_reginput = locinput;
5047 /* Now go into the branch */
5049 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
5051 PUSH_STATE_GOTO(BRANCH_next, scan);
5053 assert(0); /* NOTREACHED */
5055 PL_reginput = locinput;
5056 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5057 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5058 PUSH_STATE_GOTO(CUTGROUP_next,next);
5059 assert(0); /* NOTREACHED */
5060 case CUTGROUP_next_fail:
5063 if (st->u.mark.mark_name)
5064 sv_commit = st->u.mark.mark_name;
5066 assert(0); /* NOTREACHED */
5069 assert(0); /* NOTREACHED */
5070 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5075 REGCP_UNWIND(ST.cp);
5076 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5077 scan = ST.next_branch;
5078 /* no more branches? */
5079 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5081 PerlIO_printf( Perl_debug_log,
5082 "%*s %sBRANCH failed...%s\n",
5083 REPORT_CODE_OFF+depth*2, "",
5089 continue; /* execute next BRANCH[J] op */
5090 assert(0); /* NOTREACHED */
5097 #define ST st->u.curlym
5099 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5101 /* This is an optimisation of CURLYX that enables us to push
5102 * only a single backtracking state, no matter how many matches
5103 * there are in {m,n}. It relies on the pattern being constant
5104 * length, with no parens to influence future backrefs
5108 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5110 ST.lastparen = rex->lastparen;
5111 ST.lastcloseparen = rex->lastcloseparen;
5113 /* if paren positive, emulate an OPEN/CLOSE around A */
5115 U32 paren = ST.me->flags;
5116 if (paren > PL_regsize)
5118 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5126 ST.c1 = CHRTEST_UNINIT;
5129 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5132 curlym_do_A: /* execute the A in /A{m,n}B/ */
5133 PL_reginput = locinput;
5134 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
5135 assert(0); /* NOTREACHED */
5137 case CURLYM_A: /* we've just matched an A */
5138 locinput = st->locinput;
5139 nextchr = UCHARAT(locinput);
5142 /* after first match, determine A's length: u.curlym.alen */
5143 if (ST.count == 1) {
5144 if (PL_reg_match_utf8) {
5146 while (s < PL_reginput) {
5152 ST.alen = PL_reginput - locinput;
5155 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5158 PerlIO_printf(Perl_debug_log,
5159 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5160 (int)(REPORT_CODE_OFF+(depth*2)), "",
5161 (IV) ST.count, (IV)ST.alen)
5164 locinput = PL_reginput;
5166 if (cur_eval && cur_eval->u.eval.close_paren &&
5167 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5171 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5172 if ( max == REG_INFTY || ST.count < max )
5173 goto curlym_do_A; /* try to match another A */
5175 goto curlym_do_B; /* try to match B */
5177 case CURLYM_A_fail: /* just failed to match an A */
5178 REGCP_UNWIND(ST.cp);
5180 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5181 || (cur_eval && cur_eval->u.eval.close_paren &&
5182 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5185 curlym_do_B: /* execute the B in /A{m,n}B/ */
5186 PL_reginput = locinput;
5187 if (ST.c1 == CHRTEST_UNINIT) {
5188 /* calculate c1 and c2 for possible match of 1st char
5189 * following curly */
5190 ST.c1 = ST.c2 = CHRTEST_VOID;
5191 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5192 regnode *text_node = ST.B;
5193 if (! HAS_TEXT(text_node))
5194 FIND_NEXT_IMPT(text_node);
5197 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5199 But the former is redundant in light of the latter.
5201 if this changes back then the macro for
5202 IS_TEXT and friends need to change.
5204 if (PL_regkind[OP(text_node)] == EXACT)
5207 ST.c1 = (U8)*STRING(text_node);
5208 switch (OP(text_node)) {
5209 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5212 case EXACTFU_TRICKYFOLD:
5213 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5214 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5215 default: ST.c2 = ST.c1;
5222 PerlIO_printf(Perl_debug_log,
5223 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5224 (int)(REPORT_CODE_OFF+(depth*2)),
5227 if (ST.c1 != CHRTEST_VOID
5228 && UCHARAT(PL_reginput) != ST.c1
5229 && UCHARAT(PL_reginput) != ST.c2)
5231 /* simulate B failing */
5233 PerlIO_printf(Perl_debug_log,
5234 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5235 (int)(REPORT_CODE_OFF+(depth*2)),"",
5238 state_num = CURLYM_B_fail;
5239 goto reenter_switch;
5243 /* emulate CLOSE: mark current A as captured */
5244 I32 paren = ST.me->flags;
5246 rex->offs[paren].start
5247 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5248 rex->offs[paren].end = PL_reginput - PL_bostr;
5249 if ((U32)paren > rex->lastparen)
5250 rex->lastparen = paren;
5251 rex->lastcloseparen = paren;
5254 rex->offs[paren].end = -1;
5255 if (cur_eval && cur_eval->u.eval.close_paren &&
5256 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5265 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5266 assert(0); /* NOTREACHED */
5268 case CURLYM_B_fail: /* just failed to match a B */
5269 REGCP_UNWIND(ST.cp);
5270 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5272 I32 max = ARG2(ST.me);
5273 if (max != REG_INFTY && ST.count == max)
5275 goto curlym_do_A; /* try to match a further A */
5277 /* backtrack one A */
5278 if (ST.count == ARG1(ST.me) /* min */)
5281 locinput = HOPc(locinput, -ST.alen);
5282 goto curlym_do_B; /* try to match B */
5285 #define ST st->u.curly
5287 #define CURLY_SETPAREN(paren, success) \
5290 rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5291 rex->offs[paren].end = locinput - PL_bostr; \
5292 if (paren > rex->lastparen) \
5293 rex->lastparen = paren; \
5294 rex->lastcloseparen = paren; \
5297 rex->offs[paren].end = -1; \
5298 rex->lastparen = ST.lastparen; \
5299 rex->lastcloseparen = ST.lastcloseparen; \
5303 case STAR: /* /A*B/ where A is width 1 */
5307 scan = NEXTOPER(scan);
5309 case PLUS: /* /A+B/ where A is width 1 */
5313 scan = NEXTOPER(scan);
5315 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5316 ST.paren = scan->flags; /* Which paren to set */
5317 ST.lastparen = rex->lastparen;
5318 ST.lastcloseparen = rex->lastcloseparen;
5319 if (ST.paren > PL_regsize)
5320 PL_regsize = ST.paren;
5321 ST.min = ARG1(scan); /* min to match */
5322 ST.max = ARG2(scan); /* max to match */
5323 if (cur_eval && cur_eval->u.eval.close_paren &&
5324 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5328 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5330 case CURLY: /* /A{m,n}B/ where A is width 1 */
5332 ST.min = ARG1(scan); /* min to match */
5333 ST.max = ARG2(scan); /* max to match */
5334 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5337 * Lookahead to avoid useless match attempts
5338 * when we know what character comes next.
5340 * Used to only do .*x and .*?x, but now it allows
5341 * for )'s, ('s and (?{ ... })'s to be in the way
5342 * of the quantifier and the EXACT-like node. -- japhy
5345 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5347 if (HAS_TEXT(next) || JUMPABLE(next)) {
5349 regnode *text_node = next;
5351 if (! HAS_TEXT(text_node))
5352 FIND_NEXT_IMPT(text_node);
5354 if (! HAS_TEXT(text_node))
5355 ST.c1 = ST.c2 = CHRTEST_VOID;
5357 if ( PL_regkind[OP(text_node)] != EXACT ) {
5358 ST.c1 = ST.c2 = CHRTEST_VOID;
5359 goto assume_ok_easy;
5362 s = (U8*)STRING(text_node);
5364 /* Currently we only get here when
5366 PL_rekind[OP(text_node)] == EXACT
5368 if this changes back then the macro for IS_TEXT and
5369 friends need to change. */
5372 switch (OP(text_node)) {
5373 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5376 case EXACTFU_TRICKYFOLD:
5377 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5378 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5379 default: ST.c2 = ST.c1; break;
5382 else { /* UTF_PATTERN */
5383 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5385 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5387 to_utf8_fold((U8*)s, tmpbuf, &ulen);
5388 ST.c1 = ST.c2 = utf8n_to_uvchr(tmpbuf, UTF8_MAXLEN, 0,
5392 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5399 ST.c1 = ST.c2 = CHRTEST_VOID;
5404 PL_reginput = locinput;
5407 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5410 locinput = PL_reginput;
5412 if (ST.c1 == CHRTEST_VOID)
5413 goto curly_try_B_min;
5415 ST.oldloc = locinput;
5417 /* set ST.maxpos to the furthest point along the
5418 * string that could possibly match */
5419 if (ST.max == REG_INFTY) {
5420 ST.maxpos = PL_regeol - 1;
5422 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5425 else if (utf8_target) {
5426 int m = ST.max - ST.min;
5427 for (ST.maxpos = locinput;
5428 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5429 ST.maxpos += UTF8SKIP(ST.maxpos);
5432 ST.maxpos = locinput + ST.max - ST.min;
5433 if (ST.maxpos >= PL_regeol)
5434 ST.maxpos = PL_regeol - 1;
5436 goto curly_try_B_min_known;
5440 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5441 locinput = PL_reginput;
5442 if (ST.count < ST.min)
5444 if ((ST.count > ST.min)
5445 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5447 /* A{m,n} must come at the end of the string, there's
5448 * no point in backing off ... */
5450 /* ...except that $ and \Z can match before *and* after
5451 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5452 We may back off by one in this case. */
5453 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5457 goto curly_try_B_max;
5459 assert(0); /* NOTREACHED */
5462 case CURLY_B_min_known_fail:
5463 /* failed to find B in a non-greedy match where c1,c2 valid */
5465 PL_reginput = locinput; /* Could be reset... */
5466 REGCP_UNWIND(ST.cp);
5468 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5470 /* Couldn't or didn't -- move forward. */
5471 ST.oldloc = locinput;
5473 locinput += UTF8SKIP(locinput);
5477 curly_try_B_min_known:
5478 /* find the next place where 'B' could work, then call B */
5482 n = (ST.oldloc == locinput) ? 0 : 1;
5483 if (ST.c1 == ST.c2) {
5485 /* set n to utf8_distance(oldloc, locinput) */
5486 while (locinput <= ST.maxpos &&
5487 utf8n_to_uvchr((U8*)locinput,
5488 UTF8_MAXBYTES, &len,
5489 uniflags) != (UV)ST.c1) {
5495 /* set n to utf8_distance(oldloc, locinput) */
5496 while (locinput <= ST.maxpos) {
5498 const UV c = utf8n_to_uvchr((U8*)locinput,
5499 UTF8_MAXBYTES, &len,
5501 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5509 if (ST.c1 == ST.c2) {
5510 while (locinput <= ST.maxpos &&
5511 UCHARAT(locinput) != ST.c1)
5515 while (locinput <= ST.maxpos
5516 && UCHARAT(locinput) != ST.c1
5517 && UCHARAT(locinput) != ST.c2)
5520 n = locinput - ST.oldloc;
5522 if (locinput > ST.maxpos)
5524 /* PL_reginput == oldloc now */
5527 if (regrepeat(rex, ST.A, n, depth) < n)
5530 PL_reginput = locinput;
5531 CURLY_SETPAREN(ST.paren, ST.count);
5532 if (cur_eval && cur_eval->u.eval.close_paren &&
5533 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5536 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5538 assert(0); /* NOTREACHED */
5541 case CURLY_B_min_fail:
5542 /* failed to find B in a non-greedy match where c1,c2 invalid */
5544 REGCP_UNWIND(ST.cp);
5546 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5548 /* failed -- move forward one */
5549 PL_reginput = locinput;
5550 if (regrepeat(rex, ST.A, 1, depth)) {
5552 locinput = PL_reginput;
5553 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5554 ST.count > 0)) /* count overflow ? */
5557 CURLY_SETPAREN(ST.paren, ST.count);
5558 if (cur_eval && cur_eval->u.eval.close_paren &&
5559 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5562 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5566 assert(0); /* NOTREACHED */
5570 /* a successful greedy match: now try to match B */
5571 if (cur_eval && cur_eval->u.eval.close_paren &&
5572 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5577 if (ST.c1 != CHRTEST_VOID)
5578 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5579 UTF8_MAXBYTES, 0, uniflags)
5580 : (UV) UCHARAT(PL_reginput);
5581 /* If it could work, try it. */
5582 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5583 CURLY_SETPAREN(ST.paren, ST.count);
5584 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5585 assert(0); /* NOTREACHED */
5589 case CURLY_B_max_fail:
5590 /* failed to find B in a greedy match */
5592 REGCP_UNWIND(ST.cp);
5594 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5597 if (--ST.count < ST.min)
5599 PL_reginput = locinput = HOPc(locinput, -1);
5600 goto curly_try_B_max;
5607 /* we've just finished A in /(??{A})B/; now continue with B */
5608 st->u.eval.toggle_reg_flags
5609 = cur_eval->u.eval.toggle_reg_flags;
5610 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5612 st->u.eval.prev_rex = rex_sv; /* inner */
5613 st->u.eval.cp = regcppush(rex, 0); /* Save *all* the positions. */
5614 rex_sv = cur_eval->u.eval.prev_rex;
5615 SET_reg_curpm(rex_sv);
5616 rex = (struct regexp *)SvANY(rex_sv);
5617 rexi = RXi_GET(rex);
5618 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5620 REGCP_SET(st->u.eval.lastcp);
5621 PL_reginput = locinput;
5623 /* Restore parens of the outer rex without popping the
5625 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp);
5627 st->u.eval.prev_eval = cur_eval;
5628 cur_eval = cur_eval->u.eval.prev_eval;
5630 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5631 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5632 if ( nochange_depth )
5635 PUSH_YES_STATE_GOTO(EVAL_AB,
5636 st->u.eval.prev_eval->u.eval.B); /* match B */
5639 if (locinput < reginfo->till) {
5640 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5641 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5643 (long)(locinput - PL_reg_starttry),
5644 (long)(reginfo->till - PL_reg_starttry),
5647 sayNO_SILENT; /* Cannot match: too short. */
5649 PL_reginput = locinput; /* put where regtry can find it */
5650 sayYES; /* Success! */
5652 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5654 PerlIO_printf(Perl_debug_log,
5655 "%*s %ssubpattern success...%s\n",
5656 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5657 PL_reginput = locinput; /* put where regtry can find it */
5658 sayYES; /* Success! */
5661 #define ST st->u.ifmatch
5663 case SUSPEND: /* (?>A) */
5665 PL_reginput = locinput;
5668 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5670 goto ifmatch_trivial_fail_test;
5672 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5674 ifmatch_trivial_fail_test:
5676 char * const s = HOPBACKc(locinput, scan->flags);
5681 sw = 1 - cBOOL(ST.wanted);
5685 next = scan + ARG(scan);
5693 PL_reginput = locinput;
5697 ST.logical = logical;
5698 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5700 /* execute body of (?...A) */
5701 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5702 assert(0); /* NOTREACHED */
5704 case IFMATCH_A_fail: /* body of (?...A) failed */
5705 ST.wanted = !ST.wanted;
5708 case IFMATCH_A: /* body of (?...A) succeeded */
5710 sw = cBOOL(ST.wanted);
5712 else if (!ST.wanted)
5715 if (OP(ST.me) == SUSPEND)
5716 locinput = PL_reginput;
5718 locinput = PL_reginput = st->locinput;
5719 nextchr = UCHARAT(locinput);
5721 scan = ST.me + ARG(ST.me);
5724 continue; /* execute B */
5729 next = scan + ARG(scan);
5734 reginfo->cutpoint = PL_regeol;
5737 PL_reginput = locinput;
5739 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5740 PUSH_STATE_GOTO(COMMIT_next,next);
5741 assert(0); /* NOTREACHED */
5742 case COMMIT_next_fail:
5747 assert(0); /* NOTREACHED */
5749 #define ST st->u.mark
5751 ST.prev_mark = mark_state;
5752 ST.mark_name = sv_commit = sv_yes_mark
5753 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5755 ST.mark_loc = PL_reginput = locinput;
5756 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5757 assert(0); /* NOTREACHED */
5758 case MARKPOINT_next:
5759 mark_state = ST.prev_mark;
5761 assert(0); /* NOTREACHED */
5762 case MARKPOINT_next_fail:
5763 if (popmark && sv_eq(ST.mark_name,popmark))
5765 if (ST.mark_loc > startpoint)
5766 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5767 popmark = NULL; /* we found our mark */
5768 sv_commit = ST.mark_name;
5771 PerlIO_printf(Perl_debug_log,
5772 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5773 REPORT_CODE_OFF+depth*2, "",
5774 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5777 mark_state = ST.prev_mark;
5778 sv_yes_mark = mark_state ?
5779 mark_state->u.mark.mark_name : NULL;
5781 assert(0); /* NOTREACHED */
5783 PL_reginput = locinput;
5785 /* (*SKIP) : if we fail we cut here*/
5786 ST.mark_name = NULL;
5787 ST.mark_loc = locinput;
5788 PUSH_STATE_GOTO(SKIP_next,next);
5790 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5791 otherwise do nothing. Meaning we need to scan
5793 regmatch_state *cur = mark_state;
5794 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5797 if ( sv_eq( cur->u.mark.mark_name,
5800 ST.mark_name = find;
5801 PUSH_STATE_GOTO( SKIP_next, next );
5803 cur = cur->u.mark.prev_mark;
5806 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5808 case SKIP_next_fail:
5810 /* (*CUT:NAME) - Set up to search for the name as we
5811 collapse the stack*/
5812 popmark = ST.mark_name;
5814 /* (*CUT) - No name, we cut here.*/
5815 if (ST.mark_loc > startpoint)
5816 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5817 /* but we set sv_commit to latest mark_name if there
5818 is one so they can test to see how things lead to this
5821 sv_commit=mark_state->u.mark.mark_name;
5825 assert(0); /* NOTREACHED */
5828 if ((n=is_LNBREAK(locinput,utf8_target))) {
5830 nextchr = UCHARAT(locinput);
5835 #define CASE_CLASS(nAmE) \
5837 if (locinput >= PL_regeol) \
5839 if ((n=is_##nAmE(locinput,utf8_target))) { \
5841 nextchr = UCHARAT(locinput); \
5846 if (locinput >= PL_regeol) \
5848 if ((n=is_##nAmE(locinput,utf8_target))) { \
5851 locinput += UTF8SKIP(locinput); \
5852 nextchr = UCHARAT(locinput); \
5857 CASE_CLASS(HORIZWS);
5861 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5862 PTR2UV(scan), OP(scan));
5863 Perl_croak(aTHX_ "regexp memory corruption");
5867 /* switch break jumps here */
5868 scan = next; /* prepare to execute the next op and ... */
5869 continue; /* ... jump back to the top, reusing st */
5870 assert(0); /* NOTREACHED */
5873 /* push a state that backtracks on success */
5874 st->u.yes.prev_yes_state = yes_state;
5878 /* push a new regex state, then continue at scan */
5880 regmatch_state *newst;
5883 regmatch_state *cur = st;
5884 regmatch_state *curyes = yes_state;
5886 regmatch_slab *slab = PL_regmatch_slab;
5887 for (;curd > -1;cur--,curd--) {
5888 if (cur < SLAB_FIRST(slab)) {
5890 cur = SLAB_LAST(slab);
5892 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5893 REPORT_CODE_OFF + 2 + depth * 2,"",
5894 curd, PL_reg_name[cur->resume_state],
5895 (curyes == cur) ? "yes" : ""
5898 curyes = cur->u.yes.prev_yes_state;
5901 DEBUG_STATE_pp("push")
5904 st->locinput = locinput;
5906 if (newst > SLAB_LAST(PL_regmatch_slab))
5907 newst = S_push_slab(aTHX);
5908 PL_regmatch_state = newst;
5910 locinput = PL_reginput;
5911 nextchr = UCHARAT(locinput);
5914 assert(0); /* NOTREACHED */
5919 * We get here only if there's trouble -- normally "case END" is
5920 * the terminating point.
5922 Perl_croak(aTHX_ "corrupted regexp pointers");
5928 /* we have successfully completed a subexpression, but we must now
5929 * pop to the state marked by yes_state and continue from there */
5930 assert(st != yes_state);
5932 while (st != yes_state) {
5934 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5935 PL_regmatch_slab = PL_regmatch_slab->prev;
5936 st = SLAB_LAST(PL_regmatch_slab);
5940 DEBUG_STATE_pp("pop (no final)");
5942 DEBUG_STATE_pp("pop (yes)");
5948 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5949 || yes_state > SLAB_LAST(PL_regmatch_slab))
5951 /* not in this slab, pop slab */
5952 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5953 PL_regmatch_slab = PL_regmatch_slab->prev;
5954 st = SLAB_LAST(PL_regmatch_slab);
5956 depth -= (st - yes_state);
5959 yes_state = st->u.yes.prev_yes_state;
5960 PL_regmatch_state = st;
5963 locinput= st->locinput;
5964 nextchr = UCHARAT(locinput);
5966 state_num = st->resume_state + no_final;
5967 goto reenter_switch;
5970 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5971 PL_colors[4], PL_colors[5]));
5973 if (PL_reg_state.re_state_eval_setup_done) {
5974 /* each successfully executed (?{...}) block does the equivalent of
5975 * local $^R = do {...}
5976 * When popping the save stack, all these locals would be undone;
5977 * bypass this by setting the outermost saved $^R to the latest
5979 if (oreplsv != GvSV(PL_replgv))
5980 sv_setsv(oreplsv, GvSV(PL_replgv));
5987 PerlIO_printf(Perl_debug_log,
5988 "%*s %sfailed...%s\n",
5989 REPORT_CODE_OFF+depth*2, "",
5990 PL_colors[4], PL_colors[5])
6002 /* there's a previous state to backtrack to */
6004 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6005 PL_regmatch_slab = PL_regmatch_slab->prev;
6006 st = SLAB_LAST(PL_regmatch_slab);
6008 PL_regmatch_state = st;
6009 locinput= st->locinput;
6010 nextchr = UCHARAT(locinput);
6012 DEBUG_STATE_pp("pop");
6014 if (yes_state == st)
6015 yes_state = st->u.yes.prev_yes_state;
6017 state_num = st->resume_state + 1; /* failure = success + 1 */
6018 goto reenter_switch;
6023 if (rex->intflags & PREGf_VERBARG_SEEN) {
6024 SV *sv_err = get_sv("REGERROR", 1);
6025 SV *sv_mrk = get_sv("REGMARK", 1);
6027 sv_commit = &PL_sv_no;
6029 sv_yes_mark = &PL_sv_yes;
6032 sv_commit = &PL_sv_yes;
6033 sv_yes_mark = &PL_sv_no;
6035 sv_setsv(sv_err, sv_commit);
6036 sv_setsv(sv_mrk, sv_yes_mark);
6040 if (last_pushed_cv) {
6045 /* clean up; in particular, free all slabs above current one */
6046 LEAVE_SCOPE(oldsave);
6052 - regrepeat - repeatedly match something simple, report how many
6055 * [This routine now assumes that it will only match on things of length 1.
6056 * That was true before, but now we assume scan - reginput is the count,
6057 * rather than incrementing count on every character. [Er, except utf8.]]
6060 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
6063 register char *scan;
6065 register char *loceol = PL_regeol;
6066 register I32 hardcount = 0;
6067 register bool utf8_target = PL_reg_match_utf8;
6070 PERL_UNUSED_ARG(depth);
6073 PERL_ARGS_ASSERT_REGREPEAT;
6076 if (max == REG_INFTY)
6078 else if (max < loceol - scan)
6079 loceol = scan + max;
6084 while (scan < loceol && hardcount < max && *scan != '\n') {
6085 scan += UTF8SKIP(scan);
6089 while (scan < loceol && *scan != '\n')
6096 while (scan < loceol && hardcount < max) {
6097 scan += UTF8SKIP(scan);
6108 /* To get here, EXACTish nodes must have *byte* length == 1. That
6109 * means they match only characters in the string that can be expressed
6110 * as a single byte. For non-utf8 strings, that means a simple match.
6111 * For utf8 strings, the character matched must be an invariant, or
6112 * downgradable to a single byte. The pattern's utf8ness is
6113 * irrelevant, as since it's a single byte, it either isn't utf8, or if
6114 * it is, it's an invariant */
6117 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6119 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6120 while (scan < loceol && UCHARAT(scan) == c) {
6126 /* Here, the string is utf8, and the pattern char is different
6127 * in utf8 than not, so can't compare them directly. Outside the
6128 * loop, find the two utf8 bytes that represent c, and then
6129 * look for those in sequence in the utf8 string */
6130 U8 high = UTF8_TWO_BYTE_HI(c);
6131 U8 low = UTF8_TWO_BYTE_LO(c);
6134 while (hardcount < max
6135 && scan + 1 < loceol
6136 && UCHARAT(scan) == high
6137 && UCHARAT(scan + 1) == low)
6145 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6149 PL_reg_flags |= RF_tainted;
6150 utf8_flags = FOLDEQ_UTF8_LOCALE;
6158 case EXACTFU_TRICKYFOLD:
6160 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6162 /* The comments for the EXACT case above apply as well to these fold
6167 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6169 if (utf8_target || OP(p) == EXACTFU_SS) { /* Use full Unicode fold matching */
6170 char *tmpeol = loceol;
6171 while (hardcount < max
6172 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6173 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6180 /* XXX Note that the above handles properly the German sharp s in
6181 * the pattern matching ss in the string. But it doesn't handle
6182 * properly cases where the string contains say 'LIGATURE ff' and
6183 * the pattern is 'f+'. This would require, say, a new function or
6184 * revised interface to foldEQ_utf8(), in which the maximum number
6185 * of characters to match could be passed and it would return how
6186 * many actually did. This is just one of many cases where
6187 * multi-char folds don't work properly, and so the fix is being
6193 /* Here, the string isn't utf8 and c is a single byte; and either
6194 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6195 * doesn't affect c. Can just do simple comparisons for exact or
6198 case EXACTF: folded = PL_fold[c]; break;
6200 case EXACTFU_TRICKYFOLD:
6201 case EXACTFU: folded = PL_fold_latin1[c]; break;
6202 case EXACTFL: folded = PL_fold_locale[c]; break;
6203 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6205 while (scan < loceol &&
6206 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6214 if (utf8_target || OP(p) == ANYOFV) {
6217 inclasslen = loceol - scan;
6218 while (hardcount < max
6219 && ((inclasslen = loceol - scan) > 0)
6220 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6226 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6234 LOAD_UTF8_CHARCLASS_ALNUM();
6235 while (hardcount < max && scan < loceol &&
6236 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6238 scan += UTF8SKIP(scan);
6242 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6250 while (scan < loceol && isALNUM((U8) *scan)) {
6255 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6260 PL_reg_flags |= RF_tainted;
6263 while (hardcount < max && scan < loceol &&
6264 isALNUM_LC_utf8((U8*)scan)) {
6265 scan += UTF8SKIP(scan);
6269 while (scan < loceol && isALNUM_LC(*scan))
6279 LOAD_UTF8_CHARCLASS_ALNUM();
6280 while (hardcount < max && scan < loceol &&
6281 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6283 scan += UTF8SKIP(scan);
6287 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6294 goto utf8_Nwordchar;
6295 while (scan < loceol && ! isALNUM((U8) *scan)) {
6301 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6302 scan += UTF8SKIP(scan);
6306 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6312 PL_reg_flags |= RF_tainted;
6315 while (hardcount < max && scan < loceol &&
6316 !isALNUM_LC_utf8((U8*)scan)) {
6317 scan += UTF8SKIP(scan);
6321 while (scan < loceol && !isALNUM_LC(*scan))
6331 LOAD_UTF8_CHARCLASS_SPACE();
6332 while (hardcount < max && scan < loceol &&
6334 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6336 scan += UTF8SKIP(scan);
6342 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6351 while (scan < loceol && isSPACE((U8) *scan)) {
6356 while (scan < loceol && isSPACE_A((U8) *scan)) {
6361 PL_reg_flags |= RF_tainted;
6364 while (hardcount < max && scan < loceol &&
6365 isSPACE_LC_utf8((U8*)scan)) {
6366 scan += UTF8SKIP(scan);
6370 while (scan < loceol && isSPACE_LC(*scan))
6380 LOAD_UTF8_CHARCLASS_SPACE();
6381 while (hardcount < max && scan < loceol &&
6383 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6385 scan += UTF8SKIP(scan);
6391 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6400 while (scan < loceol && ! isSPACE((U8) *scan)) {
6406 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6407 scan += UTF8SKIP(scan);
6411 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6417 PL_reg_flags |= RF_tainted;
6420 while (hardcount < max && scan < loceol &&
6421 !isSPACE_LC_utf8((U8*)scan)) {
6422 scan += UTF8SKIP(scan);
6426 while (scan < loceol && !isSPACE_LC(*scan))
6433 LOAD_UTF8_CHARCLASS_DIGIT();
6434 while (hardcount < max && scan < loceol &&
6435 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6436 scan += UTF8SKIP(scan);
6440 while (scan < loceol && isDIGIT(*scan))
6445 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6450 PL_reg_flags |= RF_tainted;
6453 while (hardcount < max && scan < loceol &&
6454 isDIGIT_LC_utf8((U8*)scan)) {
6455 scan += UTF8SKIP(scan);
6459 while (scan < loceol && isDIGIT_LC(*scan))
6466 LOAD_UTF8_CHARCLASS_DIGIT();
6467 while (hardcount < max && scan < loceol &&
6468 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6469 scan += UTF8SKIP(scan);
6473 while (scan < loceol && !isDIGIT(*scan))
6479 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6480 scan += UTF8SKIP(scan);
6484 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6490 PL_reg_flags |= RF_tainted;
6493 while (hardcount < max && scan < loceol &&
6494 !isDIGIT_LC_utf8((U8*)scan)) {
6495 scan += UTF8SKIP(scan);
6499 while (scan < loceol && !isDIGIT_LC(*scan))
6506 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6512 LNBREAK can match two latin chars, which is ok,
6513 because we have a null terminated string, but we
6514 have to use hardcount in this situation
6516 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6525 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6530 while (scan < loceol && is_HORIZWS_latin1(scan))
6537 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6538 scan += UTF8SKIP(scan);
6542 while (scan < loceol && !is_HORIZWS_latin1(scan))
6550 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6555 while (scan < loceol && is_VERTWS_latin1(scan))
6563 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6564 scan += UTF8SKIP(scan);
6568 while (scan < loceol && !is_VERTWS_latin1(scan))
6574 default: /* Called on something of 0 width. */
6575 break; /* So match right here or not at all. */
6581 c = scan - PL_reginput;
6585 GET_RE_DEBUG_FLAGS_DECL;
6587 SV * const prop = sv_newmortal();
6588 regprop(prog, prop, p);
6589 PerlIO_printf(Perl_debug_log,
6590 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6591 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6599 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6601 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
6602 create a copy so that changes the caller makes won't change the shared one
6605 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6607 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6608 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp, altsvp));
6613 S_core_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6615 /* Returns the swash for the input 'node' in the regex 'prog'.
6616 * If <doinit> is true, will attempt to create the swash if not already
6618 * If <listsvp> is non-null, will return the swash initialization string in
6620 * If <altsvp> is non-null, will return the alternates to the regular swash
6622 * Tied intimately to how regcomp.c sets up the data structure */
6630 RXi_GET_DECL(prog,progi);
6631 const struct reg_data * const data = prog ? progi->data : NULL;
6633 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
6635 assert(ANYOF_NONBITMAP(node));
6637 if (data && data->count) {
6638 const U32 n = ARG(node);
6640 if (data->what[n] == 's') {
6641 SV * const rv = MUTABLE_SV(data->data[n]);
6642 AV * const av = MUTABLE_AV(SvRV(rv));
6643 SV **const ary = AvARRAY(av);
6644 bool invlist_has_user_defined_property;
6646 si = *ary; /* ary[0] = the string to initialize the swash with */
6648 /* Elements 3 and 4 are either both present or both absent. [3] is
6649 * any inversion list generated at compile time; [4] indicates if
6650 * that inversion list has any user-defined properties in it. */
6651 if (av_len(av) >= 3) {
6653 invlist_has_user_defined_property = cBOOL(SvUV(ary[4]));
6657 invlist_has_user_defined_property = FALSE;
6660 /* Element [1] is reserved for the set-up swash. If already there,
6661 * return it; if not, create it and store it there */
6662 if (SvROK(ary[1])) {
6665 else if (si && doinit) {
6667 sw = _core_swash_init("utf8", /* the utf8 package */
6671 0, /* not from tr/// */
6672 FALSE, /* is error if can't find
6675 invlist_has_user_defined_property);
6676 (void)av_store(av, 1, sw);
6679 /* Element [2] is for any multi-char folds. Note that is a
6680 * fundamentally flawed design, because can't backtrack and try
6681 * again. See [perl #89774] */
6682 if (SvTYPE(ary[2]) == SVt_PVAV) {
6689 SV* matches_string = newSVpvn("", 0);
6692 /* Use the swash, if any, which has to have incorporated into it all
6696 && SvTYPE(SvRV(sw)) == SVt_PVHV
6697 && (invlistsvp = hv_fetchs(MUTABLE_HV(SvRV(sw)), "INVLIST", FALSE)))
6699 invlist = *invlistsvp;
6701 else if (si && si != &PL_sv_undef) {
6703 /* If no swash, use the input nitialization string, if available */
6704 sv_catsv(matches_string, si);
6707 /* Add the inversion list to whatever we have. This may have come from
6708 * the swash, or from an input parameter */
6710 sv_catsv(matches_string, _invlist_contents(invlist));
6712 *listsvp = matches_string;
6722 - reginclass - determine if a character falls into a character class
6724 n is the ANYOF regnode
6725 p is the target string
6726 lenp is pointer to the maximum number of bytes of how far to go in p
6727 (This is assumed wthout checking to always be at least the current
6729 utf8_target tells whether p is in UTF-8.
6731 Returns true if matched; false otherwise. If lenp is not NULL, on return
6732 from a successful match, the value it points to will be updated to how many
6733 bytes in p were matched. If there was no match, the value is undefined,
6734 possibly changed from the input.
6736 Note that this can be a synthetic start class, a combination of various
6737 nodes, so things you think might be mutually exclusive, such as locale,
6738 aren't. It can match both locale and non-locale
6743 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6746 const char flags = ANYOF_FLAGS(n);
6752 PERL_ARGS_ASSERT_REGINCLASS;
6754 /* If c is not already the code point, get it */
6755 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6756 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6757 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6758 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6759 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6760 * UTF8_ALLOW_FFFF */
6761 if (c_len == (STRLEN)-1)
6762 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6768 /* Use passed in max length, or one character if none passed in or less
6769 * than one character. And assume will match just one character. This is
6770 * overwritten later if matched more. */
6772 maxlen = (*lenp > c_len) ? *lenp : c_len;
6780 /* If this character is potentially in the bitmap, check it */
6782 if (ANYOF_BITMAP_TEST(n, c))
6784 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6791 else if (flags & ANYOF_LOCALE) {
6792 PL_reg_flags |= RF_tainted;
6794 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6795 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6799 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6800 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6801 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6802 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6803 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6804 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6805 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6806 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6807 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6808 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6809 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6810 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII_LC(c)) ||
6811 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII_LC(c)) ||
6812 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6813 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6814 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6815 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6816 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6817 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6818 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6819 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6820 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6821 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6822 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6823 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6824 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6825 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6826 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6827 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6828 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK_LC(c)) ||
6829 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK_LC(c))
6830 ) /* How's that for a conditional? */
6837 /* If the bitmap didn't (or couldn't) match, and something outside the
6838 * bitmap could match, try that. Locale nodes specifiy completely the
6839 * behavior of code points in the bit map (otherwise, a utf8 target would
6840 * cause them to be treated as Unicode and not locale), except in
6841 * the very unlikely event when this node is a synthetic start class, which
6842 * could be a combination of locale and non-locale nodes. So allow locale
6843 * to match for the synthetic start class, which will give a false
6844 * positive that will be resolved when the match is done again as not part
6845 * of the synthetic start class */
6847 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6848 match = TRUE; /* Everything above 255 matches */
6850 else if (ANYOF_NONBITMAP(n)
6851 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6854 || (! (flags & ANYOF_LOCALE))
6855 || (flags & ANYOF_IS_SYNTHETIC)))))
6858 SV * const sw = core_regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6866 /* Not utf8. Convert as much of the string as available up
6867 * to the limit of how far the (single) character in the
6868 * pattern can possibly match (no need to go further). If
6869 * the node is a straight ANYOF or not folding, it can't
6870 * match more than one. Otherwise, It can match up to how
6871 * far a single char can fold to. Since not utf8, each
6872 * character is a single byte, so the max it can be in
6873 * bytes is the same as the max it can be in characters */
6874 STRLEN len = (OP(n) == ANYOF
6875 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6877 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6879 : UTF8_MAX_FOLD_CHAR_EXPAND;
6880 utf8_p = bytes_to_utf8(p, &len);
6883 if (swash_fetch(sw, utf8_p, TRUE))
6885 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6887 /* Here, we need to test if the fold of the target string
6888 * matches. The non-multi char folds have all been moved to
6889 * the compilation phase, and the multi-char folds have
6890 * been stored by regcomp into 'av'; we linearly check to
6891 * see if any match the target string (folded). We know
6892 * that the originals were each one character, but we don't
6893 * currently know how many characters/bytes each folded to,
6894 * except we do know that there are small limits imposed by
6895 * Unicode. XXX A performance enhancement would be to have
6896 * regcomp.c store the max number of chars/bytes that are
6897 * in an av entry, as, say the 0th element. Even better
6898 * would be to have a hash of the few characters that can
6899 * start a multi-char fold to the max number of chars of
6902 * If there is a match, we will need to advance (if lenp is
6903 * specified) the match pointer in the target string. But
6904 * what we are comparing here isn't that string directly,
6905 * but its fold, whose length may differ from the original.
6906 * As we go along in constructing the fold, therefore, we
6907 * create a map so that we know how many bytes in the
6908 * source to advance given that we have matched a certain
6909 * number of bytes in the fold. This map is stored in
6910 * 'map_fold_len_back'. Let n mean the number of bytes in
6911 * the fold of the first character that we are folding.
6912 * Then map_fold_len_back[n] is set to the number of bytes
6913 * in that first character. Similarly let m be the
6914 * corresponding number for the second character to be
6915 * folded. Then map_fold_len_back[n+m] is set to the
6916 * number of bytes occupied by the first two source
6917 * characters. ... */
6918 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6919 U8 folded[UTF8_MAXBYTES_CASE+1];
6920 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6921 STRLEN total_foldlen = 0; /* num bytes in fold of all
6924 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6926 /* Here, only need to fold the first char of the target
6927 * string. It the source wasn't utf8, is 1 byte long */
6928 to_utf8_fold(utf8_p, folded, &foldlen);
6929 total_foldlen = foldlen;
6930 map_fold_len_back[foldlen] = (utf8_target)
6936 /* Here, need to fold more than the first char. Do so
6937 * up to the limits */
6938 U8* source_ptr = utf8_p; /* The source for the fold
6941 U8* folded_ptr = folded;
6942 U8* e = utf8_p + maxlen; /* Can't go beyond last
6943 available byte in the
6947 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6951 /* Fold the next character */
6952 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6953 STRLEN this_char_foldlen;
6954 to_utf8_fold(source_ptr,
6956 &this_char_foldlen);
6958 /* Bail if it would exceed the byte limit for
6959 * folding a single char. */
6960 if (this_char_foldlen + folded_ptr - folded >
6966 /* Add the fold of this character */
6967 Copy(this_char_folded,
6971 source_ptr += UTF8SKIP(source_ptr);
6972 folded_ptr += this_char_foldlen;
6973 total_foldlen = folded_ptr - folded;
6975 /* Create map from the number of bytes in the fold
6976 * back to the number of bytes in the source. If
6977 * the source isn't utf8, the byte count is just
6978 * the number of characters so far */
6979 map_fold_len_back[total_foldlen]
6981 ? source_ptr - utf8_p
6988 /* Do the linear search to see if the fold is in the list
6989 * of multi-char folds. */
6992 for (i = 0; i <= av_len(av); i++) {
6993 SV* const sv = *av_fetch(av, i, FALSE);
6995 const char * const s = SvPV_const(sv, len);
6997 if (len <= total_foldlen
6998 && memEQ(s, (char*)folded, len)
7000 /* If 0, means matched a partial char. See
7002 && map_fold_len_back[len])
7005 /* Advance the target string ptr to account for
7006 * this fold, but have to translate from the
7007 * folded length to the corresponding source
7010 *lenp = map_fold_len_back[len];
7019 /* If we allocated a string above, free it */
7020 if (! utf8_target) Safefree(utf8_p);
7025 return (flags & ANYOF_INVERT) ? !match : match;
7029 S_reghop3(U8 *s, I32 off, const U8* lim)
7031 /* return the position 'off' UTF-8 characters away from 's', forward if
7032 * 'off' >= 0, backwards if negative. But don't go outside of position
7033 * 'lim', which better be < s if off < 0 */
7037 PERL_ARGS_ASSERT_REGHOP3;
7040 while (off-- && s < lim) {
7041 /* XXX could check well-formedness here */
7046 while (off++ && s > lim) {
7048 if (UTF8_IS_CONTINUED(*s)) {
7049 while (s > lim && UTF8_IS_CONTINUATION(*s))
7052 /* XXX could check well-formedness here */
7059 /* there are a bunch of places where we use two reghop3's that should
7060 be replaced with this routine. but since thats not done yet
7061 we ifdef it out - dmq
7064 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7068 PERL_ARGS_ASSERT_REGHOP4;
7071 while (off-- && s < rlim) {
7072 /* XXX could check well-formedness here */
7077 while (off++ && s > llim) {
7079 if (UTF8_IS_CONTINUED(*s)) {
7080 while (s > llim && UTF8_IS_CONTINUATION(*s))
7083 /* XXX could check well-formedness here */
7091 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7095 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7098 while (off-- && s < lim) {
7099 /* XXX could check well-formedness here */
7106 while (off++ && s > lim) {
7108 if (UTF8_IS_CONTINUED(*s)) {
7109 while (s > lim && UTF8_IS_CONTINUATION(*s))
7112 /* XXX could check well-formedness here */
7121 restore_pos(pTHX_ void *arg)
7124 regexp * const rex = (regexp *)arg;
7125 if (PL_reg_state.re_state_eval_setup_done) {
7126 if (PL_reg_oldsaved) {
7127 rex->subbeg = PL_reg_oldsaved;
7128 rex->sublen = PL_reg_oldsavedlen;
7129 #ifdef PERL_OLD_COPY_ON_WRITE
7130 rex->saved_copy = PL_nrs;
7132 RXp_MATCH_COPIED_on(rex);
7134 PL_reg_magic->mg_len = PL_reg_oldpos;
7135 PL_reg_state.re_state_eval_setup_done = FALSE;
7136 PL_curpm = PL_reg_oldcurpm;
7141 S_to_utf8_substr(pTHX_ register regexp *prog)
7145 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7148 if (prog->substrs->data[i].substr
7149 && !prog->substrs->data[i].utf8_substr) {
7150 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7151 prog->substrs->data[i].utf8_substr = sv;
7152 sv_utf8_upgrade(sv);
7153 if (SvVALID(prog->substrs->data[i].substr)) {
7154 if (SvTAIL(prog->substrs->data[i].substr)) {
7155 /* Trim the trailing \n that fbm_compile added last
7157 SvCUR_set(sv, SvCUR(sv) - 1);
7158 /* Whilst this makes the SV technically "invalid" (as its
7159 buffer is no longer followed by "\0") when fbm_compile()
7160 adds the "\n" back, a "\0" is restored. */
7161 fbm_compile(sv, FBMcf_TAIL);
7165 if (prog->substrs->data[i].substr == prog->check_substr)
7166 prog->check_utf8 = sv;
7172 S_to_byte_substr(pTHX_ register regexp *prog)
7177 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7180 if (prog->substrs->data[i].utf8_substr
7181 && !prog->substrs->data[i].substr) {
7182 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7183 if (sv_utf8_downgrade(sv, TRUE)) {
7184 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7185 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7186 /* Trim the trailing \n that fbm_compile added last
7188 SvCUR_set(sv, SvCUR(sv) - 1);
7189 fbm_compile(sv, FBMcf_TAIL);
7197 prog->substrs->data[i].substr = sv;
7198 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7199 prog->check_substr = sv;
7206 * c-indentation-style: bsd
7208 * indent-tabs-mode: nil
7211 * ex: set ts=8 sts=4 sw=4 et: